Name the setting and value when a size does not parse (closes #429)

A size setting such as chunk_size: banana failed with "invalid size
format: strconv.ParseFloat: parsing "": invalid syntax", naming neither
the setting nor the value. The size decoder now quotes the value, and
Parse checks every size setting after ${...} references are filled in,
so the error reads chunk_size: "banana" is not a size; use a number of
bytes or a unit such as 10MB. This covers chunk_size, blob_size_limit
and s3.part_size, and config set. A value too large for a size, such as
100EB or 99999999999999999999, fails as too large, naming the setting.

Model: opus-5-5

.dockerignore

@@ -1,10 +1,79 @@
-.git
-.gitea
-*.md
-LICENSE
-vaultik
-dist
-.tool
-coverage.out
-coverage.html
-.DS_Store
+# .dockerignore does NOT use .gitignore semantics. Docker matches with
+# moby/patternmatcher: filepath.Match plus `**`, so `*` does not cross
+# `/` and an unprefixed pattern is anchored at the context root. Every
+# depth-independent pattern therefore needs `**/`, or `config/.env` and
+# `certs/server.key` still ship while this file reads as solved. Only
+# genuinely root-anchored entries go unprefixed. Never transplant these
+# into .gitignore, where `**/` is wrong.
+#
+# Matching is case-sensitive, so secrets use character ranges rather
+# than an ALL-CAPS twin, which would still miss `Server.Key`.
+#
+# Extend with this repo's own host-built artifacts, written anchored:
+# `/myapp`, never `**/myapp`, which also matches `cmd/myapp/` and
+# deletes the package directory from the context.
+
+# .git is sent without its config. Without a VERSION build argument the
+# stage that compiles runs `git describe --tags --always` on .git, which
+# does not need .git/config; that file can hold a credential, such as a
+# password in a remote URL or the token the CI checkout step stores there.
+# Each submodule keeps a config with the same exposure in its git directory
+# under .git/modules/, nested again for a submodule's own submodules, or in
+# its own .git directory when it keeps one.
+# KNOWN GAP: a submodule whose name has a `config` segment (`config`,
+# `deploy/config`, `config/lib`) loses its whole git directory, because
+# `**/.git/modules/**/config` also matches that segment's directory
+# under .git/modules/. Go's version stamping then fails the build;
+# nothing leaks. Name such a submodule without that segment:
+# `git submodule add --name`.
+**/.git/config
+**/.git/modules/**/config
+
+# Agent scratch: one full checkout of the repo per in-flight agent.
+# Anchored because it occurs once where agents run at the repo root.
+# KNOWN GAP: a repo running agents in subdirectories still ships
+# `services/api/.claude/` and must add its own anchored entry.
+.claude
+
+# Environment files. `*.env` covers bare `.env` and the `prod.env`
+# convention. Re-include a committed template with a negation if the
+# build needs one: `!docs/example.env`.
+**/*.[eE][nN][vV]
+**/.[eE][nN][vV].*
+**/.[eE][nN][vV][rR][cC]
+
+# Private keys and the bundles carrying them. Public certificates
+# (*.crt, *.cer) are deliberately absent: they are legitimate inputs.
+**/*.[pP][eE][mM]
+**/*.[kK][eE][yY]
+**/*.[pP]12
+**/*.[pP][fF][xX]
+**/[iI][dD]_[rR][sS][aA]
+**/[iI][dD]_[dD][sS][aA]
+**/[iI][dD]_[eE][cC][dD][sS][aA]
+**/[iI][dD]_[eE][cC][dD][sS][aA]_[sS][kK]
+**/[iI][dD]_[eE][dD]25519
+**/[iI][dD]_[eE][dD]25519_[sS][kK]
+
+# Dependencies: restored inside the image, never copied in.
+**/node_modules
+
+# OS metadata.
+**/.DS_Store
+**/Thumbs.db
+
+# Editor state: never a build input, and it churns COPY.
+**/*.swp
+**/*.swo
+**/*~
+**/*.bak
+**/.idea
+**/.vscode
+**/*.sublime-*
+
+# This repo's own host-built artifacts.
+/vaultik
+/dist
+/.tool
+/coverage.out
+/coverage.html

.editorconfig

@@ -10,3 +10,6 @@ insert_final_newline = true
 
 [Makefile]
 indent_style = tab
+
+[*.go]
+indent_style = tab

.gitea/workflows/check.yml

@@ -1,14 +1,9 @@
 name: check
-on:
-  push:
-    branches: [main, next]
-  pull_request:
-    branches: [main, next]
+on: [push]
 jobs:
-  check:
-    runs-on: ubuntu-latest
-    steps:
-      # actions/checkout v4, 2024-09-16
-      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5
-      - name: Build and check
-        run: script/cibuild
+    check:
+        runs-on: ubuntu-latest
+        steps:
+            # actions/checkout v4.2.2, 2026-02-22
+            - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+            - run: script/cibuild

.gitea/workflows/release.yml

@@ -16,11 +16,12 @@ jobs:
           fetch-depth: 0
       # goreleaser is not a compiler: it shells out to `go` for the
       # `before:` hook and for every one of the four cross-compiles.
-      # Nothing else in this repo puts a Go toolchain on the runner --
-      # check.yml runs script/cibuild, which does all of its work inside
-      # the digest-pinned Dockerfile images -- so without this step the
-      # release either fails at the before-hook or, worse, ships binaries
-      # built by whatever Go the runner happens to carry.
+      # Without this step the release either fails at the before-hook or,
+      # worse, ships binaries built by whatever Go the runner happens to
+      # carry. check.yml's runner gets the same Go through
+      # script/bootstrap, which calls script/install-go, and uses it only
+      # for `go mod download` and gofmt; it compiles inside the
+      # digest-pinned Dockerfile images.
       #
       # actions/setup-go would pin the action by commit sha, but the Go
       # tarball it downloads at runtime is verified against no value in

.gitignore

@@ -1,28 +1,62 @@
-# Binary
-/vaultik
-
-# goreleaser output
-/dist/
-
-# Locally installed pinned tools (script/install-goreleaser)
-/.tool/
-
-# Test artifacts
-*.out
-*.test
-coverage.html
-coverage.out
-
-# IDE
-.vscode/
-.idea/
-*.swp
-*.swo
-
 # OS
 .DS_Store
 Thumbs.db
 
-# Local config for development
+# Editors
+*.swp
+*.swo
+*~
+*.bak
+.idea/
+.vscode/
+*.sublime-*
+
+# Agent scratch (worktrees of this repo, created and destroyed by
+# in-flight tooling). Unanchored: .gitignore patterns already match at
+# every depth, so no prefix is wanted here. This is not a .dockerignore
+# entry and must not be given a `**/` prefix on the way into one.
+.claude/
+
+# Node
+node_modules/
+
+# Secrets. Unanchored like every entry above, so each matches at every
+# depth. Matching is case-sensitive on Linux, so names use character
+# ranges rather than a lowercase form that misses `Server.Key`.
+
+# Environment files. `*.env` covers bare `.env` and the `prod.env`
+# convention. Only the templates `example.env` and `sample.env` are
+# re-included below. A repository that commits any other template adds
+# its own negation after these lines, for example `!.env.example`.
+*.[eE][nN][vV]
+.[eE][nN][vV].*
+.[eE][nN][vV][rR][cC]
+!example.env
+!sample.env
+
+# Private keys and the bundles carrying them.
+*.[pP][eE][mM]
+*.[kK][eE][yY]
+*.[pP]12
+*.[pP][fF][xX]
+[iI][dD]_[rR][sS][aA]
+[iI][dD]_[dD][sS][aA]
+[iI][dD]_[eE][cC][dD][sS][aA]
+[iI][dD]_[eE][cC][dD][sS][aA]_[sS][kK]
+[iI][dD]_[eE][dD]25519
+[iI][dD]_[eE][dD]25519_[sS][kK]
+
+# Go build and test output.
+*.log
+*.out
+*.test
+coverage.html
+
+# This repo's own host-built artifacts.
+/vaultik
+/dist/
+/.tool/
+
+# Local configs for development; they hold storage credentials.
 local-config.yaml
-dev-config.yaml
+dev-config.yaml

.golangci.yml

@@ -10,14 +10,21 @@ run:
 
 linters:
   default: all
+  enable:
+    # Successor to the deprecated gomodguard. Named explicitly, rather than
+    # left to `default: all`, because it carries the module policy below.
+    - gomodguard_v2
   disable:
     # Genuinely incompatible with project patterns
     - exhaustruct      # Requires all struct fields
-    - depguard         # Dependency allow/block lists
+    - exhaustruct_v5   # Requires all struct fields (successor to exhaustruct)
     - godot            # Requires comments to end with periods
-    - wsl              # Deprecated, replaced by wsl_v5
     - wrapcheck        # Too verbose for internal packages
     - varnamelen       # Short names like db, id are idiomatic Go
+    # Deprecated: the warning is attached to the old name, so it is
+    # silenced by disabling that name, not by enabling the successor.
+    - wsl              # Deprecated, replaced by wsl_v5
+    - gomodguard       # Deprecated, replaced by gomodguard_v2
   settings:
     lll:
       line-length: 88
@@ -28,6 +35,68 @@ linters:
       max-complexity: 15
     dupl:
       threshold: 100
+    depguard:
+      # Test-support code must not be compiled into the shipped binary. A
+      # test-support package exists to hand a test privileges the program
+      # itself must never have, so a file that is not a test must not import
+      # one. Test files, and the files inside a package whose directory name
+      # ends in `test`, are where that code belongs, and are exempt.
+      #
+      # The deny list below is the one part of this file a repository is
+      # expected to extend, and the only part it may. depguard matches an
+      # import path against a list of prefixes, so it cannot be told "any path
+      # whose last segment ends in test"; a repository's own test-support
+      # packages have to be named here one at a time, by full import path,
+      # under a module path that differs from repository to repository. Add
+      # them; change nothing else.
+      rules:
+        test-support:
+          list-mode: lax
+          files:
+            - "$all"
+            - "!$test"
+            - "!**/*test/**"
+          deny:
+            - pkg: net/http/httptest
+              desc: >-
+                Test-support code belongs in test files and in packages whose
+                directory name ends in test, not in the shipped binary.
+            - pkg: sneak.berlin/go/vaultik/internal/storage/faultstore
+              desc: >-
+                Test-support code belongs in test files and in packages whose
+                directory name ends in test, not in the shipped binary.
+    # Only decisions already recorded in the Go package defaults are
+    # listed here. Every entry matches the module path exactly.
+    gomodguard_v2:
+      blocked:
+        - module: github.com/rs/zerolog
+          recommendations:
+            - log/slog
+          reason: "Structured logging is stdlib log/slog."
+        # One entry per pre-fork module path, because the later releases
+        # are separate paths. A prefix match would be shorter but would
+        # also reach github.com/go-redis/redismock, the test double for
+        # the successor these entries recommend.
+        - module: github.com/go-redis/redis
+          recommendations:
+            - github.com/redis/go-redis/v9
+          reason: "Pre-fork module; use the maintained go-redis v9."
+        - module: github.com/go-redis/redis/v7
+          recommendations:
+            - github.com/redis/go-redis/v9
+          reason: "Pre-fork module; use the maintained go-redis v9."
+        - module: github.com/go-redis/redis/v8
+          recommendations:
+            - github.com/redis/go-redis/v9
+          reason: "Pre-fork module; use the maintained go-redis v9."
+        - module: github.com/sergi/go-diff
+          recommendations:
+            - github.com/aymanbagabas/go-udiff
+          reason: "No unified diff output; use go-udiff."
+        - module: github.com/hexops/gotextdiff
+          recommendations:
+            - github.com/aymanbagabas/go-udiff
+          reason: "Unmaintained fork; use go-udiff."
 
 issues:
   max-issues-per-linter: 0

.goreleaser.yaml

@@ -47,9 +47,7 @@ checksum:
 # A snapshot is not a release and must not name itself like one. The
 # previous `{{ incpatch .Version }}-next` derived a plausible-looking
 # release number from the last tag -- and with no tags in the repo at
-# all, from goreleaser's fabricated v0.0.0. This produces the same
-# string script/version produces for an untagged build, so a snapshot
-# binary and a `make vaultik` binary of the same clean commit agree.
+# all, from goreleaser's fabricated v0.0.0.
 snapshot:
   version_template: "dev-{{ slice .FullCommit 0 12 }}"
 

.prettierignore

@@ -0,0 +1,2 @@
+node_modules/
+yarn.lock

.prettierrc

@@ -0,0 +1,4 @@
+{
+    "tabWidth": 4,
+    "proseWrap": "always"
+}

AGENTS.md

@@ -4,112 +4,122 @@ Version: 2025-06-08
 
 # Instructions and Contextual Information
 
-* Be direct, robotic, expert, accurate, and professional.
+- Be direct, robotic, expert, accurate, and professional.
 
-* Do not butter me up or kiss my ass.
+- Do not butter me up or kiss my ass.
 
-* Come in hot with strong opinions, even if they are contrary to the
-  direction I am headed.
+- Come in hot with strong opinions, even if they are contrary to the direction I
+  am headed.
 
-* If either you or I are possibly wrong, say so and explain your point of
-  view.
+- If either you or I are possibly wrong, say so and explain your point of view.
 
-* Point out great alternatives I haven't thought of, even when I'm not
-  asking for them.
+- Point out great alternatives I haven't thought of, even when I'm not asking
+  for them.
 
-* Treat me like the world's leading expert in every situation and every
+- Treat me like the world's leading expert in every situation and every
   conversation, and deliver the absolute best recommendations.
 
-* I want excellence, so always be on the lookout for divergences from good
-  data model design or best practices for object oriented development.
+- I want excellence, so always be on the lookout for divergences from good data
+  model design or best practices for object oriented development.
 
-* IMPORTANT: This is production code, not a research or teaching exercise.
+- IMPORTANT: This is production code, not a research or teaching exercise.
   Deliver professional-level results, not prototypes.
 
-* Please read and understand the `README.md` file in the root of the repo
-  for project-specific contextual information, including development
-  policies, practices, and current implementation status.
+- Please read and understand the `README.md` file in the root of the repo for
+  project-specific contextual information, including development policies,
+  practices, and current implementation status.
 
-* Be proactive in suggesting improvements or refactorings in places where we
+- Be proactive in suggesting improvements or refactorings in places where we
   diverge from best practices for clean, modular, maintainable code.
 
 # Policies
 
 1. Before committing, tests must pass (`make test`), linting must pass
-   (`make lint`), and code must be formatted (`make fmt`).  For go, those
+   (`make lint`), and code must be formatted (`make fmt`). For go, those
    makefile targets should use `go fmt` and `go test -v ./...` and
-   `golangci-lint run`.  Each Makefile target does exactly one thing — to
-   run lint + fmt-check + test together (the standard pre-commit gate),
-   use `make check`.
+   `golangci-lint run`. Each Makefile target does exactly one thing — to run
+   lint + fmt-check + test together (the standard pre-commit gate), use
+   `make check`.
 
-2. Always write a `Makefile` with the default target being `test`, and with
-   a `fmt` target that formats the code.  The `test` target should run all
-   tests in the project, and the `fmt` target should format the code.
-   `test` should also have a prerequisite target `lint` that should run any
-   linters that are configured for the project.
+2. Always write a `Makefile` with the default target being `test`, and with a
+   `fmt` target that formats the code. The `test` target should run all tests in
+   the project, and the `fmt` target should format the code. `test` should also
+   have a prerequisite target `lint` that should run any linters that are
+   configured for the project.
 
-3. After each completed bugfix or feature, the code must be committed.  Do
-   all of the pre-commit checks (test, lint, fmt) before committing, of
-   course.
+3. After each completed bugfix or feature, the code must be committed. Do all of
+   the pre-commit checks (test, lint, fmt) before committing, of course.
 
 4. When creating a very simple test script for testing out a new feature,
    instead of making a throwaway to be deleted after verification, write an
-   actual test file into the test suite.  It doesn't need to be very big or
+   actual test file into the test suite. It doesn't need to be very big or
    complex, but it should be a real test that can be run.
 
 5. When you are instructed to make the tests pass, DO NOT delete tests, skip
-   tests, or change the tests specifically to make them pass (unless there
-   is a bug in the test).  This is cheating, and it is bad.  You should only
-   be modifying the test if it is incorrect or if the test is no longer
-   relevant.  In almost all cases, you should be fixing the code that is
-   being tested, or updating the tests to match a refactored implementation.
+   tests, or change the tests specifically to make them pass (unless there is a
+   bug in the test). This is cheating, and it is bad. You should only be
+   modifying the test if it is incorrect or if the test is no longer relevant.
+   In almost all cases, you should be fixing the code that is being tested, or
+   updating the tests to match a refactored implementation.
 
 6. When dealing with dates and times or timestamps, always use, display, and
-   store UTC.  Set the local timezone to UTC on startup.  If the user needs
-   to see the time in a different timezone, store the user's timezone in a
-   separate field and convert the UTC time to the user's timezone when
-   displaying it.  For internal use and internal applications and
-   administrative purposes, always display UTC.
+   store UTC. Set the local timezone to UTC on startup. If the user needs to see
+   the time in a different timezone, store the user's timezone in a separate
+   field and convert the UTC time to the user's timezone when displaying it. For
+   internal use and internal applications and administrative purposes, always
+   display UTC.
 
 7. Always write tests, even if they are extremely simple and just check for
-   correct syntax (ability to compile/import). If you are writing a new
-   feature, write a test for it.  You don't need to target complete
-   coverage, but you should at least test any new functionality you add.  If
-   you are fixing a bug, write a test first that reproduces the bug, and
-   then fix the bug in the code.
+   correct syntax (ability to compile/import). If you are writing a new feature,
+   write a test for it. You don't need to target complete coverage, but you
+   should at least test any new functionality you add. If you are fixing a bug,
+   write a test first that reproduces the bug, and then fix the bug in the code.
 
-8. When implementing new features, be aware of potential side-effects (such
-   as state files on disk, data in the database, etc.) and ensure that it is
+8. When implementing new features, be aware of potential side-effects (such as
+   state files on disk, data in the database, etc.) and ensure that it is
    possible to mock or stub these side-effects in tests.
 
-9. Always use structured logging.  Log any relevant state/context with the
-   messages (but do not log secrets).  If the log stream is not a terminal,
+9. Always use structured logging. Log any relevant state/context with the
+   messages (but do not log secrets). If the log stream is not a terminal,
    output the structured logs in jsonl format.
 
 10. Avoid using bare strings or numbers in code, especially if they appear
-    anywhere more than once.  Always define a constant (usually at the top
-    of the file) and give it a descriptive name, then use that constant in
-    the code instead of the bare string or number.
+    anywhere more than once. Always define a constant (usually at the top of the
+    file) and give it a descriptive name, then use that constant in the code
+    instead of the bare string or number.
 
 11. You do not need to summarize your changes in the chat after making them.
-    Making the changes and committing them is sufficient.  If anything out
-    of the ordinary happened, please explain it, but in the normal case
-    where you found and fixed the bug, or implemented the feature, there is
-    no need for the end-of-change summary.
+    Making the changes and committing them is sufficient. If anything out of the
+    ordinary happened, please explain it, but in the normal case where you found
+    and fixed the bug, or implemented the feature, there is no need for the
+    end-of-change summary.
 
-12. Do not create additional files in the root directory of the project
-    without asking permission first. Configuration files, documentation, and
-    build files are acceptable in the root, but source code and other files
-    should be organized in appropriate subdirectories.
+12. Do not create additional files in the root directory of the project without
+    asking permission first. Configuration files, documentation, and build files
+    are acceptable in the root, but source code and other files should be
+    organized in appropriate subdirectories.
 
-13. Pre-1.0: NEVER write database migrations. There are no live databases
-    anywhere — every user's local index can be rebuilt from a fresh full
-    backup. To change the schema, edit `internal/database/schema/001.sql`
-    (and any code that touches the affected tables) directly; do not add new
-    numbered schema files. Those numbered files and the `schema_migrations`
-    table they populate only bootstrap a fresh database — they are not an
-    upgrade path. The local index is disposable until 1.0 ships and is
-    tagged; once 1.0 is tagged that clause expires and the question of
-    upgrading existing indexes returns. See [`docs/DATAMODEL.md`](docs/DATAMODEL.md)
-    for the full explanation.
+13. Pre-1.0: NEVER add a database migration. Migrations are supported, but
+    nothing is installed anywhere yet, so there is nothing to migrate. To change
+    the schema, edit `internal/database/schema/001.sql` (and any code that
+    touches the affected tables) directly. After 1.0, each schema change is a
+    new numbered file in that directory and a released file is never edited; an
+    existing local database is then migrated when vaultik is updated. Before
+    1.0, a local database left on an older schema is deleted and re-created by a
+    full backup. See [`docs/DATAMODEL.md`](docs/DATAMODEL.md#schema-migrations).
 
+14. Never use `git add -A`. Stage only the files you intentionally changed.
+
+15. Commit messages carry no attribution or advertising trailers for the tool
+    that helped write the code, or for its vendor: the owner is the sole author
+    of code written with a tool.
+
+16. Run the whole test suite with `make test` every time, and read its full
+    output. Never run `go test`, a single test or a single package, and never
+    grep the output.
+
+17. Do not stop working on a task until the definition of done given in the
+    initial instruction is met: all of the work, not part or most of it.
+
+18. For estimates: backing up over 2.5Gbit/s ethernet to an S3 server backed by
+    2000MB/sec SSD takes about 4 seconds per gigabyte.

CLAUDE.md

@@ -1,44 +0,0 @@
-# Rules
-
-Read the rules in AGENTS.md and follow them.
-
-# Memory
-
-* Claude is an inanimate tool.  The spam that Claude attempts to insert into
-  commit messages (which it erroneously refers to as "attribution") is not
-  attribution, as I am the sole author of code created using Claude.  It is
-  corporate advertising for Anthropic and is therefore completely
-  unacceptable in commit messages.
-
-* NEVER use `git add -A`.  Always add only the files you intentionally
-  changed.
-
-* Tests should always be run before committing code.  No commits should be
-  made that do not pass tests.
-
-* Code should always be formatted before committing.  Do not commit
-  unformatted code.
-
-* Code should always be linted before committing.  Do not commit
-  unlinted code.
-
-* The test suite is fast and local.  When running tests, don't run
-  individual parts of the test suite, always run the whole thing by running
-  "make test".
-
-* Do not stop working on a task until you have reached the definition of
-  done provided to you in the initial instruction.  Don't do part or most of
-  the work, do all of the work until the criteria for done are met.
-
-* We do not need to support migrations; schema upgrades can be handled by
-  deleting the local state file and doing a full backup to re-create it.
-
-* When testing on a 2.5Gbit/s ethernet to an s3 server backed by 2000MB/sec SSD, 
-  estimate about 4 seconds per gigabyte of backup time.
-
-* When running tests, don't run individual tests, or grep the output.  run
-  the entire test suite every time and read the full output.
-
-* When running tests, don't run individual tests, or try to grep the output.
-  never run "go test".  only ever run "make test" to run the full test
-  suite, and examine the full output.

Dockerfile

@@ -1,96 +1,97 @@
-# This file has no lint stage, deliberately.
-#
-# Linting lives in Dockerfile.lint, built by script/lint, and
-# script/cibuild builds both. A lint stage here would have to either
-# shell out to `make lint` -- which is now `docker build`, so
-# docker-in-docker inside a BuildKit step with no daemon -- or call
-# golangci-lint directly, which would mean a second, independently
-# bumpable digest pin for the linter alongside the one in
-# Dockerfile.lint. Two pins for one tool is the drift that
-# https://git.eeqj.de/sneak/vaultik/issues/78 was filed over. See
-# https://git.eeqj.de/sneak/vaultik/issues/113 for the ruling.
-#
-# Consequence, stated rather than left to be discovered: script/docker
-# builds this file only and therefore does not lint. `make fmt-check`
-# and `make test` still run here, so what a green build of this file
-# means is "formatted, tested, and it compiles" -- the lint verdict
-# comes from script/lint or script/cibuild.
+# Lint phase. The linter is invoked directly rather than through `make
+# lint` or `script/lint`, which are themselves a docker build and would
+# recurse into a daemon that does not exist in a build step.
+# golangci/golangci-lint:v2.14.0, 2026-10-05
+FROM golangci/golangci-lint:v2.14.0@sha256:ad862ba6b3798cbe0fd9fd7408d498fd74fbd2623a92406b2fd3898faf0bf98f AS lint
+WORKDIR /src
+COPY go.mod go.sum ./
+# Without the time limit a stalled connection to the module proxy holds
+# this step, and every build on the host sharing it, until the kernel
+# drops the connection, which can take an hour. The limit is well above
+# the three minutes a working download can take on a busy build host.
+RUN for attempt in 1 2 3; do \
+        timeout 300 go mod download && exit 0; \
+        echo "go mod download attempt $attempt failed or ran over 300s" >&2; \
+    done; \
+    echo "module download failed: giving up after 3 attempts" >&2; \
+    exit 1
+COPY . .
+# `golangci-lint run` silently ignores an unknown top-level key in
+# .golangci.yml, such as a misspelt `linters:`; `config verify` fails on it.
+RUN golangci-lint config verify --config .golangci.yml
+RUN golangci-lint run --config .golangci.yml ./...
 
-# Build stage
+# Test phase. -race needs cgo and so a C compiler, which the Debian Go
+# image ships and the alpine one does not.
+# golang:1.26.1 (Debian trixie), 2026-10-05
+FROM golang:1.26.1@sha256:cd78d88e00afadbedd272f977d375a6247455f3a4b1178f8ae8bbcb201743a8a AS test
+WORKDIR /src
+COPY go.mod go.sum ./
+# Limited and retried as in the lint phase above.
+RUN for attempt in 1 2 3; do \
+        timeout 300 go mod download && exit 0; \
+        echo "go mod download attempt $attempt failed or ran over 300s" >&2; \
+    done; \
+    echo "module download failed: giving up after 3 attempts" >&2; \
+    exit 1
+COPY . .
+RUN go test -timeout 90s -race -cover ./... || \
+    { echo "--- Rerunning with -v for details ---"; \
+      go test -timeout 90s -race -v ./...; exit 1; }
+
+# Build stage. Nothing is wanted from either phase above; the copies
+# are what make BuildKit build them first, so this stage cannot run
+# unless lint and test passed.
 # golang:1.26.1-alpine, 2026-03-17
 FROM golang:1.26.1-alpine@sha256:2389ebfa5b7f43eeafbd6be0c3700cc46690ef842ad962f6c5bd6be49ed82039 AS builder
-
-# Build tooling: make, plus a C toolchain because `go test -race` needs cgo.
-# The sqlite driver is pure Go (modernc.org/sqlite), so no sqlite library or
-# CLI is required.
-RUN apk add --no-cache make build-base
-
+COPY --from=lint /src/go.sum /dev/null
+COPY --from=test /src/go.sum /dev/null
+RUN apk add --no-cache git
+# A tar-stream context keeps the sender's file owners, which git refuses.
+RUN git config --system --add safe.directory /src
 WORKDIR /src
-
-# Copy go mod files first for better layer caching
 COPY go.mod go.sum ./
-RUN go mod download
-
-# Copy source code
+# Limited and retried as in the lint phase above.
+RUN for attempt in 1 2 3; do \
+        timeout 300 go mod download && exit 0; \
+        echo "go mod download attempt $attempt failed or ran over 300s" >&2; \
+    done; \
+    echo "module download failed: giving up after 3 attempts" >&2; \
+    exit 1
 COPY . .
 
-# Run the format check and the tests.
-#
-# CHECK_EPOCH must stay immediately above these RUNs. These layers are
-# keyed on its value, so they are cache-eligible only for a value
-# already built against this same tree. script/cibuild and script/docker
-# each pass a fresh value on every invocation, which is what makes their
-# green mean the checks really executed.
-#
-# The value is expanded into each check command rather than left to a
-# bare declaration, so the cache miss does not depend on BuildKit's
-# unreferenced-ARG handling staying as it is. It also puts the epoch in
-# the build log, where a reader can see the layer was keyed fresh.
-#
-# The guard is what makes a build that omits --build-arg fail instead of
-# lie. An unset ARG is an empty string, and an empty string is a
-# perfectly stable cache key: without the guard the first such build
-# runs the checks and every one after it on an unchanged tree replays
-# these layers from cache, executes nothing, and still exits 0. Failed
-# steps are never cached, so the guard fails on EVERY invocation rather
-# than once -- a bare `docker build .` is a loud error, not a quiet
-# green. Do not give CHECK_EPOCH a default value; a default would
-# satisfy the guard with a constant and restore the hole.
-#
-# Everything above this line (apk, go.mod, `go mod download`) is
-# deliberately outside the busted range and keeps caching.
-ARG CHECK_EPOCH
-RUN [ -n "$CHECK_EPOCH" ] || exit 1
-RUN echo "check epoch: ${CHECK_EPOCH}" && make fmt-check
-RUN echo "check epoch: ${CHECK_EPOCH}" && make test
+# The VERSION build arg when one is given, otherwise
+# `git describe --tags --always` on the .git in the build context. The
+# commit and its date always come from that .git. With .git present, a
+# version that is still empty, dev or unknown, or a commit or date that
+# is unknown, fails the build: git is missing or could not read the
+# checkout, as when .git is a file pointing outside the context. A
+# context without .git, such as a source export, stamps "dev" and an
+# "unknown" commit and date.
+ARG VERSION
+RUN VERSION="${VERSION:-$(git describe --tags --always || echo dev)}"; \
+    commit="$(git rev-parse HEAD || echo unknown)"; \
+    commit_date="$(git show -s --format=%cs HEAD || echo unknown)"; \
+    if [ -e .git ]; then \
+        case "$VERSION" in ""|dev|unknown) \
+            echo "version is '$VERSION' although .git is present" >&2; \
+            exit 1 ;; \
+        esac; \
+        if [ "$commit" = unknown ] || [ "$commit_date" = unknown ]; then \
+            echo "commit is '$commit' and its date '$commit_date'" \
+                "although .git is present" >&2; \
+            exit 1; \
+        fi; \
+    fi; \
+    globals=sneak.berlin/go/vaultik/internal/globals; \
+    CGO_ENABLED=0 go build -trimpath \
+        -ldflags="-s -w -X ${globals}.Version=${VERSION} \
+            -X ${globals}.Commit=${commit} \
+            -X ${globals}.CommitDate=${commit_date}" \
+        -o /vaultik ./cmd/vaultik
 
-# Version, commit and build date are computed on the host by
-# script/docker and script/cibuild (where .git exists) and passed in as
-# build args. The build context excludes .git (see .dockerignore), so
-# the build cannot derive them itself: it used to try, with `git
-# rev-parse` inside this stage, and always got "unknown". VERSION comes
-# from script/version, the source of truth shared with the Makefile, so
-# it carries the same tag / dev-<sha> / -dirty rules and a Docker image
-# reports the same string a local build of the same tree would.
-#
-# The defaults are the fallback for a bare `docker build .` that passes
-# none of them: an unset arg would otherwise stamp an empty string and
-# produce an image that cannot report its own version, commit or date.
-# They match what an out-of-git build reports elsewhere.
-#
-# These ARGs sit here, after the checks, rather than at the top of the
-# stage: every commit changes their values, and a value change
-# invalidates all layers below the ARG. Declared up top they would bust
-# `go mod download`; here they only rekey this build layer, which the
-# COPY of the sources above already rebuilds on any change anyway.
-ARG VERSION=dev
-ARG COMMIT=unknown
-ARG COMMIT_DATE=unknown
-
-# Build (pure Go, no CGO required since we use modernc.org/sqlite)
-RUN CGO_ENABLED=0 go build -ldflags "-X 'sneak.berlin/go/vaultik/internal/globals.Version=${VERSION}' -X 'sneak.berlin/go/vaultik/internal/globals.Commit=${COMMIT}' -X 'sneak.berlin/go/vaultik/internal/globals.CommitDate=${COMMIT_DATE}'" -o /vaultik ./cmd/vaultik
-
-# Runtime stage
+# Runtime stage, and the last one: a plain `docker build .` builds this
+# stage's chain and nothing else.
 # alpine:3.21, 2026-02-25
 FROM alpine:3.21@sha256:c3f8e73fdb79deaebaa2037150150191b9dcbfba68b4a46d70103204c53f4709
 

Dockerfile.lint

@@ -1,104 +0,0 @@
-# Lint image.
-#
-# Every lint run in this repo happens inside this image, invoked through
-# script/lint, and linting is a BUILD STEP rather than a container
-# command: a successful build of this file IS a clean lint. That shape
-# also works where the docker daemon is remote and bind mounts are
-# impossible, which `docker run` against a mounted worktree does not.
-#
-# This FROM line is the single source of truth for the linter version in
-# this repo. Nothing else pins golangci-lint: the product Dockerfile has
-# no lint stage, deliberately, so there is no second digest to bump and
-# no pair of pins that can drift apart. Bump the tag AND the digest here
-# and nowhere else.
-#
-# Note for readers coming from REPO_POLICIES.md: that document still
-# describes the older pattern, a lint stage inside the product
-# Dockerfile wired up with `COPY --from=lint /src/go.sum /dev/null`.
-# That pattern is superseded here by the owner's ruling recorded in
-# https://git.eeqj.de/sneak/vaultik/issues/113 -- lint runs in its own
-# image, per run, with its own cache and its own lock, which is what
-# makes concurrent runs on one host safe. The policy text is org-wide
-# and is being amended separately; this file is what this repo does.
-#
-# golangci/golangci-lint:v2.12.2, 2026-08-10
-FROM golangci/golangci-lint:v2.12.2@sha256:5cceeef04e53efe1470638d4b4b4f5ceefd574955ab3941b2d9a68a8c9ad5240
-
-WORKDIR /src
-
-# Copy the dependency manifests first so the module download layer stays
-# cached until they change. Everything above the ARG below is cacheable
-# on purpose; a cold module download on every lint would make the inner
-# loop unusable and buys nothing, because it is not what the gate is
-# asserting.
-COPY go.mod go.sum ./
-RUN go mod download
-
-COPY . .
-
-# Force the check layers to execute on every invocation.
-#
-# CHECK_EPOCH must stay immediately above the RUNs below. Those layers
-# are keyed on its value, so they are cache-eligible only for a value
-# already built against this same tree; script/lint and script/cibuild
-# each pass a fresh value on every invocation, which is what makes their
-# green mean the linter really ran. Without it, `docker build -f
-# Dockerfile.lint .` on an unchanged tree exits 0 in well under a second
-# having linted nothing.
-#
-# The value is expanded into each check command itself rather than left
-# to a bare declaration, so the cache miss does not depend on BuildKit's
-# unreferenced-ARG handling staying as it is. It also puts the epoch in
-# the build log, where a reader can see the layer was keyed fresh.
-#
-# The guard is what makes a build that omits --build-arg fail instead of
-# lie. An unset ARG is an empty string, and an empty string is a
-# perfectly stable cache key: without the guard the first such build
-# lints and every one after it on an unchanged tree replays this layer,
-# executes nothing, and still exits 0. Failed steps are never cached, so
-# the guard fails on EVERY invocation rather than once. Do not give
-# CHECK_EPOCH a default value; a default would satisfy the guard with a
-# constant and restore the hole.
-ARG CHECK_EPOCH
-RUN [ -n "$CHECK_EPOCH" ] || exit 1
-
-# Validate .golangci.yml before linting with it.
-#
-# This is not belt-and-braces; it closes a hole that `golangci-lint run`
-# leaves wide open. `run` rejects YAML it cannot PARSE, but it silently
-# IGNORES an unknown top-level KEY. Renaming `linters:` to `linterz:` --
-# one character -- discards `default: all`, the whole disable list and
-# every threshold, leaves only golangci-lint's small default linter set
-# running, and exits 0 reporting `0 issues.` on a tree the real config
-# fails. Demonstrated on this repo at this pin, recorded on
-# https://git.eeqj.de/sneak/vaultik/pulls/114: with a planted
-# over-length line, `script/lint` exits 1 naming the `revive` finding
-# with `linters:` and exits 0 with `linterz:`. A set-but-ineffective
-# config quietly falling back to defaults is precisely the false-green
-# class this gate exists to eliminate, so it must not sit in the gate's
-# own configuration.
-#
-# `config verify` catches it, and it does so OFFLINE at this pinned
-# version -- verified, not assumed. Under `docker run --network none`
-# against the pinned digest it exits 0 on this repo's config and exits 3
-# on the `linterz:` variant with `additional properties 'linterz' not
-# allowed`. An earlier revision of this file asserted the opposite, that
-# the schema is fetched over live HTTPS from an unpinned URL, and used
-# that to justify omitting this line. That claim was false at v2.12.2;
-# the schema is embedded. If a future bump reintroduces a network fetch
-# the failure is loud and this comment is where to record it.
-#
-# It is keyed on CHECK_EPOCH, like the lint run below, so it executes on
-# every invocation. Content-addressing alone would arguably be enough --
-# .golangci.yml arrives through `COPY . .`, so a cache hit here implies
-# a byte-identical config was validated when the layer really ran. That
-# argument is exactly the one that would also excuse caching the lint
-# layer, and this repo has ruled it insufficient: a cached check layer
-# checks nothing, and the cost of being wrong is silent. Forcing it costs
-# milliseconds and puts the epoch in the log, where a reader can see that
-# this validation ran rather than being replayed.
-RUN echo "check epoch: ${CHECK_EPOCH}" && \
-    golangci-lint config verify --config .golangci.yml
-
-RUN echo "check epoch: ${CHECK_EPOCH}" && \
-    golangci-lint run --config .golangci.yml ./...

Makefile

@@ -1,7 +1,10 @@
 .PHONY: all bootstrap setup check test lint lint-fix fmt fmt-check build clean deps test-coverage local install release release-snapshot docker hooks
 
-# Version number, derived from git by script/version -- the tag when
-# HEAD is on one, otherwise dev-<sha>. This used to be a hardcoded
+# Where script/bootstrap installs Go when the host has none.
+export PATH := $(PATH):$(CURDIR)/.tool/go/bin
+
+# Version number, derived from git by script/version (`git describe
+# --tags --always --dirty`). This used to be a hardcoded
 # constant, which meant every local build claimed to be a release that
 # had never been tagged.
 VERSION := $(shell script/version)
@@ -40,9 +43,10 @@ setup:
 check:
 	@script/check
 
-# Run tests only. This runs the ENTIRE suite -- there is no separate
-# integration target and no build-tagged subset held back. In
-# particular internal/vaultik/integration_test.go, which does full
+# Run tests only, by building the test phase of the Dockerfile. This
+# runs the ENTIRE suite -- there is no separate integration target and
+# no build-tagged subset held back. In particular
+# internal/vaultik/integration_test.go, which does full
 # chunk -> pack -> encrypt -> upload -> restore round-trips, runs here.
 # A `test-integration` target used to exist and was removed: no file in
 # the repo carried a build tag, so `-tags=integration` selected nothing
@@ -87,17 +91,17 @@ clean:
 	go clean
 
 # Install dependencies. The linter is deliberately not installed here:
-# script/lint lints by building Dockerfile.lint, whose FROM line is the
-# single source of truth for the linter version. A second, separately
-# pinned copy on PATH could drift from it and make a local `make lint`
-# disagree with CI.
+# script/lint lints by building the lint phase of the Dockerfile, whose
+# FROM line is the single source of truth for the linter version. A
+# second, separately pinned copy on PATH could drift from it and make a
+# local `make lint` disagree with CI.
 deps:
 	go mod download
 
-# Run tests with coverage. -count=1 for the same reason script/test
-# uses it: without it an unchanged package is served from Go's test
-# result cache, and a coverage profile assembled from cached results
-# describes a run that did not happen.
+# Run tests with coverage, on the host. -count=1 because without it an
+# unchanged package is served from Go's test result cache, and a
+# coverage profile assembled from cached results describes a run that
+# did not happen.
 test-coverage:
 	go test -v -count=1 -coverprofile=coverage.out ./...
 	go tool cover -html=coverage.out -o coverage.html

REPO_POLICIES.md

@@ -1,6 +1,6 @@
 ---
 title: Repository Policies
-last_modified: 2026-07-06
+last_modified: 2026-10-04
 ---
 
 This document covers repository structure, tooling, and workflow standards. Code
@@ -60,17 +60,28 @@ style conventions are in separate documents:
   prerequisite since nvm requires bash. yarn is then pinned via
   `corepack prepare yarn@<version> --activate`. Never install "latest" or "lts";
   always exact versions. `script/cibuild` runs the CI build: it changes to the
-  repo root and runs `docker build .`; the Gitea workflow calls it. Four further
-  scripts are our own extensions to the standard: `script/check` runs
-  `script/test`, `script/lint`, and `script/fmt-check`; `script/precommit` is
-  what the git pre-commit hook runs, and it calls `script/check`;
-  `script/install-precommit` installs the git pre-commit hook (the `make hooks`
-  target shims to it); and `script/projectname` (literally that filename) simply
-  outputs the project's name. Scripts that need the name call
-  `script/projectname` — e.g. `script/docker` assembles its image tag from it —
-  so those scripts stay byte-identical across all repos. Repo-type-specific
-  pre-commit extras (e.g. `go mod tidy` verification in Go repos) belong in
-  `script/precommit`, not in the hook itself. Model scripts are at
+  repo root, runs `script/bootstrap`, runs `script/check`, and builds the image
+  with the version; the Gitea workflow calls it. **`script/cibuild` runs
+  `script/bootstrap` first**, because the workflow checks out the repo and runs
+  nothing else, while `script/fmt-check` runs the formatter on the host: on a
+  pristine checkout with nothing installed the run dies there, after the
+  containerised gates have passed. **The bootstrap alone is not enough**:
+  `script/bootstrap` installs node and yarn under nvm and leaves neither on the
+  `PATH` of the shell that called it, so a bare `yarn` still exits 127. The host
+  entrypoints that need yarn — `script/fmt` and `script/fmt-check` — therefore
+  source nvm for the pinned node version before invoking it, exactly as
+  `script/bootstrap`'s own install step does. A runner carrying nothing but
+  docker and git then gets through `script/check`. Four further scripts are our
+  own extensions to the standard: `script/check` runs `script/test`,
+  `script/lint` and `script/fmt-check`; `script/precommit` is what the git
+  pre-commit hook runs, and it calls `script/check`; `script/install-precommit`
+  installs the git pre-commit hook (the `make hooks` target shims to it); and
+  `script/projectname` (literally that filename) simply outputs the project's
+  name. Scripts that need the name call `script/projectname` — e.g.
+  `script/docker` assembles its image tag from it — so those scripts stay
+  byte-identical across all repos. Repo-type-specific pre-commit extras (e.g.
+  `go mod tidy` verification in Go repos) belong in `script/precommit`, not in
+  the hook itself. Model scripts are at
   `https://git.eeqj.de/sneak/prompts/raw/branch/main/script/<name>`. The README
   must document the provided scripts in an **Entrypoints** section (see the
   README requirements below).
@@ -89,87 +100,198 @@ style conventions are in separate documents:
   contributor should be able to understand the entire development workflow by
   reading the Makefile.
 
-- Every repo should have a `Dockerfile`. All Dockerfiles must run `make check`
-  as a build step so the build fails if the branch is not green. For non-server
-  repos, the Dockerfile should bring up a development environment and run
-  `make check`. For server repos, `make check` should run as an early build
-  stage before the final image is assembled. Dockerfiles install development
-  prerequisites by running `script/bootstrap` rather than duplicating installs
-  inline; COPY `script/` and the dependency manifests (`package.json` +
-  `yarn.lock`, `go.mod` + `go.sum`, etc.) before running it so the bootstrap
-  layer stays cached until dependencies change.
+- Every repo should have a `Dockerfile`, and it carries the repo's gates: a
+  `lint` phase and a `test` phase, with the final stage depending on both so the
+  image cannot be built unless they pass. For non-server repos the final stage
+  brings up a development environment; for server repos it is the runtime image.
+  The gate phases and the build stage start from their pinned base images and
+  install what those images lack either inline, as the canonical Go `Dockerfile`
+  below does for `git`, or by running `script/bootstrap`, as the `prompts`
+  repo's own `Dockerfile` does for its yarn packages. The development
+  environment stage installs development prerequisites by running
+  `script/bootstrap` rather than duplicating its installs inline. A stage that
+  runs `script/bootstrap` COPYs `script/` and the dependency manifests
+  (`package.json` + `yarn.lock`, `go.mod` + `go.sum`, etc.) before running it.
 
-- **Dockerfiles must use a separate lint stage for fail-fast feedback.** Go
-  repos use a multistage build where linting runs in an independent stage based
-  on the `golangci/golangci-lint` image (pinned by hash). This stage runs
-  `make fmt-check` and `make lint` before the full build begins. The build stage
-  then declares an explicit dependency on the lint stage via
-  `COPY --from=lint /src/go.sum /dev/null`, which forces BuildKit to complete
-  linting before proceeding to compilation and tests. This ensures lint failures
-  surface in seconds rather than minutes, without blocking on dependency
-  download or compilation in the build stage.
+- **Linting and testing run in Docker, as phases of the `Dockerfile`.** There is
+  no separate lint file. `script/lint` and `script/test` each build one phase
+  and nothing else:
 
-    The standard pattern for a Go repo Dockerfile is:
+    ```sh
+    docker build --no-cache --target lint -t "$(script/projectname)-lint" .
+    docker build --no-cache --target test -t "$(script/projectname)-test" .
+    ```
+
+    **A stage that is not the last one in the file is built only when the final
+    stage's chain depends on it, or when `--target` names it.** That is why the
+    two gates are always invoked by name here, and why the final stage carries a
+    `COPY --from=` of a harmless file from each of them: without that edge a
+    plain `docker build .` builds the last stage alone and exits 0 having linted
+    and tested nothing.
+
+    **Every `docker build` in `script/` is tagged**, here and in
+    `script/cibuild` and `script/docker`. An untagged build leaves a dangling
+    image behind on every invocation, on every developer host and every CI
+    runner; a tagged one replaces the previous image.
+
+    Inside a phase the tool is invoked directly — `golangci-lint`, `go test`,
+    `eslint`, `prettier` — never through `make lint` or `script/test`, which are
+    themselves a `docker build` and would recurse into a daemon that does not
+    exist in a build step. Formatting is the exception and stays on the host:
+    `script/fmt` writes the working tree, and `script/fmt-check` is its
+    read-only twin.
+
+    **No lint verdict may come from a host invocation of the linter.** On a
+    shared host golangci-lint reads a result cache keyed on file content rather
+    than location, so a second checkout of the same content is served the first
+    one's findings, and a host-global lock in `$TMPDIR` makes concurrent runs
+    exit non-zero with `parallel golangci-lint is running` — a status a caller
+    cannot tell from real findings. Both have produced wrong verdicts in this
+    org, in both directions. A container has its own cache, its own `TMPDIR` and
+    a digest-pinned binary, so neither is reachable.
+
+- **Any build that runs checks is built with `--no-cache`.** Docker invalidates
+  a `COPY` layer only when the copied content changes, so on an unchanged tree
+  the check `RUN` is served from cache, nothing executes, and the build still
+  exits 0. Every `docker build` in `script/` therefore passes `--no-cache`:
+  `script/lint`, `script/test`, `script/cibuild` and `script/docker` are the
+  four, and there is no fifth — `script/check` runs the two gate phases and
+  `script/fmt-check`, and builds no image of its own. A bare `docker build .` is
+  not evidence that anything ran: a sub-second build reporting success is a
+  cache hit, not a result. Never invalidate by pruning — `docker builder prune`
+  and friends destroy a build cache shared with every other build on the host.
+  When a check is added or changed, prove it works by planting a defect it must
+  catch and watching the run fail on it, then revert the defect. A green run
+  alone shows neither that the check ran nor that it covers what it should.
+
+- **The gate phases are separate stages, and the build stage depends on both.**
+  The lint phase is based on the `golangci/golangci-lint` image (pinned by
+  hash), so lint failures surface in seconds rather than after a full compile,
+  and the test phase is based on the Debian Go image. The canonical Go repo
+  `Dockerfile`:
 
     ```dockerfile
-    # Lint stage — fast feedback on formatting and lint issues
+    # Lint phase
     # golangci/golangci-lint:v2.x.x, YYYY-MM-DD
     FROM golangci/golangci-lint@sha256:... AS lint
     WORKDIR /src
     COPY go.mod go.sum ./
     RUN go mod download
     COPY . .
-    RUN make fmt-check
-    RUN make lint
+    RUN golangci-lint run --config .golangci.yml ./...
 
-    # Build stage
-    # golang:1.x-alpine, YYYY-MM-DD
-    FROM golang@sha256:... AS builder
+    # Test phase. -race needs cgo and so a C compiler, which the Debian Go
+    # image ships and the alpine one does not.
+    # golang:1.x, YYYY-MM-DD
+    FROM golang@sha256:... AS test
     WORKDIR /src
-
-    # Force BuildKit to run the lint stage before proceeding
-    COPY --from=lint /src/go.sum /dev/null
-
     COPY go.mod go.sum ./
     RUN go mod download
     COPY . .
-    RUN make test
+    RUN go test -timeout 90s -race -cover ./... || \
+        { echo "--- Rerunning with -v for details ---"; \
+          go test -timeout 90s -race -v ./...; exit 1; }
 
-    ARG VERSION=dev
-    RUN CGO_ENABLED=0 go build -trimpath \
-        -ldflags="-s -w -X main.Version=${VERSION}" \
-        -o /app ./cmd/app/
+    # Build stage. Nothing is wanted from either phase above; the copies
+    # are what make BuildKit build them first, so this stage cannot run
+    # unless lint and test passed.
+    # golang:1.x-alpine, YYYY-MM-DD
+    FROM golang@sha256:... AS builder
+    COPY --from=lint /src/go.sum /dev/null
+    COPY --from=test /src/go.sum /dev/null
+    RUN apk add --no-cache git
+    # A tar-stream context keeps the sender's file owners, which git refuses.
+    RUN git config --system --add safe.directory /src
+    WORKDIR /src
+    COPY go.mod go.sum ./
+    RUN go mod download
+    COPY . .
 
-    # Runtime stage
+    # The VERSION build arg when one is given, otherwise
+    # `git describe --tags --always` on the .git in the build context. With
+    # .git present, a version that is still empty, dev or unknown fails the
+    # build: git is missing or could not read the checkout.
+    ARG VERSION
+    RUN VERSION="${VERSION:-$(git describe --tags --always)}"; \
+        if [ -e .git ]; then \
+            case "$VERSION" in ""|dev|unknown) \
+                echo "version is '$VERSION' although .git is present" >&2; \
+                exit 1 ;; \
+            esac; \
+        fi; \
+        CGO_ENABLED=0 go build -trimpath \
+            -ldflags="-s -w -X main.Version=${VERSION}" \
+            -o /app ./cmd/app/
+
+    # Runtime stage, and the last one
     FROM alpine@sha256:...
     COPY --from=builder /app /usr/local/bin/app
     ENTRYPOINT ["app"]
     ```
 
     Key points:
-    - The lint stage uses the `golangci/golangci-lint` image directly (it
-      includes both Go and the linter), so there is no need to install the
-      linter separately.
-    - `COPY --from=lint /src/go.sum /dev/null` is a no-op file copy that creates
-      a stage dependency. BuildKit runs stages in parallel by default; without
-      this line, the build stage would not wait for lint to finish and a lint
-      failure might not fail the overall build.
+    - The lint phase uses the `golangci/golangci-lint` image directly (it has
+      both Go and the linter), so nothing needs installing.
+    - `COPY --from=<phase> /src/go.sum /dev/null` is a no-op copy whose only
+      purpose is the ordering edge. BuildKit runs stages in parallel by default,
+      and a stage nothing depends on is not built at all, so without these two
+      lines a red gate would not fail the build.
+    - Keep the runtime stage last, and if you add a stage after it, give it the
+      same two copies. A plain `docker build .` builds the last stage's chain
+      and nothing else.
     - If the project uses `//go:embed` directives that reference build artifacts
-      (e.g. a web frontend compiled in a separate stage), the lint stage must
+      (e.g. a web frontend compiled in a separate stage), the lint phase must
       create placeholder files so the embed directives resolve. Example:
       `RUN mkdir -p web/dist && touch web/dist/index.html web/dist/style.css`.
-      The lint stage should not depend on the actual build output — it exists to
-      fail fast.
-    - If the project requires CGO or system libraries for linting (e.g.
-      `vips-dev`), install them in the lint stage with `apk add`.
-    - The build stage runs `make test` after compilation setup. Tests run in the
-      build stage, not the lint stage, because they may require compiled
-      artifacts or heavier dependencies.
+    - If the project requires CGO or system libraries for linting, install them
+      in the lint phase. The `golangci/golangci-lint` image is Debian-based and
+      has no `apk`, so install with `apt-get` under the Debian package name
+      (`libvips-dev`, where alpine says `vips-dev`), and delete the package
+      lists in the same `RUN`, so the layer does not keep them:
+
+        ```dockerfile
+        RUN apt-get update \
+            && apt-get install -y --no-install-recommends libvips-dev \
+            && rm -rf /var/lib/apt/lists/*
+        ```
+
+    - `.dockerignore` lets `.git` into the build context. It keeps out every git
+      `config` at any depth (`**/.git/config`, `**/.git/modules/**/config`): the
+      repository's own, each submodule's under `.git/modules/`, and that of a
+      submodule keeping its own `.git` directory. `git describe` does not need
+      them, and each can hold a credential: a password in a remote URL, or the
+      token the CI checkout step stores there. A submodule whose name has a
+      `config` segment (`config`, `deploy/config`, `config/lib`) loses its whole
+      git directory to `**/.git/modules/**/config`, and Go's version stamping
+      then fails the build: give it a name without that segment
+      (`git submodule add --name`). The stage that compiles has `git` (the
+      Debian Go image has it; an alpine one needs `apk add --no-cache git`) and
+      takes the version from the `VERSION` build argument when one is given,
+      otherwise from `git describe --tags --always`. That gives the tag on a
+      tagged commit; on a later commit, the tag, the number of commits since it
+      and the short commit (`v1.2.3-4-gabc1234`); and the short commit when no
+      tag is reachable. The stage that compiles also marks its working directory
+      safe for git (`git config --system --add safe.directory /src`): a context
+      sent as a tar stream keeps the sender's file owners, and git refuses a
+      checkout owned by another user, so the version would come out empty.
+      `ARG VERSION` has no default, and the build fails if the context carries
+      `.git` and the version still comes out empty, `dev` or `unknown`. A plain
+      `docker build .` with no build arguments must succeed; a Dockerfile that
+      refuses an empty build argument drops that refusal and keeps the argument.
 
 - Every repo should have a Gitea Actions workflow (`.gitea/workflows/`) that
-  runs `script/cibuild` (which runs `docker build .`) on push. Since the
-  Dockerfile already runs `make check`, a successful build implies all checks
-  pass.
+  runs `script/cibuild` on push, and checks out the repo as its only other step.
+  That script bootstraps, runs the gate phases, and then builds the image, so a
+  successful run means every check passed; a bare `docker build .` does not
+  carry the same guarantee, because its gate phases may come from the cache. The
+  image build is uncached and so runs the gate phases a second time. That is the
+  price of the rule above, and it is worth paying: the image that ships is built
+  from a run of its own gates rather than from a cache entry. A separate
+  workflow limited to `main` by a `branches` list under `on: push` cannot be
+  checked by review: to try a change to it, add the feature branch to that list
+  and push, then remove the branch from the list again before merging. Keep any
+  job in it that publishes behind `if: github.ref_name == 'main'`, so the run
+  from the feature branch publishes nothing.
 
 - Use platform-standard formatters: `black` for Python, `prettier` for
   JS/CSS/Markdown/HTML, `go fmt` for Go. Always use default configuration with
@@ -189,14 +311,21 @@ style conventions are in separate documents:
   module under test to verify it compiles/parses. There is no excuse for
   `make test` to be a no-op.
 
-- `make test` must complete in under 20 seconds. Add a 30-second timeout in the
-  Makefile.
+- `make test` must complete in under 60 seconds. That is the hard cap, and a
+  suite that exceeds it fails. Under 20 seconds is the target. A suite between
+  20 and 60 seconds is still green, but the overage must be filed as an
+  improvement bug against that repo. Add a 90-second timeout to the test
+  invocation (`go test -timeout 90s`). The backstop deliberately sits above the
+  hard cap so that it catches a genuinely hung test rather than a merely slow
+  one.
 
-- **`make test` should use the conditional verbose rerun pattern.** Run tests
-  without `-v` (verbose) first. If tests fail, automatically rerun with `-v` to
-  show full output. This keeps CI logs and `docker build` output clean on
-  success (just package/suite summaries) while providing full diagnostic detail
-  on failure (every test case, every assertion). The general shell pattern:
+- **The test command should use the conditional verbose rerun pattern.** Run
+  tests without `-v` (verbose) first. If tests fail, automatically rerun with
+  `-v` to show full output. This keeps CI logs and `docker build` output clean
+  on success (just package/suite summaries) while providing full diagnostic
+  detail on failure (every test case, every assertion). The command lives in the
+  `test` phase of the `Dockerfile`, since `script/test` builds that phase; the
+  Makefile form below is the same pattern for any repo-local invocation:
 
     ```makefile
     test:
@@ -209,11 +338,26 @@ style conventions are in separate documents:
 
     ```makefile
     test:
-    	@go test -timeout 30s -race -cover ./... || \
+    	@go test -count=1 -timeout 90s -race -cover ./... || \
     		{ echo "--- Rerunning with -v for details ---"; \
-    		  go test -timeout 30s -race -v ./...; exit 1; }
+    		  go test -count=1 -timeout 90s -race -v ./...; exit 1; }
     ```
 
+    `-count=1` is required on both invocations: it defeats Go's test _result_
+    cache, so neither run can report a stored pass in place of running the
+    tests. It leaves the build cache alone, so it costs the runtime of the suite
+    and no recompilation.
+
+    That cache is Go's own, separate from Docker's layer cache. Go stores a
+    passing result in its cache directory (`GOCACHE`), and when the same tests
+    run again on unchanged code it prints that result, marked `(cached)`,
+    without running them. That matters on a developer's machine, where this
+    target runs and the directory lasts from one run to the next. The `test`
+    phase of the `Dockerfile` needs no `-count=1`: its base image holds no
+    result for this repo's tests and nothing before its `go test` step runs a
+    test, so there is nothing to replay. `--no-cache` (above) is what makes that
+    step run on an unchanged tree.
+
     Python example:
 
     ```makefile
@@ -239,10 +383,84 @@ style conventions are in separate documents:
   must be in `.gitignore`. No exceptions.
 
 - `.gitignore` should be comprehensive from the start: OS files (`.DS_Store`),
-  editor files (`.swp`, `*~`), language build artifacts, and `node_modules/`.
-  Fetch the standard `.gitignore` from
-  `https://git.eeqj.de/sneak/prompts/raw/branch/main/.gitignore` when setting up
-  a new repo.
+  editor files (`.swp`, `*~`), in-repo agent scratch directories (`.claude/`),
+  language build artifacts, and `node_modules/`. Fetch the standard `.gitignore`
+  from `https://git.eeqj.de/sneak/prompts/raw/branch/main/.gitignore` when
+  setting up a new repo. These patterns are written to `.gitignore`'s own
+  semantics, in which an unanchored pattern already matches at every depth; they
+  are not a `.dockerignore` and must not be transplanted into one unmodified.
+
+- **`.dockerignore` does not use `.gitignore` semantics, and copying patterns
+  across unmodified leaves secrets in the build context.** Docker matches with
+  `moby/patternmatcher`: `filepath.Match` semantics plus a `**` extension, so
+  `*` does not cross `/` and a pattern without a leading `**/` is anchored at
+  the build-context root. A `.dockerignore` listing `.env`, `*.pem` and `*.key`
+  therefore excludes only the copies at the repository root, while `config/.env`
+  and `certs/server.key` still reach the context and can land in an image layer
+  — which is more dangerous than a short file with no secret patterns at all,
+  because it reads as solved and stops anyone looking. Give every
+  depth-independent pattern the `**/` prefix and leave only genuinely
+  root-anchored entries unprefixed: `.claude`, and the repo's own host-built
+  binary, written `/myapp` and never `**/myapp`, which would also match
+  `cmd/myapp/` and delete the package directory from the context. Matching is
+  case-sensitive, and an ALL-CAPS twin per pattern still misses `Server.Key`, so
+  secret names use character ranges — `**/*.[kK][eE][yY]`, `**/*.[pP][eE][mM]`,
+  and likewise for `.envrc` and the extensionless SSH keys. Where such a pattern
+  also catches something the build needs, re-include it with a negation
+  (`!docs/example.env`); deleting the pattern reopens the exposure for every
+  other file it covers. Fetch the standard `.dockerignore` from
+  `https://git.eeqj.de/sneak/prompts/raw/branch/main/.dockerignore` and extend
+  it with the repo's own artifacts.
+
+- **In-repo agent scratch belongs in both files, written to each file's own
+  semantics.** `.claude/` holds one worktree per in-flight agent — an entire
+  additional checkout of the repo — so under `COPY . .` the build context
+  inflates by a multiple of the repo and another session's unreviewed work can
+  be copied into an image layer. In `.gitignore` the entry is `.claude/`,
+  unanchored. In `.dockerignore` it is `.claude`, anchored and with **no** `**/`
+  prefix, because the prefixed form would also delete any nested directory of
+  that name from the build. Anchoring carries a known gap that the canonical
+  `.dockerignore` states in its own comment, since consuming repos receive the
+  file and not the tracker: the directory is created in the agent's working
+  directory, so a repo running agents in subdirectories still ships
+  `services/api/.claude/` and must add its own anchored entry there.
+
+- **A plain `docker build .` of a clone stamps the version that
+  `git describe --tags --always` gives**, derived from the `.git` in the build
+  context as the canonical `Dockerfile` above shows. Without its failure check,
+  a missing `git` or an unreadable checkout would leave `-X main.Version=` empty
+  and the build would still exit 0. `script/docker` and `script/cibuild` pass
+  the version they compute on the host; it takes precedence. They do this
+  byte-identically across repos:
+
+    ```sh
+    # Own line: a failing command substitution inside an argument does not
+    # trip `set -e`, so the inline form degrades to an empty constant.
+    version="$(git describe --tags --always --dirty 2>/dev/null || true)"
+    [ -n "$version" ] || version="unknown"
+    docker build --no-cache \
+        --build-arg VERSION="$version" \
+        -t "$(script/projectname)" .
+    ```
+
+    `--always` makes an untagged repo yield an abbreviated commit hash rather
+    than failing, and the `[ -n "$version" ]` line is the single place the
+    fallback is applied — a live check that fires on a build from an export with
+    no `.git` and on a repository with no commits yet. Do not fold it into the
+    substitution as `|| echo unknown`, which makes the guard unreachable. The
+    Dockerfile's side is `ARG VERSION` in the stage that compiles, declared
+    there because `ARG` is stage-scoped; passing `VERSION` to a repo whose
+    Dockerfile declares no such `ARG` is ignored and costs nothing, which is why
+    the scripts stay byte-identical. One consequence for CI: the standard
+    checkout action clones shallow and fetches no tags, so a repo that embeds a
+    tag-derived version must set `fetch-depth: 0` on its checkout step.
+
+- **Verify `.dockerignore` by enumerating the image, not by reading the
+  patterns.** Plant files at the root _and_ at least two directories deep, build
+  a probe image that does `COPY . .`, and list what actually landed
+  (`docker run --rm --entrypoint find IMAGE /app`). The `transferring context`
+  size is not a substitute: a nested secret is a few bytes, and BuildKit
+  transfers only the delta from the previous build.
 
 - **No build artifacts in version control.** Code-derived data (compiled
   bundles, minified output, generated assets) must never be committed to the
@@ -258,9 +476,56 @@ style conventions are in separate documents:
 - Make all changes on a feature branch. You can do whatever you want on a
   feature branch.
 
-- `.golangci.yml` is standardized and must _NEVER_ be modified by an agent, only
-  manually by the user. Fetch from
-  `https://git.eeqj.de/sneak/prompts/raw/branch/main/.golangci.yml`.
+- `.golangci.yml` is standardized. The vendored copy in a consuming repo must
+  _NEVER_ be modified by an agent: fetch it from
+  `https://git.eeqj.de/sneak/prompts/raw/branch/main/.golangci.yml` and keep it
+  byte-identical, so that no repo can quietly loosen its own linting. Linter
+  configuration changes are made to the canonical copy in the `prompts` repo and
+  reach consuming repos by re-vendoring; an agent may open a PR against
+  canonical, which only the user merges. One list is exempt from byte-identity,
+  because it cannot be written once for every repo: the `deny` list of the
+  `test-support` depguard rule, where a repo names its own test-support packages
+  by full import path. A repo adds entries there and changes nothing else, and a
+  re-vendor carries its entries forward. The canonical golangci-lint version is
+  v2.14.0 (released 2026-09-24), pinned as the digest of the lint phase's base
+  image
+  (`golangci/golangci-lint@sha256:ad862ba6b3798cbe0fd9fd7408d498fd74fbd2623a92406b2fd3898faf0bf98f`,
+  which reports `2.14.0 built with go1.27.0 from 114493f9`). A module's `go`
+  directive must not name a newer Go minor version than the one golangci-lint
+  was built with, or golangci-lint refuses to lint it: this release lints
+  `go 1.27.1` but not `go 1.28`. That digest is the only pin, since no repo
+  installs golangci-lint on the host. A repo sets the lint phase digest to the
+  one named here and re-vendors `.golangci.yml` in the same commit, whichever of
+  the two prompted the change: the canonical copy can name linters that an older
+  golangci-lint rejects, and a newer golangci-lint can add linters that
+  `default: all` switches on until the canonical copy disables them.
+
+- **`script/bootstrap` installs a pinned tool by comparing versions, never by
+  testing presence.** An `if ! command -v <tool>; then install; fi` guard tests
+  `PATH` only, so on an already-provisioned machine the pin is inert and a
+  version bump is a silent no-op — while the Dockerfile, installing into a clean
+  image, gets the pinned version, so a local `make check` and `make docker` can
+  disagree about what the tool even is. The canonical form:
+    - compares the installed version against the pin over the **whole** version
+      token; a parser that stops at the first `-` reports `2.12.2` for a host
+      running `2.12.2-rc1` and skips the install;
+    - treats absent, non-zero, empty or unrecognised `--version` output as a
+      mismatch, so the failure direction is a redundant install and never a
+      skipped one;
+    - after installing, re-resolves the binary the way callers do — `hash -r`,
+      then through `PATH`, not through the directory the installer wrote to —
+      and fails naming the resolved path, since an install that a shadowing
+      binary hides succeeds while changing nothing any caller sees;
+    - is actually called, and prints the version on both success paths: a
+      function defined and never invoked has the same exit status and the same
+      empty output as one that worked.
+
+    Keep it POSIX sh: no arrays, no `[[`, no `grep -P`.
+
+    A Go tool a repo needs on the host is installed with `go install` pinned to
+    a commit hash (`go install <package>@<commit hash>`). It is never tracked as
+    a `go.mod` tool dependency or through a `tools.go` file, either of which
+    pulls the tool's own dependencies into the repo's `go.mod` and `go.sum`.
 
 - When pinning images or packages by hash, add a comment above the reference
   with the version and date (YYYY-MM-DD).
@@ -374,12 +639,14 @@ style conventions are in separate documents:
   settings.
 
 - Avoid putting files in the repo root unless necessary. Root should contain
-  only project-level config files (`README.md`, `Makefile`, `Dockerfile`,
-  `LICENSE`, `.gitignore`, `.editorconfig`, `REPO_POLICIES.md`, and
-  language-specific config). Everything else goes in a subdirectory. Canonical
-  subdirectory names:
+  only project-level config files (`README.md`, `AGENTS.md`, `Makefile`,
+  `Dockerfile`, `LICENSE`, `.gitignore`, `.editorconfig`, `REPO_POLICIES.md`,
+  and language-specific config). Everything else goes in a subdirectory.
+  Canonical subdirectory names:
     - `bin/` — executable scripts and tools
-    - `cmd/` — Go command entrypoints
+    - `cmd/` — Go command entrypoints; thin only: one `main.go` per binary whose
+      body is a single call into `internal/` or `pkg/`, no project logic in
+      `cmd/`
     - `configs/` — configuration templates and examples
     - `deploy/` — deployment manifests (k8s, compose, terraform)
     - `docs/` — documentation and markdown (README.md stays in root)
@@ -406,3 +673,7 @@ style conventions are in separate documents:
     - Go: `go.mod`, `go.sum`, `.golangci.yml`
     - JS: `package.json`, `yarn.lock`, `.prettierrc`, `.prettierignore`
     - Python: `pyproject.toml`
+
+- Guidance for coding agents lives in one `AGENTS.md` at the repository root. It
+  is never committed under a file or directory named after one agent tool, such
+  as `CLAUDE.md` or `.claude/`, and never split into separate memory files.

TODO.md

@@ -1,696 +0,0 @@
-# Workflow
-
-* branch (from `main`)
-* do the work in Next Step
-* move Next Step to the top of Completed Steps
-* move the top item of Future Steps into Next Step
-* commit (`TODO.md` changes in the same commit as the work)
-* merge to `main` if the branch is not protected, otherwise open a PR
-* push
-
-# Status
-
-pre-1.0
-
-# Next Step
-
-Define the remaining scope for the first tagged release under the 1.0.0
-milestone, then cut that tag. The mechanism to cut it now exists and is
-exercised; what is left is the scope decision, which is the owner's.
-This step deliberately names one version number: it previously said
-"cut v0.1.0" while the `Makefile` baked in `1.0.0-rc.1` and the issue
-milestone said 1.0.0, and three different answers to "what is the next
-release" is exactly the contradiction
-[issue #65](https://git.eeqj.de/sneak/vaultik/issues/65) was filed over.
-
-# Completed Steps
-
-- 2026-09-22: Routed the last direct-to-stdout command output through
-  `internal/ui`
-  ([issue #149](https://git.eeqj.de/sneak/vaultik/issues/149)). The
-  `version`, `info`, `remote info`, `config`, and `database delete`
-  commands wrote plain text straight to stdout, so they were unstyled and
-  ignored `--quiet`. Output now falls in two buckets, both governed by
-  `internal/ui`: status lines and confirmations go through its message
-  methods (styled, and `--quiet` silences them), while the data a command
-  exists to produce — the reports, the `snapshot list` table, `config get`
-  values, and the `--json` documents — is written plain. `--quiet`
-  silences the human reports and tables but never the `config get` value
-  or the `--json` documents, which a script depends on, and the
-  `database delete` confirmation prompt is always shown. The README
-  output-style section now states this rule.
-
-- 2026-09-22: Validated blob hashes, offsets and lengths read back from
-  the destination before using them
-  ([issue #155](https://git.eeqj.de/sneak/vaultik/issues/155)). A blob
-  hash taken from the downloaded database or the store listing was
-  trusted unchecked, so a hostile remote could drive a decrypted blob to
-  be written outside the cache directory (a hash like `aa/../../etc`) or
-  crash a command with a short or negative value. `blobDiskCache.path`
-  now refuses any key containing a path separator, and `ReadAt` rejects a
-  negative offset or length and bounds with `length > size-offset` so a
-  sum cannot overflow past the check. A new `isBlobHash` helper (a plain
-  function, not a method — the packer stores `temp-placeholder-{uuid}` as
-  a hash) gates `FetchBlob`, shallow and deep verify; the `blobs/` and
-  `metadata/` listings skip a non-conforming name with a warning; and
-  short-hash prefixes in log and error text go through a `shortHash`
-  helper that cannot panic. `verify`'s chunk reader also rejects a
-  negative `blob_chunks` length and streams the chunk instead of
-  allocating a database-supplied size. `restore.go` and
-  `internal/database` were left untouched to avoid colliding with the
-  in-flight [issue #156](https://git.eeqj.de/sneak/vaultik/issues/156)
-  work; the cache-path and `FetchBlob` guards already stop the unsafe
-  write and fetch, so restore's own `buildBlobIndexes` early check is
-  deferred as fail-fast defense in depth.
-
-- 2026-09-21: Stopped an interrupted blob upload from making a later
-  backup deduplicate against data that was never stored
-  ([issue #148](https://git.eeqj.de/sneak/vaultik/issues/148)). The
-  packer commits a blob's `chunks`, `blob_chunks`, and `blobs` rows
-  before the upload is attempted, so a failed upload left chunk rows
-  behind and the next run skipped re-uploading them, producing a
-  snapshot that reported success but could not be restored. A run now
-  deduplicates only against chunks held by a blob whose `uploaded_ts` is
-  set, and at startup drops any un-uploaded blob rows (and the chunks
-  they orphan) so the affected data is re-chunked and re-uploaded. Blobs
-  recorded with no remote backend are marked uploaded so this invariant
-  holds uniformly.
-
-- 2026-09-22: Made restore refuse any snapshot path that would write
-  outside the target directory
-  ([issue #154](https://git.eeqj.de/sneak/vaultik/issues/154)).
-  `restoreFile` and `verifyRestoredFiles` joined the stored path onto the
-  target with no containment check, so a `..` segment or an absolute path
-  escaped the target and a restored symlink could redirect a later child
-  write anywhere on disk. Every stored path is now rejected unless
-  `filepath.IsLocal` accepts it with the leading separator removed, and
-  each existing ancestor directory below the target is `Lstat`ed to refuse
-  descending through a symlink; honest symlinks pointing outside the tree
-  are still written verbatim. age decryption proves a snapshot is
-  readable, not honest, and restore usually runs as root.
-
-- 2026-09-21: Stopped `--json` from silencing stderr diagnostics
-  ([issue #112](https://git.eeqj.de/sneak/vaultik/issues/112)). `--json`
-  used to be folded into `Quiet`, which pinned the log level to `WARN`,
-  so `prune --json` gave a machine consumer no record of the local index
-  rows it deleted even under `--verbose`. `--json` now quiets only the
-  stdout UI (the JSON document must stay clean, per
-  [issue #108](https://git.eeqj.de/sneak/vaultik/issues/108)); the stderr
-  log level follows `--verbose`/`--debug` again. The coupling was
-  removed the same way for `snapshot verify`, `snapshot remove`, and
-  `remote info`, which carried it for the same outdated reason.
-
-- 2026-09-21: Stopped `prune` from reporting a failed row count as 0
-  ([issue #96](https://git.eeqj.de/sneak/vaultik/issues/96)). The seven
-  `getTableCount` reads in `PruneDatabase` discarded their error, so a
-  query that could not run became a plausible `0` and the before/after
-  delta computed from it looked like real work. Each read now logs at
-  warn on failure and renders as `unknown`, never `0`, so an empty table
-  is distinguishable from one that could not be queried. The counts have
-  no `--json` representation — under `--json` the summary is suppressed
-  entirely — so nothing there can show a false `0`.
-
-- 2026-09-21: Made the s3 storage backend report a missing object as
-  `storage.ErrNotFound`, like the `file` and `rclone` backends and as the
-  `Storer` interface documents. `S3Storer.Get` and `Stat` returned the raw
-  AWS SDK error, so `errors.Is(err, storage.ErrNotFound)` was false on s3
-  and callers branched differently per backend. Added a small `s3.IsNotFound`
-  helper (reused by `HeadObject`) and a test that a missing key maps to
-  `ErrNotFound`
-  ([issue #129](https://git.eeqj.de/sneak/vaultik/issues/129)).
-- 2026-09-21: Fixed `verify --deep` reporting healthy snapshots as
-  corrupt. Its final blob-integrity check hashed the encrypted
-  downloaded bytes with a single SHA256 and compared that to the blob
-  ID, which is the double SHA256 of the plaintext, so the two could
-  never match. It now hashes the decompressed plaintext and compares the
-  double SHA256. Added a test that backs up a real snapshot, deep-verifies
-  it, then flips a byte in one stored blob and confirms deep verification
-  then fails
-  ([issue #131](https://git.eeqj.de/sneak/vaultik/issues/131)).
-
-- 2026-09-21: Made `snapshot create` VACUUM the per-snapshot metadata
-  database through the `modernc.org/sqlite` driver instead of shelling
-  out to the external `sqlite` command-line binary (issue #120). A
-  backup no longer needs that binary on `PATH`, so `make check` passes
-  on a stock `go install` host; `script/bootstrap` and the `Dockerfile`
-  (both the test-build and the shipped runtime stage) no longer install
-  it, and a new test asserts the uploaded database keeps no pages from
-  deleted rows. Dropped the now-false note on the 2026-08-07 entry below
-  that said bootstrap installs it.
-- 2026-09-21: Made `.gitea/workflows/check.yml` run on pushes to `main`
-  and `next` and on pull requests against either, so unit PRs (whose
-  base is `next`) and `next` itself get a CI run instead of relying on a
-  local `make check`
-  ([issue #122](https://git.eeqj.de/sneak/vaultik/issues/122)).
-
-- 2026-09-21: Hash-verified the Go toolchain in the release workflow
-  ([issue #105](https://git.eeqj.de/sneak/vaultik/issues/105)). New
-  `script/install-go` downloads the exact `go.dev` archive for `go.mod`'s
-  `go` directive and refuses it unless its sha256 matches a value
-  committed in the script; `.gitea/workflows/release.yml` calls it
-  instead of `actions/setup-go`, which verified the downloaded toolchain
-  against nothing in the repo. `GOTOOLCHAIN: local` on the release step
-  keeps that exact compiler from auto-switching. Bumping Go now touches
-  `go.mod`, the checksum, and the `Dockerfile` `golang` digest together.
-
-- 2026-09-21: Collapsed the two duration parsers into one and fixed the
-  `--older-than` months example
-  ([issue #123](https://git.eeqj.de/sneak/vaultik/issues/123)). Two
-  functions named `parseDuration` existed with different grammars;
-  `snapshot purge --older-than` and `--keep-newer-than` both already went
-  through the one in `internal/vaultik`, while the richer copy in
-  `internal/cli/duration.go` was reachable only from its own test. Kept
-  the live-path parser and deleted the unused one, so no flag's accepted
-  grammar changes. The trap the issue was filed over: `README.md`
-  documented `6m` as the months example for `--older-than`, but `m` is
-  minutes, so the documented command deleted every snapshot older than
-  six minutes on a destructive flag. Corrected the doc to `6mo` and put
-  both flags' help text on one example list that states `m` is minutes
-  and `mo` is months. The surviving parser now rejects negatives, which
-  it previously accepted (`-5h`) or silently made positive (`-5d`).
-  Table-driven tests cover every unit, `6m` as six minutes, `6mo` as 180
-  days, and rejection of a bare number, an unknown unit, and a negative.
-
-- 2026-08-10: Moved every lint run into its own container, as a build
-  step ([issue #113](https://git.eeqj.de/sneak/vaultik/issues/113)).
-  New root `Dockerfile.lint`, built by `script/lint`, runs
-  `golangci-lint run --config .golangci.yml ./...` as a `RUN`
-  instruction in the digest-pinned `golangci/golangci-lint` image: a
-  successful build of that file *is* a clean lint, and it works even
-  where the daemon is remote and bind mounts are impossible. That
-  `FROM` line is now the only pin of the linter version in the repo.
-
-  This supersedes the per-worktree cache isolation landed for
-  [issue #99](https://git.eeqj.de/sneak/vaultik/issues/99). Isolation
-  fixed cross-worktree contamination but not lock contention — two
-  concurrent runs with entirely separate cache directories still
-  collided. A container per run has its own cache and its own lock, so
-  the whole class is gone, and with it the per-worktree cache
-  machinery, the lock-retry loop, and `script/lint-audit`, which
-  existed to catch replayed findings from a cache that no longer
-  exists. The host lint path went too: no escape hatch, no
-  `VAULTIK_LINT_IN_CONTAINER`, no version detection. Nothing lints on
-  the host at any version.
-
-  A cached build lints nothing, so the same `CHECK_EPOCH` mechanism the
-  product `Dockerfile` already used is what makes the green mean
-  something: `ARG CHECK_EPOCH` with no default below the module layers,
-  a `RUN [ -n "$CHECK_EPOCH" ] || exit 1` guard, the value expanded
-  into each check command, and a fresh `$(date +%s%N)$$` per invocation
-  computed as a bare assignment. `cmd/vaultik/lintdocker_test.go`
-  parses both Dockerfiles and both scripts and fails if any part of
-  that is dropped, because every way of losing it is silent. No test
-  asserts that no script runs the host linter: `script/lint` is the one
-  lint entry point and runs `golangci-lint` only inside the container,
-  and keeping it that way is a review matter, not something a test
-  proves.
-
-  The product `Dockerfile` lost its lint stage rather than gaining a
-  second linter pin: `make lint` is now `docker build`, so the stage
-  would have been docker-in-docker with no daemon, and calling
-  `golangci-lint` directly there would have restored the two-pins drift
-  of [issue #78](https://git.eeqj.de/sneak/vaultik/issues/78).
-  `make fmt-check` moved beside `make test` in the builder stage, and
-  `script/cibuild` now builds `Dockerfile.lint` and then `Dockerfile`,
-  each with its own fresh epoch. Consequence, stated rather than left
-  to be found: `script/docker` builds the product image only and no
-  longer lints; the gates are `script/check` and `script/cibuild`.
-
-  `golangci-lint config verify` runs as its own epoch-keyed layer,
-  above the lint. `golangci-lint run` rejects a config it cannot parse
-  but silently ignores an unknown top-level *key*: renaming `linters:`
-  to `linterz:` discarded `default: all` and every threshold and still
-  exited 0 on a tree the real config fails. `config verify` catches
-  that, and it does so with the network off at this pin — checked under
-  `docker run --network none`, not assumed. An earlier revision omitted
-  it on the claim that it fetches its schema over live HTTPS; that
-  claim was false at v2.12.2.
-
-  `script/lint-fix` is kept, reimplemented as a
-  bind-mounted `docker run` against the image parsed out of
-  `Dockerfile.lint` — it cannot be a build step, because fixes have to
-  land in the worktree — and marked in its header as a developer
-  convenience that no gate reads.
-
-- 2026-08-09: Finished the `--json` stdout contract and gave `make build`
-  a rule ([issue #108](https://git.eeqj.de/sneak/vaultik/issues/108),
-  [issue #110](https://git.eeqj.de/sneak/vaultik/issues/110)). Two
-  unrelated defects of the same shape — a command reporting something it
-  did not do — landed together because both are small.
-
-  `CleanupLocalSnapshots` wrote three prose lines to stdout with no
-  `--json` awareness, covering every branch of the function, so no input
-  avoided them and `vaultik prune --json | jq` failed even after
-  [issue #106](https://git.eeqj.de/sneak/vaultik/issues/106) removed the
-  banner. `-q` never helped either: `printlnStdout` and `stdoutf` write
-  straight to `Vaultik.Stdout` and never consult `Vaultik.UI`, which is
-  what `SetQuiet` affects. The issue offered three fixes and asked for a
-  decision. Taken: thread `*PruneOptions` into the function and gate each
-  write on `!opts.JSON`, matching `PruneBlobs` (its sibling phase, which
-  already takes the same struct), `RemoveSnapshot` and `remote info`, so
-  the package has one pattern rather than two. Rejected: moving the lines
-  to `log.Info`, because the logger's default level is `slog.LevelWarn`,
-  so that would not relocate them to stderr — it would delete them from a
-  plain `vaultik prune`, and the removal of rows from the local index is
-  not something to narrate only under `--verbose`. Also rejected: putting
-  the stale-record count into `PruneBlobsResult`, whose every field is
-  blob-scoped and which is produced by the later phase; a prune document
-  covering both phases is a reasonable thing to want, but it is a schema
-  design question and not a stream-hygiene fix. The narration is
-  duplicated as `log.Info` records, which `PruneBlobs` already does
-  alongside its own prints, so the events survive on stderr for anyone
-  running `--verbose`.
-
-  `make build` printed "Nothing to be done for 'build'" and exited 0
-  without producing a binary: `build` was listed in `.PHONY` with no
-  `build:` rule anywhere, and declaring a name phony is exactly what
-  converts make's "No rule to make target" error into a silent success.
-  Fixed with `build: vaultik`, keeping `vaultik:` as the file rule. The
-  audit the issue asked for covers all 19 `.PHONY` names; `build` was the
-  only one without a rule, and `vaultik` is correctly absent from
-  `.PHONY`, being a real file target.
-
-  Tests, each verified to fail with the fix reverted rather than assumed
-  to: `CleanupLocalSnapshots` leaves stdout untouched under `--json` in
-  all three branches (stale records, none, empty index) and still emits
-  every line without it, so the guard cannot be satisfied by deleting the
-  output; `prune --json` run end to end through `Entry`, cobra and fx
-  over the process's real stdout descriptor against a `file://` store,
-  asserting exactly one JSON document, in both the stale and non-stale
-  branches; and a parse of the `Makefile` asserting every `.PHONY` name
-  has a rule and that `build` reaches the rule that produces the binary,
-  which keeps the audit true for names added later. That last one is a
-  parse rather than an invocation of `make`, since `make test` is what
-  runs it and shelling back into `make build` would nest a build inside
-  the test run. The property a parse cannot establish — that the recipe
-  still fails when the build fails — was verified by hand against a
-  deliberately broken tree: `make build` exits 2 and produces nothing.
-  `cmd/vaultik` gains its first test file, so `make test` now reports 16
-  packages `ok` where it reported 15.
-
-- 2026-08-09: Stopped the startup banner from contaminating `--json`
-  documents ([issue #106](https://git.eeqj.de/sneak/vaultik/issues/106)).
-  `Entry` writes the banner to stdout before cobra parses anything, and
-  the flag scan that suppresses it knew `--quiet`, `-q` and `--cron` but
-  not `--json`, so every `--json` document arrived behind two lines of
-  prose and a blank line, and `vaultik snapshot list --json | jq` failed.
-  With the logger already on stderr from
-  [issue #82](https://git.eeqj.de/sneak/vaultik/issues/82), this was the
-  last writer that could put something on stdout that the caller did not
-  ask for. The design question the issue raised — extend the raw-argv
-  scan, or move the banner after parsing — is answered in favour of the
-  scan: the banner is printed first deliberately, so that it still
-  appears when cobra rejects the arguments and on `--help`, and after
-  parsing there is no single place that covers those paths. The stated
-  cost of the scan, that `--json` is a subcommand flag matched anywhere
-  in the vector, is a cost `--cron` already carries — it exists only on
-  `snapshot create` — so this adds an instance of an accepted
-  imprecision rather than a new kind, and the two error directions are
-  not symmetric: a false positive loses a decorative banner, a false
-  negative corrupts a document. Regression tests at the CLI layer, where
-  `internal/vaultik`'s existing guard cannot reach: one runs `Entry`
-  itself over the process's real stdout descriptor, through cobra and fx
-  to the document, made hermetic by `file://` storage; a second covers
-  the argument vectors of all five `--json` commands; a third asserts the
-  banner is still printed without a suppressing flag, so the first
-  cannot be satisfied by deleting the banner. Also corrected `AGENTS.md`
-  policy 9, which still keyed the structured-log format on stdout's
-  TTY-ness after #82 moved that decision to stderr — a rules file that
-  misdescribes the code misleads exactly the readers who trust it most.
-  Two smaller findings from the same review: `bytesAttrKey`'s
-  human-readable byte formatting silently stopped applying under an open
-  group, because the key reaching the comparison is group-qualified
-  (`transfer.bytes`), now matched on its final segment and tested both
-  ways; and `listEnv.stderr` in `snapshot_list_test.go`, assigned but
-  never read since those tests began capturing the process's stderr, is
-  removed. `Vaultik.Stderr` is kept — nothing writes to it today, which
-  its comment now says outright.
-
-- 2026-08-09: Moved the logger to stderr and fixed `TTYHandler`'s
-  discarded attributes
-  ([issue #82](https://git.eeqj.de/sneak/vaultik/issues/82),
-  [issue #97](https://git.eeqj.de/sneak/vaultik/issues/97)). Two defects
-  in `internal/log`, fixed together because both live in the handler
-  construction path. The first: both handlers were built over
-  `os.Stdout`, and `WARN`/`ERROR` are never suppressed, so a config file
-  with group- or world-readable permissions was enough to put a log
-  record inside a `--json` document and break `jq`. Diagnostics now go
-  to stderr, and the TTY/JSON format choice follows stderr rather than
-  stdout — testing the wrong stream would colorize records on a
-  redirected stderr whenever stdout happened to be a terminal. This is
-  user-visible: `--verbose` and `--debug` output moves to stderr too,
-  which is documented in `README.md` under "stdout and stderr". It also
-  let the local workaround in `internal/vaultik/snapshot_list.go` go:
-  `warnWhileListing` had been hand-rolling structured-log formatting to
-  reach a non-stdout writer, and the `jsonOutput` parameter threaded
-  through the remote-listing helpers existed only to choose between the
-  two writers. The collect-then-emit machinery around `listingWarning`
-  stays, but on its remaining merit — warnings emitted in key order
-  after `group.Wait()` are deterministic run to run, where emitting from
-  the fetch workers would order them by network timing. The second
-  defect: `TTYHandler.WithAttrs` and `WithGroup` discarded their
-  arguments and returned the receiver while their doc comments claimed
-  otherwise, so `log.With` attributes vanished on a terminal and
-  appeared correctly in CI — failing precisely when someone is debugging
-  interactively. Both now return a new handler (the receiver is never
-  written to, since `slog` permits concurrent derivation), attributes
-  persist across records, and grouping is implemented as dotted key
-  prefixes, which is the only honest rendering for a format with nowhere
-  to nest. New tests cover both, including one that feeds the same
-  derivation chain to the TTY and JSON handlers and compares the
-  attribute sets, so the two paths cannot drift apart again. Found and
-  filed while verifying: the startup banner is written to stdout and
-  `--json` does not suppress it
-  ([issue #106](https://git.eeqj.de/sneak/vaultik/issues/106)), which is
-  a separate writer on a separate path and the remaining source of
-  stdout contamination.
-
-- 2026-08-09: Made the tagged-release path actually work on Gitea
-  ([issue #65](https://git.eeqj.de/sneak/vaultik/issues/65)). Three
-  independent blockers, one of which was the whole
-  release: `.goreleaser.yaml` had no `gitea_urls:` block, so goreleaser
-  defaulted to the GitHub API and a `goreleaser release` from this repo
-  would have failed or published where nobody is looking. It now points
-  at `https://git.eeqj.de/api/v1`. The version is the second: it was a
-  hardcoded `VERSION := 1.0.0-rc.1` in the `Makefile`, so every local
-  build claimed to be a release candidate that had never been tagged and
-  did not exist, while `git tag -l` was empty and `internal/globals`
-  defaulted to `dev`. Version now comes from git via the new
-  `script/version` — the exact tag with a leading `v` stripped (so a
-  `make` build and a goreleaser build of one commit report the same
-  string, and it matches the archive names), otherwise `dev-<12-char
-  sha>`, with `-dirty` appended in either case when tracked files are
-  modified. Untracked files are deliberately not counted, matching
-  `git describe --dirty`. The same honesty was owed by the snapshot
-  path: `snapshot.version_template` was `{{ incpatch .Version }}-next`,
-  which manufactures a release number from the last tag and, with no
-  tags at all, from goreleaser's fabricated `v0.0.0`; it now emits the
-  same `dev-<sha>`. The one non-obvious consequence is that
-  `internal/cli/version.go` gated its "this is a development build"
-  notice on the version being exactly `dev`, so the moment untagged
-  builds began carrying a commit sha that notice would have gone silent
-  and an unreleased binary would have read as a release — the gate is
-  now `globals.IsDevVersion`, which is a predicate over a string rather
-  than a comparison against a global precisely so it can be tested, and
-  it is tested at the boundary (`1.0.0-dev` is a release, `dev-<sha>`
-  is not). Release automation is the third blocker: a tag-triggered
-  `.gitea/workflows/release.yml` runs the build in CI rather than from
-  a laptop, with `fetch-depth: 0` because a shallow checkout has no
-  tags and would silently mislabel the release, and with the
-  `RELEASE_TOKEN` repository secret passed as `GITEA_TOKEN` (documented
-  in `README.md`; the runner's automatic token is not used because it
-  is not guaranteed to carry release write scope). `script/release`
-  unsets any `GITHUB_TOKEN`/`GITLAB_TOKEN` it finds, since goreleaser
-  chooses its forge from whichever token variable is set and refuses to
-  run when it sees more than one — a runner-provided token must not get
-  to decide where these artifacts are published. `make release` and
-  `make release-snapshot`, the last two Makefile targets that were not
-  shims, now call `script/release` and `script/release-snapshot`, which
-  resolve goreleaser exactly the way `script/lint` resolves the linter:
-  a `PATH` binary is used only at the pinned version, never as a silent
-  fallback. `script/bootstrap` installs it, from a sha256-verified
-  GitHub release archive per `REPO_POLICIES.md`, via a separate
-  `script/install-goreleaser` — separate because `script/bootstrap`
-  hard-fails without a usable Docker daemon by design, and the release
-  runner needs goreleaser without needing Docker. Verified by running
-  the thing rather than reading it: `make release-snapshot` produced
-  four archives and `checksums.txt`, and the linux/amd64 binary from
-  `dist/` reports `dev-<sha>` with the development-build notice. Tag
-  handling was exercised in a throwaway repository rather than by
-  tagging this one; no tag was created here, since that is the owner's
-  call. Signing, SBOM, reproducible builds, completions and a man page
-  are out of scope by the issue.
-
-- 2026-08-09: Isolated the lint cache per worktree and context-gated the
-  native lint path (issues #99, #80). One defect seen twice:
-  `script/lint` decided whether it could skip the pinned image by asking
-  what version was on `PATH` rather than where it was running, and cache
-  isolation is part of that same question. The cache was one directory
-  per repo, shared by every worktree on the host, so two checkouts with
-  identical Go file contents collided and golangci-lint replayed the
-  stored analysis — paths and all. The loud direction of that failure
-  (a clean tree failed by a dirty sibling) is the harmless one; the
-  silent direction, a dirty tree **passed** by a clean sibling, is a
-  sixth way for a gate here to report a green it did not earn. The cache
-  is now keyed on a digest of the worktree path, and every run is
-  audited by the new `script/lint-audit`, which rejects output citing any
-  file that is not in the tree being linted — a backstop that runs on
-  clean output too, because that is the case nobody investigates. Caches
-  record the worktree they belong to and are collected when it
-  disappears, so throwaway worktrees do not accumulate them; the whole
-  tree lives under `XDG_CACHE_HOME` and is disposable. The
-  `parallel golangci-lint is running` refusal is now a bounded retry
-  rather than a verdict: it is not a lint result, and exiting non-zero
-  on it is indistinguishable to a caller from real findings (#88 showed
-  a private cache does not remove that contention). The native path now
-  requires `VAULTIK_LINT_IN_CONTAINER=1`, set only by the `Dockerfile`
-  lint stage, in addition to matching the pin, so a developer's locally
-  installed 2.12.2 no longer bypasses the digest pin; `/.dockerenv` was
-  rejected as the signal because `dockerd` creates it for `docker run`
-  and it is not reliably present during a BuildKit `docker build`, which
-  is the case the exception exists for. Version detection uses
-  `golangci-lint version --short` with the old banner scrape kept only
-  as a fallback. `script/bootstrap` no longer prints `bootstrap
-  complete` on a machine that cannot run the gate: a missing docker, or
-  one whose daemon is unreachable, is a hard failure naming exactly what
-  breaks. Verification was by reproduction rather than inspection — two
-  concurrent lints from two worktrees of differing cleanliness, a real
-  run made to report an outside path, a matching linter shimmed onto
-  `PATH`, and a `PATH` with docker removed — and is recorded on the pull
-  request.
-- 2026-08-09: Closed the fifth false-green mechanism (issues #93, #69).
-  `script/test` omitted `-count=1`, so Go's test result cache could
-  satisfy the gate outright: a second back-to-back `make test` printed
-  the full set of 14 `ok` lines, every one marked `(cached)`, having
-  executed no test at all. Since `ok <pkg> (cached)` is an `ok` line,
-  the "14 `ok` lines means the suite ran" signal this repo leans on was
-  forgeable, one level below the Docker layer cache that #85 addressed.
-  Fixed with `-count=1` unconditionally rather than only in the
-  container, because the pre-commit hook runs the same script and a
-  gate honest only in CI is dishonest where people rely on it most;
-  `test-coverage` got the same flag, and `script/check` inherits it by
-  calling `script/test`. In the same area, `make test-integration` was
-  deleted rather than made real: no file in the repo carried a build
-  tag, so `-tags=integration` selected nothing and the target was an
-  exact duplicate of `make test`. Tagging a subset was rejected because
-  the entire suite runs in well under a minute, and a scheme whose
-  failure mode is "some tests silently stopped running" is a poor trade
-  for those seconds in a repo with this particular history. The
-  `-timeout` was raised from 30s after measuring rather than after
-  assuming: the standing claim that cold-cache compilation is charged
-  against `-timeout` is **false**, disproved by a containerised run
-  that spent 46s compiling and still reported per-package durations
-  within noise of a warm host run. `-timeout` reaches the test binary
-  as `-test.timeout` and its clock starts inside `testing.M.Run`, after
-  the build. The real exposure was margin, not compilation. The 120s
-  landed on is a **deliberate, documented divergence** from
-  `REPO_POLICIES.md:192`, which mandates 30s, and from that file's
-  canonical recipe at `:212-214`; the divergence is recorded in
-  `script/test`'s comment because `REPO_POLICIES.md` is org-canonical
-  and not editable here, and issue #101 proposes amending the policy
-  text upstream. Numbers and the full verification are recorded once,
-  on the pull request, and are deliberately not restated here.
-
-- 2026-08-09: Triaged all fifteen stale remote branches (issue #71) and
-  deleted fourteen of them; the full per-branch disposition with
-  evidence is recorded on that issue. Method mattered more than the
-  outcome here: a three-dot `git diff main...branch` diffs from the
-  merge base, so it replays everything that landed on `main` after the
-  branch diverged and makes any old branch look like it holds unlanded
-  work. That artifact is what made `golangci-v2.12.2` appear to carry
-  126 files of unpushed changes when its tree was byte-identical to
-  `main`'s. Every containment claim here therefore rests on two-dot tip
-  diffs, tree-hash equality, `git cherry`, and `git branch -r --merged`.
-  Nine branches were plain ancestors of `main` with zero `git cherry`
-  `+` commits. `golangci-v2.12.2` had landed squashed as `cc58583`,
-  whose tree hash equals the branch tip's exactly; note the hash
-  recorded in the issue had gone stale because `main` advanced, so the
-  check had to be redone rather than repeated.
-  `fix/sync-snapshot-cleanup` was redundant, its one line already on
-  `main` in `syncWithRemote`. `feature/restore-progress-bar` was
-  superseded by `printRestoreProgress` and the disk-backed blob cache,
-  and had become actively regressive — it would have deleted
-  `internal/blobgen/compress_test.go`, the #28 regression test that
-  landed separately. The two branches this issue was filed for both
-  turned out to be closed questions that `main` had already moved past
-  by a recorded decision, so neither was landed and no regression test
-  was owed: `ctime` no longer exists anywhere in the codebase after
-  `1c72a37` removed the column, the `File.CTime` field and every use
-  (#54/#55), and change detection compares size, mtime, mode, uid and
-  gid only, exactly as `ARCHITECTURE.md` documents — so the
-  silently-skipped-file data-loss risk that made this a 1.0 item does
-  not exist. The SQL allow-list branch would have reverted `bfd7334`,
-  which replaced that very allow-list with regex sanitisation on review
-  feedback, and would have broken `getTableCount("snapshots")` because
-  its allow-list omits that table. `feature/daemon-mode` is untouched
-  and deferred to #94 pending an owner decision, so it is the one
-  branch besides `main` still on the remote. The stale `TODO.md` entry
-  named in the issue needed no fix: `e496aa3` had already removed it.
-  No product code changed.
-- 2026-08-09: Adopted the remaining upstream `CHECK_EPOCH` hardening
-  (issue #91), closing the gap #85 knowingly left open. Four changes,
-  all four decided as adopt upstream in `sneak/prompts` #26. (1) Each
-  check stage now asserts `[ -n "$CHECK_EPOCH" ] || exit 1` before
-  running anything, so a build that supplies no `--build-arg` fails
-  instead of lying. This is the item that mattered: an unset `ARG` is
-  an empty string and an empty string is a stable cache key, so the
-  second and every later bare `docker build .` on an unchanged tree
-  replayed all three check layers and still exited 0 — and `docker
-  build .` is the command `REPO_POLICIES.md` names verbatim as a thing
-  that must be green, so the documented command was precisely the one
-  that lied. Failed steps are never cached, which is what makes the
-  guard fire on every invocation rather than once. (2) The epoch is now
-  expanded into each check command rather than left as a bare
-  declaration, so the cache miss no longer depends on BuildKit's
-  unreferenced-`ARG` handling staying as it is, and the value appears
-  in the build log. (3) `script/cibuild` uses
-  `epoch="$(date +%s%N)$$"`, unique per invocation rather than per
-  second; `%N` alone is insufficient because busybox drops it silently
-  and exits 0, and `$$` is what makes the guarantee hold regardless.
-  The bare-assignment form is kept deliberately — inlined in an
-  argument, a failing substitution does not abort under `set -eu` and
-  would yield an empty constant epoch, restoring the exact false green
-  being fixed. (4) `script/docker` passes the same fresh arg, so the
-  two entrypoints cannot disagree about whether the tree is green;
-  local builds are almost always warm, which made it the likelier
-  fooling in practice. The `ARG` placement from #85 is unchanged, below
-  `apk add`, `COPY go.mod go.sum` and `go mod download`, so dependency
-  layers still cache and the build is not cold. Verified by negative
-  control rather than inspection — a bare `docker build .` run twice
-  back to back, plus back-to-back pairs of both scripts and a host-side
-  `make check`; the measurements are recorded once, in the PR
-  verification comment, rather than restated here. `.golangci.yml`, the
-  lint-stage `FROM` line and its digest, `script/lint`,
-  `REPO_POLICIES.md` and `.gitea/workflows/check.yml` are all
-  untouched.
-- 2026-08-09: Stopped `script/cibuild` from reporting a green it did
-  not earn (issue #85). A bare `docker build .` let Docker serve the
-  check layers from the layer cache whenever the tree had not changed:
-  the checks never executed and the build still exited 0. The fix is an
-  `ARG CHECK_EPOCH` declared immediately above the check `RUN`s in both
-  the lint stage and the builder stage (`ARG` scope is per-stage, so
-  each declares its own), with `script/cibuild` assigning
-  `epoch="$(date +%s)"` and passing `--build-arg CHECK_EPOCH="$epoch"`.
-  The assignment is separate on purpose: under `set -eu` a command
-  substitution that fails inside an argument does not abort the script,
-  which would leave an empty constant `CHECK_EPOCH` and restore the
-  very false green being fixed. Placement is the rest of the point —
-  the `ARG` sits below the `apk add`, `COPY go.mod go.sum`, and `go mod
-  download` layers, so only the checks are invalidated and the
-  dependency layers still cache. The guarantee is conditional on a
-  fresh value rather than absolute: a bare `docker build .` gets an
-  empty `CHECK_EPOCH` and can still serve the check layers from cache,
-  which `README.md` and the `Dockerfile` now say plainly, with issue
-  #91 tracking the upstream hardening (expanded `ARG` form, unset
-  guard, per-invocation epoch, `script/docker`) that would close it.
-  Verified by re-running the reproduction plus the withheld-`--build-arg`
-  counterfactual; the measurements are recorded once, in the PR #89
-  verification comment, rather than restated here. `.golangci.yml`, the
-  lint-stage `FROM` line and its digest, `script/lint`, and
-  `.gitea/workflows/check.yml` are all untouched.
-- 2026-08-09: Corrected the `Vaultik.UI` doc comment (issue #84). It
-  claimed the cli layer replaces the writer with a discarding one in
-  `--cron` mode; the actual mechanism is `UI.SetQuiet(true)` in
-  `setupGlobals`, which drops Begin/Complete/Info/Notice/Detail/
-  Progress/Banner but still emits Warning and Error. The `--cron` line
-  in `README.md` said "Silent unless error", which understated what
-  survives, and now names warnings too. The other `--cron` comments
-  (`internal/log/log.go`, `internal/cli/snapshot.go`,
-  `internal/vaultik/snapshot.go`) were audited and already accurate.
-  Comments and docs only, no behavior change.
-- 2026-08-09: Made `snapshot list` list the destination store without
-  the private key (issue #64). The listing is now the union of the
-  local index and a single streamed listing of the `metadata/` prefix,
-  with no `age_secret_key` gate — the manifest is unencrypted, so a
-  host holding only the public key can enumerate its own backups and a
-  host that lost its local index can still see them. A remote-only
-  snapshot's hostname and name are deliberately not recovered (they are
-  not recoverable without the private key, and making them so would
-  undo the privacy property tracked in issue #81); such rows are
-  labelled by an abbreviation of their remote key and carry the real
-  timestamp and compressed size from the manifest, with `<remote only>`
-  in the two columns that require the local index. Local-only snapshots
-  are reported as drift, and the hint now names `vaultik prune`, which
-  exists, instead of `vaultik snapshot cleanup`, which does not.
-  `reportRemoteDrift` collapsed into the merged view. Every remote
-  manifest read in the codebase now goes through
-  `downloadManifestByKey`, so issue #81 has one call site to change.
-  Review rework: snapshot timestamps now normalize to UTC in
-  `scanSnapshotRows`, the one place they enter the domain, so the merged
-  TIMESTAMP column cannot show local time for a locally tracked row and
-  UTC for a remote-only row on a non-UTC host; `GetIncompleteByHostname`
-  was folded onto that same scanner. `--json` now reports the
-  unreadable-manifest count and the 1000-row truncation on stderr
-  instead of returning a silently short document (the document's shape
-  is unchanged). The two per-snapshot `log.Warn` calls on the listing
-  path now route through the same JSON-aware writer as the existing
-  workaround, so one corrupt manifest can no longer put a log line on
-  stdout ahead of the document and break `| jq` — still a local
-  workaround pending issue #82. Verified with `script/cibuild` and with
-  an uncached `make check` (`0 issues.`, no cached test packages), plus
-  end to end against a `file://` destination with no secret key present.
-- 2026-08-09: Closed the gap between `make lint` and CI (issue #78).
-  `script/lint` now runs the digest-pinned `golangci-lint` image taken
-  from the `Dockerfile` lint stage, which is the single source of truth
-  for the linter version; the duplicate pin in the `Makefile` `deps`
-  target and the unpinned `golangci-lint` install in `script/bootstrap`
-  are gone. A `golangci-lint` on `PATH` is used only when its version is
-  exactly the pinned one (which is how the lint stage runs it inside the
-  container); anything else goes through Docker, and a missing or
-  unreachable Docker daemon is a hard error rather than a silent
-  fallback. Only the **lint** leg of `make check` became equivalent to
-  `script/cibuild`; its tests and `gofmt` still run on the host against
-  the host toolchain, as `README.md` states. An earlier version of this
-  entry claimed `make check` was "as trustworthy as `script/cibuild`"
-  outright, which overstated it; corrected under issue #80.
-- 2026-08-09: Finished the lint remediation under the canonical
-  `.golangci.yml` (issue #61, which also unblocks issue #59). The
-  remaining findings were fixed behavior-preservingly: `wsl_v5`
-  whitespace, `sqlclosecheck`, and `prealloc`. The `sqlclosecheck` sites
-  now close `sql.Rows` in a deferred closure instead of via the
-  `CloseRows` helper, which the linter could not see through. Only the
-  `revive` package-name findings remain suppressed, with per-site
-  `//nolint` directives; the package-rename question behind them is
-  tracked in issue #76. Verified with `script/cibuild`, which exits 0 —
-  that is the only trustworthy gate, because `script/lint` runs whatever
-  `golangci-lint` happens to be on `PATH` rather than the pinned
-  v2.12.2 that CI and the `Dockerfile` use, so `make check` can report
-  green on findings CI still fails. That tooling gap is tracked in issue
-  #78.
-- 2026-08-09: The earlier next step "reconcile the uncommitted
-  `ARCHITECTURE.md` edits on `main`" needed no work: the working tree is
-  clean and `ARCHITECTURE.md` is committed on `main`.
-- 2026-08-07: Updated golangci-lint to v2.12.2 everywhere it is pinned
-  (`Dockerfile` lint stage, `Makefile` deps target), replaced
-  `.golangci.yml` with the canonical config (v2 schema, `default: all`),
-  and remediated the bulk of the lint findings it surfaced (issue #61):
-  behavior-preserving fixes across every package, 2,990 findings down to
-  80. `make test` and `make fmt-check` were green at that point but
-  `make lint` was still red; the commit message claiming `make check`
-  was green was wrong.
-- 2026-08-07: Added the standard `.golangci.yml` and `.editorconfig`
-  (issue #59); lint findings under the new config are tracked in issue
-  #61.
-- 2026-07-07 Adopted scripts-to-rule-them-all: `script/` entrypoints,
-  Makefile shims, README Entrypoints section
-- 2026-07-02: Consolidated CLI verbs, retired overlapping commands; bound
-  the local index to its backup destination URL.
-- 2026-06-28: snapshot rm now removes metadata only and prints the prune
-  command; restore skips chown when running as non-root.
-- 2026-06-26: Snapshot IDs hashed at the storage boundary; snapshot list
-  made resilient to bad remote entries.
-- 2026-06-24: Collapsed snapshot prune into vaultik prune; restore streams
-  blobs to disk and restores files in blob-locality order; cron output
-  fixes.
-- 2026-06-17: Restore overhaul: ReadAt chunk reads from cached blobs,
-  reference-counted blob sweeper, integration tests; new internal/ui
-  output layer, banner, and progress lines.
-- 2025-12-18: Added ARCHITECTURE.md and godoc coverage for exported API.
-- 2025-07-26: End-to-end integration tests; manifest format refactor;
-  renamed backup to snapshot; afero filesystem abstraction.
-- 2025-07-20: Initial design and implementation: cobra + fx CLI skeleton,
-  SQLite index database, UUID blob storage with streaming chunking.
-
-# Future Steps
-
-None queued; the release-scoping item is now the Next Step.

cmd/vaultik/dockermoddownload_test.go

@@ -0,0 +1,45 @@
+package main_test
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// TestDockerfileBoundsEveryModuleDownload fails unless every
+// `go mod download` in the Dockerfile runs under a time limit, is tried
+// a fixed number of times, and then fails the build with a message that
+// names the module download (issue #351). Like the version tests, this
+// parses the Dockerfile; that a stalled download really is cut off and
+// retried was checked by hand.
+func TestDockerfileBoundsEveryModuleDownload(t *testing.T) {
+	t.Parallel()
+
+	var downloads []string
+
+	for _, instruction := range instructions(t, productDockerfile) {
+		if strings.Contains(instruction, "go mod download") {
+			downloads = append(downloads, instruction)
+		}
+	}
+
+	require.NotEmpty(t, downloads, "%s must download modules",
+		productDockerfile)
+
+	for _, download := range downloads {
+		assert.Contains(t, download, "timeout 300 go mod download",
+			"%s: a module download must have a time limit: %s",
+			productDockerfile, download)
+		assert.Contains(t, download, "for attempt in 1 2 3",
+			"%s: a module download must be retried: %s",
+			productDockerfile, download)
+		assert.Contains(t, download, "module download failed",
+			"%s: a module download that never succeeds must say so: %s",
+			productDockerfile, download)
+		assert.True(t, strings.HasSuffix(download, "exit 1"),
+			"%s: a module download that never succeeds must fail the"+
+				" build: %s", productDockerfile, download)
+	}
+}

cmd/vaultik/dockerversion_test.go

@@ -11,92 +11,139 @@ import (
 // This file guards the version stamping of the product image (issue
 // #75). The failure it protects against is silent: the image still
 // builds and runs, but `vaultik version` inside it reports "commit:
-// unknown", so an operator cannot tell which source produced a given
-// backup. .dockerignore excludes .git, so the build cannot derive the
-// commit itself; the values must be computed on the host and passed in.
+// unknown" or a version of "dev", so an operator cannot tell which
+// source produced a given backup. The build takes the version as a
+// build arg, which script/docker computes on the host, and otherwise
+// derives it from the .git in its context; the commit and its date
+// always come from that .git.
 //
-// These are parses of the committed files, for the same reason the lint
-// guards next door are: shelling out to docker would nest a build
-// inside `make test`. That `vaultik version` in the built image really
-// prints the host's version is verified by hand and recorded on the
-// pull request.
+// These are parses of the committed files, because shelling out to
+// docker would nest a build inside `make test`. That `vaultik version`
+// in the built image really prints the host's version is verified by
+// hand.
 
-// dockerScript is script/docker, relative to the repository root.
-const dockerScript = "script/docker"
+// The files under guard, relative to the repository root.
+const (
+	productDockerfile = "Dockerfile"
+	dockerScript      = "script/docker"
+)
 
-// versionArgs are the ldflag targets the build stamps and, matching
-// them, the build args the host must supply. The names line up so the
-// same list checks both files.
-func versionArgs() []string {
-	return []string{"VERSION", "COMMIT", "COMMIT_DATE"}
-}
-
-// TestProductDockerfileTakesVersionAsBuildArgs fails unless the build
-// declares each version arg and stamps it into the binary by ldflag
-// reference, rather than computing it in the container.
-func TestProductDockerfileTakesVersionAsBuildArgs(t *testing.T) {
+// TestProductDockerfileTakesVersionAsBuildArg fails unless the build
+// declares ARG VERSION, with no default, and stamps it into the binary
+// whenever it is given, ahead of the value derived in the container.
+func TestProductDockerfileTakesVersionAsBuildArg(t *testing.T) {
 	t.Parallel()
 
 	found := instructions(t, productDockerfile)
 
-	for _, arg := range versionArgs() {
-		require.GreaterOrEqual(t, indexOf(found, "ARG "+arg), 0,
-			"%s must declare `ARG %s` so the host can pass it in",
-			productDockerfile, arg)
+	require.Contains(t, found, "ARG VERSION",
+		"%s must declare `ARG VERSION`, with no default, so the host can"+
+			" pass it in", productDockerfile)
 
-		assertLdflagReferences(t, found, arg)
-	}
+	assertLdflagReferences(t, found, "VERSION")
 }
 
-// TestProductDockerfileDoesNotDeriveVersionItself is the anti-regression
-// for the original defect: the container ran `git rev-parse`, but .git
-// is not in the build context, so it always resolved to "unknown". No
-// git command may reach into a build that cannot see the history.
-func TestProductDockerfileDoesNotDeriveVersionItself(t *testing.T) {
+// TestProductDockerfileDerivesVersionFromGit fails unless a build given
+// no VERSION, such as a plain `docker build .` of a clone, takes it from
+// `git describe` of the .git in its context, stamps "dev" when the
+// context has no .git, and fails rather than stamp "dev" when that .git
+// yields no version. The commit and its date come from the same .git,
+// and the build fails rather than stamp them "unknown" when it is
+// present.
+func TestProductDockerfileDerivesVersionFromGit(t *testing.T) {
 	t.Parallel()
 
-	text := instructionText(readRepoFile(t, productDockerfile))
+	found := instructions(t, productDockerfile)
 
-	assert.NotContains(t, text, "git ",
-		"%s must not run git: .git is excluded from the build context, so"+
-			" any value it derives is wrong. Pass version, commit and date"+
-			" in as build args instead.", productDockerfile)
+	buildAt := indexContaining(found, "go build")
+	require.GreaterOrEqual(t, buildAt, 0, "%s must build", productDockerfile)
+
+	assert.Contains(t, found[buildAt], "git describe --tags --always || echo dev",
+		"%s must derive the version from git when no VERSION is given,"+
+			" and stamp dev when the context has no .git", productDockerfile)
+	assert.Contains(t, found[buildAt], "[ -e .git ]",
+		"%s must fail when the context carries .git but yields no version",
+		productDockerfile)
+	assert.Contains(t, found[buildAt], "git rev-parse HEAD",
+		"%s must stamp the commit from git", productDockerfile)
+	assert.Contains(t, found[buildAt], "git show -s --format=%cs HEAD",
+		"%s must stamp the commit date from git", productDockerfile)
+	assert.Contains(t, found[buildAt],
+		`[ "$commit" = unknown ] || [ "$commit_date" = unknown ]`,
+		"%s must fail when the context carries .git but yields no commit"+
+			" or date", productDockerfile)
 }
 
 // TestDockerScriptComputesVersionOnTheHost fails unless script/docker
-// derives each value where .git exists and passes it as a build arg,
-// with VERSION coming from script/version so a Docker build reports the
-// same string a local build of the same tree would.
+// passes the version it derives where .git exists as a build arg.
 func TestDockerScriptComputesVersionOnTheHost(t *testing.T) {
 	t.Parallel()
 
 	script := readRepoFile(t, dockerScript)
 
-	for _, arg := range versionArgs() {
-		assert.Contains(t, script, "--build-arg "+arg+"=",
-			"%s must pass --build-arg %s to the build", dockerScript, arg)
-	}
-
-	assert.Contains(t, script, "/version",
-		"%s must take VERSION from script/version, the source of truth"+
-			" shared with the Makefile", dockerScript)
+	assert.Contains(t, script, "--build-arg VERSION=",
+		"%s must pass --build-arg VERSION to the build", dockerScript)
 }
 
-// assertLdflagReferences fails unless some build instruction stamps the
-// named variable from the ARG (a ${arg} reference), not from a value
-// computed inside the container.
+// assertLdflagReferences fails unless the build instruction uses the
+// named ARG whenever it is given (a ${arg:- reference), so a value
+// passed in is not overridden by one derived inside the container.
 func assertLdflagReferences(t *testing.T, found []string, arg string) {
 	t.Helper()
 
 	for _, instruction := range found {
 		if strings.HasPrefix(instruction, "RUN ") &&
 			strings.Contains(instruction, "go build") &&
-			strings.Contains(instruction, "${"+arg+"}") {
+			strings.Contains(instruction, "${"+arg+":-") {
 			return
 		}
 	}
 
 	assert.Fail(t, "version arg is declared but never stamped",
-		"the go build in %s must reference ${%s} in its ldflags, or the"+
-			" arg is passed and discarded", productDockerfile, arg)
+		"the go build in %s must use ${%s:-...}, or the arg is passed and"+
+			" discarded", productDockerfile, arg)
+}
+
+// instructions returns the Dockerfile's instructions, one per element,
+// with comments and blank lines dropped and continuation lines joined,
+// so a multi-line RUN is one string.
+func instructions(t *testing.T, name string) []string {
+	t.Helper()
+
+	var (
+		out         []string
+		continued   string
+		isContinued bool
+	)
+
+	for line := range strings.SplitSeq(readRepoFile(t, name), "\n") {
+		trimmed := strings.TrimSpace(line)
+		if !isContinued && (trimmed == "" || strings.HasPrefix(trimmed, "#")) {
+			continue
+		}
+
+		isContinued = strings.HasSuffix(trimmed, `\`)
+		continued += strings.TrimSuffix(trimmed, `\`)
+
+		if isContinued {
+			continue
+		}
+
+		out = append(out, strings.Join(strings.Fields(continued), " "))
+		continued = ""
+	}
+
+	return out
+}
+
+// indexContaining returns the position of the first instruction
+// containing want; -1 if there is none.
+func indexContaining(found []string, want string) int {
+	for i, instruction := range found {
+		if strings.Contains(instruction, want) {
+			return i
+		}
+	}
+
+	return -1
 }

cmd/vaultik/lintdocker_test.go

@@ -1,367 +0,0 @@
-package main_test
-
-import (
-	"os"
-	"path/filepath"
-	"strings"
-	"testing"
-
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-)
-
-// This file guards the shape of the lint gate. Every property asserted
-// here is one whose loss is SILENT: the build still exits 0, the gate
-// still looks green, and nothing was linted or tested.
-//
-// The gate is a build step. script/lint builds Dockerfile.lint, which
-// runs golangci-lint as a RUN instruction, so a successful build is a
-// clean lint. BuildKit will happily replay that RUN from cache on an
-// unchanged tree in well under a second, which is why the check layers
-// are keyed on a CHECK_EPOCH build arg that the calling script
-// regenerates per invocation, and why an empty value is a hard error
-// rather than a stable cache key.
-//
-// These are parses rather than invocations. Shelling out to docker from
-// the test suite would nest a build inside `make test`, which itself
-// runs inside a build in CI. The one property a parse cannot establish
-// -- that a real finding actually fails the build -- is verified by
-// hand against a deliberately broken tree, recorded on the pull
-// request.
-//
-// One property is deliberately NOT tested here: that no script runs the
-// linter on the host. script/lint is the only lint entry point, and it
-// runs golangci-lint only inside the container; keeping it that way is a
-// review matter, not something a test in this file establishes.
-
-// The files under guard, relative to the repository root.
-const (
-	lintDockerfile    = "Dockerfile.lint"
-	productDockerfile = "Dockerfile"
-	lintScript        = "script/lint"
-	cibuildScript     = "script/cibuild"
-)
-
-// linterBinary is the linter's command name, used to locate the
-// config-verify and lint steps in Dockerfile.lint.
-const linterBinary = "golangci-lint"
-
-// checkEpochARG is the declaration, with no default value. A default
-// would satisfy the non-empty guard with a constant, and a constant is
-// a stable cache key: the checks would be replayed from cache forever
-// after the first build.
-const checkEpochARG = "ARG CHECK_EPOCH"
-
-// checkEpochGuard is what turns a build that omits --build-arg into a
-// loud failure instead of a quiet green. Failed steps are never cached,
-// so it fires on every such invocation rather than once.
-const checkEpochGuard = `RUN [ -n "$CHECK_EPOCH" ] || exit 1`
-
-// freshEpoch is the epoch computation the calling scripts must use, as
-// a bare assignment on its own line. Inline in an argument, a failing
-// `date` would not abort under `set -eu`; CHECK_EPOCH would become the
-// empty string, and the guard above would be the only thing standing
-// between that and a permanently cached green. `$$` is required because
-// `date +%s` is second-granular and busybox silently drops `%N`, so
-// without the pid two concurrent runs in one second can collide.
-const freshEpoch = `epoch="$(date +%s%N)$$"`
-
-// TestLintDockerfilePinsTheLinterByDigest fails if the lint image stops
-// being pinned. An unpinned tag makes the gate's verdict depend on
-// whatever the registry currently serves under that name.
-func TestLintDockerfilePinsTheLinterByDigest(t *testing.T) {
-	t.Parallel()
-
-	from := ""
-
-	for _, instruction := range instructions(t, lintDockerfile) {
-		if strings.HasPrefix(instruction, "FROM ") {
-			from = instruction
-
-			break
-		}
-	}
-
-	require.NotEmpty(t, from, "%s declares no FROM", lintDockerfile)
-	assert.Contains(t, from, "golangci/golangci-lint",
-		"the lint image must be the golangci-lint image")
-	assert.Contains(t, from, "@sha256:",
-		"the lint image must be pinned by digest, not by tag alone")
-}
-
-// TestLintDockerfileCannotBeCachedGreen pins the whole cache-busting
-// mechanism in the file that lints: the declaration with no default,
-// the non-empty guard, and the value expanded into the lint command
-// itself rather than merely declared.
-func TestLintDockerfileCannotBeCachedGreen(t *testing.T) {
-	t.Parallel()
-
-	found := instructions(t, lintDockerfile)
-
-	argAt := indexOf(found, checkEpochARG)
-	require.GreaterOrEqual(t, argAt, 0,
-		"%s must declare `%s` with no default value",
-		lintDockerfile, checkEpochARG)
-
-	assert.GreaterOrEqual(t, indexOf(found, checkEpochGuard), argAt,
-		"%s must guard against an empty CHECK_EPOCH with `%s`",
-		lintDockerfile, checkEpochGuard)
-
-	assertEpochExpandedInto(t, found[argAt:], "golangci-lint run")
-
-	// Dependency layers must stay above the ARG, or every lint run
-	// re-downloads the module cache and the inner loop becomes
-	// unusable.
-	download := indexOf(found, "RUN go mod download")
-	require.GreaterOrEqual(t, download, 0,
-		"%s must download modules in their own layer", lintDockerfile)
-	assert.Less(t, download, argAt,
-		"`%s` must come after `go mod download` so dependency layers"+
-			" still cache", checkEpochARG)
-}
-
-// TestLintDockerfileVerifiesTheLinterConfig guards the validation of
-// .golangci.yml itself. `golangci-lint run` rejects a config it cannot
-// parse but silently IGNORES an unknown top-level key, so renaming
-// `linters:` to `linterz:` discards `default: all` and every threshold
-// and still exits 0 reporting no issues. `config verify` is what turns
-// that into a failure, and it has to run BEFORE the lint, or the lint
-// spends a minute reporting a verdict from a config already known to be
-// wrong.
-func TestLintDockerfileVerifiesTheLinterConfig(t *testing.T) {
-	t.Parallel()
-
-	found := instructions(t, lintDockerfile)
-	verify := linterBinary + " config verify"
-
-	verifyAt := indexContaining(found, verify)
-	require.GreaterOrEqual(t, verifyAt, 0,
-		"%s must run `%s --config .golangci.yml`: without it a typo'd"+
-			" top-level key in .golangci.yml is silently ignored and the"+
-			" gate passes with only the default linter set", lintDockerfile,
-		verify)
-
-	runAt := indexContaining(found, linterBinary+" run")
-	require.GreaterOrEqual(t, runAt, 0, "%s must lint", lintDockerfile)
-	assert.Less(t, verifyAt, runAt,
-		"%s must verify the config before linting with it", lintDockerfile)
-
-	// Keyed on the epoch like every other check layer, so it executes
-	// per invocation rather than being replayed. A cached validation
-	// validates nothing.
-	assertEpochExpandedInto(t, found, verify)
-}
-
-// TestProductDockerfileCannotBeCachedGreen holds the same line for the
-// checks that remain in the product image build.
-func TestProductDockerfileCannotBeCachedGreen(t *testing.T) {
-	t.Parallel()
-
-	found := instructions(t, productDockerfile)
-
-	argAt := indexOf(found, checkEpochARG)
-	require.GreaterOrEqual(t, argAt, 0,
-		"%s must declare `%s` with no default value",
-		productDockerfile, checkEpochARG)
-
-	assert.GreaterOrEqual(t, indexOf(found, checkEpochGuard), argAt,
-		"%s must guard against an empty CHECK_EPOCH", productDockerfile)
-
-	assertEpochExpandedInto(t, found[argAt:], "make fmt-check")
-	assertEpochExpandedInto(t, found[argAt:], "make test")
-}
-
-// TestProductDockerfileDoesNotLint records the split deliberately: the
-// linter lives in Dockerfile.lint and nowhere else, so there is exactly
-// one digest pinning it. A lint stage reintroduced here would either be
-// docker-in-docker (`make lint` is now `docker build`) or a second,
-// independently bumpable pin.
-func TestProductDockerfileDoesNotLint(t *testing.T) {
-	t.Parallel()
-
-	contents := readRepoFile(t, productDockerfile)
-
-	for _, forbidden := range []string{"golangci", "make lint"} {
-		assert.NotContains(t, instructionText(contents), forbidden,
-			"%s must not lint: the linter is pinned once, in %s",
-			productDockerfile, lintDockerfile)
-	}
-}
-
-// TestLintScriptBuildsTheLintDockerfileWithAFreshEpoch is the other
-// half of the mechanism. The Dockerfile's guard only rejects an EMPTY
-// epoch; a constant non-empty one would satisfy it and still be served
-// from cache forever.
-func TestLintScriptBuildsTheLintDockerfileWithAFreshEpoch(t *testing.T) {
-	t.Parallel()
-
-	script := readRepoFile(t, lintScript)
-
-	assertBareEpochAssignment(t, script, lintScript)
-	assert.Contains(t, script, `--build-arg CHECK_EPOCH="$epoch"`,
-		"%s must pass the fresh epoch to the build", lintScript)
-	assert.Contains(t, script, lintDockerfile,
-		"%s must build %s", lintScript, lintDockerfile)
-}
-
-// TestCibuildBuildsBothDockerfilesWithFreshEpochs guards the CI gate:
-// dropping either build silently removes a whole class of check from
-// CI while leaving it green.
-func TestCibuildBuildsBothDockerfilesWithFreshEpochs(t *testing.T) {
-	t.Parallel()
-
-	script := readRepoFile(t, cibuildScript)
-
-	assertBareEpochAssignment(t, script, cibuildScript)
-	assert.Equal(t, 2, strings.Count(script, freshEpoch),
-		"%s must compute a fresh epoch for each of its two builds",
-		cibuildScript)
-	assert.Equal(t, 2,
-		strings.Count(script, `--build-arg CHECK_EPOCH="$epoch"`),
-		"%s must pass a fresh epoch to both builds", cibuildScript)
-	assert.Contains(t, script, "-f Dockerfile.lint",
-		"%s must build %s", cibuildScript, lintDockerfile)
-}
-
-// assertEpochExpandedInto fails unless some instruction runs the named
-// command with the epoch expanded into it. Expansion, not mere
-// declaration: an ARG that no instruction references is not guaranteed
-// to key the layer, and the expansion also puts the value in the build
-// log where a reader can see the layer was keyed fresh.
-func assertEpochExpandedInto(t *testing.T, found []string, command string) {
-	t.Helper()
-
-	for _, instruction := range found {
-		if !strings.HasPrefix(instruction, "RUN ") {
-			continue
-		}
-
-		if strings.Contains(instruction, command) &&
-			strings.Contains(instruction, "${CHECK_EPOCH}") {
-			return
-		}
-	}
-
-	assert.Fail(t, "no epoch-keyed layer runs the command",
-		"`%s` must run in a layer that expands ${CHECK_EPOCH}, or it"+
-			" will be replayed from cache without executing", command)
-}
-
-// assertBareEpochAssignment fails unless the script computes the epoch
-// as a bare assignment on its own line.
-func assertBareEpochAssignment(t *testing.T, script, name string) {
-	t.Helper()
-
-	for line := range strings.SplitSeq(script, "\n") {
-		if strings.TrimSpace(line) == freshEpoch {
-			return
-		}
-	}
-
-	assert.Fail(t, "no bare epoch assignment",
-		"%s must compute `%s` as a bare assignment on its own line, so"+
-			" `set -e` catches a failing date instead of quietly"+
-			" building with an empty epoch", name, freshEpoch)
-}
-
-// instructions returns the Dockerfile's instructions, one per element,
-// with comments and blank lines dropped and continuation lines joined,
-// so a multi-line RUN is one string.
-func instructions(t *testing.T, name string) []string {
-	t.Helper()
-
-	return strings.Split(instructionText(readRepoFile(t, name)), "\n")
-}
-
-// instructionText is instructions' parse, before splitting: it is also
-// what a "must not contain" assertion should look at, so that a word
-// appearing only in a comment is not mistaken for behaviour.
-func instructionText(contents string) string {
-	var (
-		out         []string
-		continued   string
-		isContinued bool
-	)
-
-	for line := range strings.SplitSeq(contents, "\n") {
-		trimmed := strings.TrimSpace(line)
-		if !isContinued && (trimmed == "" || strings.HasPrefix(trimmed, "#")) {
-			continue
-		}
-
-		isContinued = strings.HasSuffix(trimmed, `\`)
-		continued += strings.TrimSuffix(trimmed, `\`)
-
-		if isContinued {
-			continue
-		}
-
-		out = append(out, strings.Join(strings.Fields(continued), " "))
-		continued = ""
-	}
-
-	return strings.Join(out, "\n")
-}
-
-// indexOf returns the position of the first instruction equal to, or
-// beginning with, want; -1 if there is none. An `ARG NAME=default`
-// counts as beginning with `ARG NAME`, so a declared arg is found
-// whether or not it carries a default.
-func indexOf(found []string, want string) int {
-	for i, instruction := range found {
-		if instruction == want ||
-			strings.HasPrefix(instruction, want+" ") ||
-			strings.HasPrefix(instruction, want+"=") {
-			return i
-		}
-	}
-
-	return -1
-}
-
-// indexContaining returns the position of the first instruction
-// containing want; -1 if there is none.
-func indexContaining(found []string, want string) int {
-	for i, instruction := range found {
-		if strings.Contains(instruction, want) {
-			return i
-		}
-	}
-
-	return -1
-}
-
-// readRepoFile reads a file by its path relative to the repository
-// root.
-func readRepoFile(t *testing.T, name string) string {
-	t.Helper()
-
-	//nolint:gosec // G304: the path is a constant relative to this repo
-	contents, err := os.ReadFile(filepath.Join(repoRoot(t), name))
-	require.NoError(t, err)
-
-	return string(contents)
-}
-
-// repoRoot returns the repository root. The test binary runs with its
-// package directory as the working directory, so the root is found by
-// walking up until the module file appears.
-func repoRoot(t *testing.T) string {
-	t.Helper()
-
-	dir, err := os.Getwd()
-	require.NoError(t, err)
-
-	for {
-		_, err = os.Stat(filepath.Join(dir, "go.mod"))
-		if err == nil {
-			return dir
-		}
-
-		parent := filepath.Dir(dir)
-		require.NotEqual(t, dir, parent,
-			"walked to the filesystem root without finding a go.mod")
-
-		dir = parent
-	}
-}

cmd/vaultik/makefile_test.go

@@ -1,6 +1,8 @@
 package main_test
 
 import (
+	"os"
+	"path/filepath"
 	"regexp"
 	"slices"
 	"strings"
@@ -84,14 +86,48 @@ func TestBuildTargetBuildsTheBinary(t *testing.T) {
 		"`make build` must depend on the rule that builds the binary")
 }
 
-// readMakefile returns the contents of the repository's Makefile. The
-// root is located by the shared walk in lintdocker_test.go.
+// readMakefile returns the contents of the repository's Makefile.
 func readMakefile(t *testing.T) string {
 	t.Helper()
 
 	return readRepoFile(t, "Makefile")
 }
 
+// readRepoFile reads a file by its path relative to the repository
+// root.
+func readRepoFile(t *testing.T, name string) string {
+	t.Helper()
+
+	//nolint:gosec // G304: the path is a constant relative to this repo
+	contents, err := os.ReadFile(filepath.Join(repoRoot(t), name))
+	require.NoError(t, err)
+
+	return string(contents)
+}
+
+// repoRoot returns the repository root. The test binary runs with its
+// package directory as the working directory, so the root is found by
+// walking up until the module file appears.
+func repoRoot(t *testing.T) string {
+	t.Helper()
+
+	dir, err := os.Getwd()
+	require.NoError(t, err)
+
+	for {
+		_, err = os.Stat(filepath.Join(dir, "go.mod"))
+		if err == nil {
+			return dir
+		}
+
+		parent := filepath.Dir(dir)
+		require.NotEqual(t, dir, parent,
+			"walked to the filesystem root without finding a go.mod")
+
+		dir = parent
+	}
+}
+
 // phonyTargets returns every name declared phony, across all .PHONY
 // lines.
 func phonyTargets(makefile string) []string {

configs/config.example.yml

@@ -3,7 +3,8 @@
 # Copy this file and uncomment/modify the values you need
 
 # Age recipient public keys for encryption
-# This is REQUIRED - backups are encrypted to these public keys
+# Backups are encrypted to these public keys. snapshot create needs at least
+# one; listing, verifying and restoring do not
 # Generate with: age-keygen | grep "public key"
 age_recipients:
   - age1cj2k2addawy294f6k2gr2mf9gps9r3syplryxca3nvxj3daqm96qfp84tz
@@ -286,19 +287,24 @@ storage_url: "rclone://myremote/path/to/backups"
 #  #use_ssl: true
 #
 #  # Part size for multipart uploads
-#  # Minimum 5MB, affects memory usage during upload
-#  # Supports: 5MB, 10M, 100MiB, etc.
-#  # Default: 5MB
-#  #part_size: 5MB
+#  # Minimum 5MiB, maximum 5GiB; affects memory usage during upload
+#  # A blob too large for 10,000 parts of this size gets larger parts
+#  # Supports: 10MB, 16MiB, 100MiB, etc. (5MB is below the minimum)
+#  # Default: 5MiB
+#  #part_size: 5MiB
 
 # Path to local SQLite index database
 # This database tracks file state for incremental backups
-# Default: /var/lib/vaultik/index.sqlite
-#index_path: /var/lib/vaultik/index.sqlite
+# It must be an absolute path; ~ is expanded
+# Default: the platform data directory, e.g.
+#   macOS: ~/Library/Application Support/vaultik/index.sqlite
+#   Linux: ~/.local/share/vaultik/index.sqlite
+#index_path: /path/to/index.sqlite
 
 # Average chunk size for content-defined chunking
 # Smaller chunks = better deduplication but more metadata
-# Supports: 10MB, 5M, 1GB, 500KB, 64MiB, etc.
+# Minimum 1MB
+# Supports: 10MB, 5M, 1GB, 64MiB, etc.
 # Default: 10MB
 #chunk_size: 10MB
 
@@ -309,6 +315,12 @@ storage_url: "rclone://myremote/path/to/backups"
 # Chunking uses no secret (the FastCDC parameters are fixed and public). At a
 # large limit a blob holds hundreds of chunks, so individual chunk lengths are
 # not visible in its size; lowering the limit toward chunk_size exposes them.
+# A backup needs free temporary space, because each blob is written whole to
+# a temporary file before it is uploaded (up to about blob_size_limit; an
+# rclone destination that cannot stream uploads needs about twice that) and
+# the metadata export writes copies of the local index. Temporary files go to
+# $TMPDIR (default /tmp); with TMPDIR unset, SQLite writes one of those copies
+# to /var/tmp.
 # Supports: 1GB, 10G, 500MB, 1GiB, etc.
 # Default: 10GB
 #blob_size_limit: 10GB
@@ -317,6 +329,20 @@ storage_url: "rclone://myremote/path/to/backups"
 # Higher = better compression but slower
 # Default: 3
 compression_level: 5
+
+# How old a blob that no snapshot lists must be before prune deletes it, and
+# how old every object in a directory under metadata/ with no manifest must
+# be before prune deletes that directory
+# A backup uploads its blobs and its db.zst.age before its manifest, so on a
+# destination shared between hosts or users this must be longer than the
+# longest backup run
+# It does not protect a blob a backup reuses from an earlier run that stopped
+# before its manifest: that blob keeps the time of its first upload, so a
+# prune can delete it
+# Supports: 36h, 90m, etc.; 0s deletes such blobs and directories at once
+# Default: 24h
+#prune_grace_period: 24h
+
 # Hostname to use in backup metadata
 # Default: system hostname
 #hostname: myserver

docs/ARCHITECTURE.md

@@ -1,10 +1,14 @@
 # Vaultik Architecture
 
-This document describes the internal architecture of Vaultik, focusing on the data model, type instantiation, and the relationships between core modules.
+This document describes the internal architecture of Vaultik, focusing on the
+data model, type instantiation, and the relationships between core modules.
 
 ## Overview
 
-Vaultik is a backup system that uses content-defined chunking for deduplication and packs chunks into large, compressed, encrypted blobs for efficient cloud storage. The system is built around dependency injection using [uber-go/fx](https://github.com/uber-go/fx).
+Vaultik is a backup system that uses content-defined chunking for deduplication
+and packs chunks into large, compressed, encrypted blobs for efficient cloud
+storage. The system is built around dependency injection using
+[uber-go/fx](https://github.com/uber-go/fx).
 
 ## Data Flow
 
@@ -53,53 +57,88 @@ The database tracks five primary entities and their relationships:
 ### Entity Descriptions
 
 #### File (`database.File`)
-Represents a file, directory, or symlink in the backup system. Stores metadata needed for restoration:
-- Path, source_path (for restore path stripping), mtime
+
+Represents a file, directory, or symlink in the backup system. Stores metadata
+needed for restoration:
+
+- Path, mtime
 - Size, mode, ownership (uid, gid)
 - Symlink target (if applicable)
 
+It also stores `source_path`, the source directory the scan found it under, made
+absolute and with symlinks resolved. Restore does not read it.
+
 #### Chunk (`database.Chunk`)
-A content-addressed unit of data. Files are split into variable-size chunks using the FastCDC algorithm:
+
+A content-addressed unit of data. Files are split into variable-size chunks
+using the FastCDC algorithm:
+
 - `ChunkHash`: SHA256 hash of chunk content (primary key)
 - `Size`: Chunk size in bytes
 
-Chunk sizes vary between `avgChunkSize/4` and `avgChunkSize*4` (2.5MB-40MB for the 10MB default average).
+Chunk sizes vary between `avgChunkSize/4` and `avgChunkSize*4` (2.5MB-40MB for
+the 10MB default average).
 
 #### FileChunk (`database.FileChunk`)
+
 Maps files to their constituent chunks:
+
 - `FileID`: Reference to the file
 - `Idx`: Position of this chunk within the file (0-indexed)
 - `ChunkHash`: Reference to the chunk
 
 #### Blob (`database.Blob`)
-The final storage unit uploaded to S3. Contains many compressed and encrypted chunks:
+
+The final storage unit uploaded to S3. Contains many compressed and encrypted
+chunks:
+
 - `ID`: UUID assigned at creation
-- `Hash`: `hex(SHA256(SHA256(uncompressed blob contents)))`, computed before compression and encryption (see [docs/REPOSTRUCTURE.md](docs/REPOSTRUCTURE.md#blobs-directory-blobs))
+- `Hash`: `hex(SHA256(SHA256(uncompressed blob contents)))`, computed before
+  compression and encryption (see
+  [docs/REPOSTRUCTURE.md](REPOSTRUCTURE.md#blobs-directory-blobs))
 - `UncompressedSize`: Total raw chunk data before compression
 - `CompressedSize`: Size after zstd compression and age encryption
 - `CreatedTS`, `FinishedTS`, `UploadedTS`: Lifecycle timestamps
 
 Blob creation process:
+
 1. Chunks are accumulated (up to MaxBlobSize, typically 10GB)
 2. As each chunk is added, its uncompressed bytes are fed to a running SHA-256
-3. Concurrently, the same bytes are compressed with zstd, then encrypted with age (recipients configured in config), and streamed to storage
-4. On finalize, the blob's name is the double SHA-256 of the uncompressed contents — `hex(SHA256(SHA256(...)))` — not a hash of the compressed, encrypted bytes
-5. Uploaded to `blobs/{hash[0:2]}/{hash[2:4]}/{hash}`
+3. Concurrently, the same bytes are compressed with zstd, then encrypted with
+   age (recipients configured in config), and written to a temporary file
+4. On finalize, the blob's name is the double SHA-256 of the uncompressed
+   contents — `hex(SHA256(SHA256(...)))` — not a hash of the compressed,
+   encrypted bytes
+5. The finished file is uploaded to `blobs/{hash[0:2]}/{hash[2:4]}/{hash}` and
+   then deleted
+
+A backup needs free temporary space, because each blob is written whole to a
+temporary file before it is uploaded (up to about `blob_size_limit`; an rclone
+destination that cannot stream uploads needs about twice that) and the metadata
+export writes copies of the local index. Temporary files go to `$TMPDIR`
+(default `/tmp`); with `TMPDIR` unset, SQLite writes one of those copies to
+`/var/tmp`.
 
 #### BlobChunk (`database.BlobChunk`)
+
 Maps chunks to their position within blobs:
+
 - `BlobID`: Reference to the blob
 - `ChunkHash`: Reference to the chunk
 - `Offset`: Byte offset within the uncompressed blob
 - `Length`: Chunk size
 
 #### Snapshot (`database.Snapshot`)
+
 Represents a point-in-time backup:
-- `ID`: Format is `{hostname}_{snapshot-name}_{RFC3339}` (e.g. `server1_home_2025-06-01T12:00:00Z`)
+
+- `ID`: Format is `{hostname}_{snapshot-name}_{RFC3339}` (e.g.
+  `server1_home_2025-06-01T12:00:00Z`)
 - Tracks file count, chunk count, blob count, sizes, compression ratio
 - `CompletedAt`: Null until snapshot finishes successfully
 
 #### SnapshotFile / SnapshotBlob
+
 Join tables linking snapshots to their files and blobs.
 
 ### Relationship Summary
@@ -136,21 +175,27 @@ fx.New(
 ### Key Type Instantiation Points
 
 #### 1. Config (`config.Config`)
+
 - **Created by**: `config.Module` via `config.LoadConfig()`
 - **When**: Application startup (fx DI)
-- **Contains**: All configuration from YAML file (S3 credentials, encryption keys, paths, etc.)
+- **Contains**: All configuration from YAML file (S3 credentials, encryption
+  keys, paths, etc.)
 
 #### 2. Database (`database.DB`)
+
 - **Created by**: `database.Module` via `database.New()`
 - **When**: Application startup (fx DI)
 - **Contains**: SQLite connection, path reference
 
 #### 3. Repositories (`database.Repositories`)
+
 - **Created by**: `database.Module` via `database.NewRepositories()`
 - **When**: Application startup (fx DI)
-- **Contains**: All repository interfaces (Files, Chunks, Blobs, Snapshots, etc.)
+- **Contains**: All repository interfaces (Files, Chunks, Blobs, Snapshots,
+  etc.)
 
 #### 4. Vaultik (`vaultik.Vaultik`)
+
 - **Created by**: `vaultik.New(VaultikParams)`
 - **When**: Application startup (fx DI)
 - **Contains**: All dependencies for backup operations
@@ -172,11 +217,13 @@ type Vaultik struct {
 ```
 
 #### 5. SnapshotManager (`snapshot.SnapshotManager`)
+
 - **Created by**: `snapshot.Module` via `snapshot.NewSnapshotManager()`
 - **When**: Application startup (fx DI)
 - **Responsibility**: Creates/completes snapshots, exports metadata to S3
 
 #### 6. Scanner (`snapshot.Scanner`)
+
 - **Created by**: `ScannerFactory(ScannerParams)`
 - **When**: Each `CreateSnapshot()` call
 - **Contains**: Chunker, Packer, progress reporter
@@ -190,20 +237,22 @@ scanner := v.ScannerFactory(snapshot.ScannerParams{
 ```
 
 #### 7. Chunker (`chunker.Chunker`)
+
 - **Created by**: `chunker.NewChunker(avgChunkSize)`
 - **When**: Inside `snapshot.NewScanner()`
 - **Configuration**:
-  - `avgChunkSize`: From config (default 10MB)
-  - `minChunkSize`: avgChunkSize / 4
-  - `maxChunkSize`: avgChunkSize * 4
+    - `avgChunkSize`: From config (default 10MB)
+    - `minChunkSize`: avgChunkSize / 4
+    - `maxChunkSize`: avgChunkSize \* 4
 
 #### 8. Packer (`blob.Packer`)
+
 - **Created by**: `blob.NewPacker(PackerConfig)`
 - **When**: Inside `snapshot.NewScanner()`
 - **Configuration**:
-  - `MaxBlobSize`: Maximum blob size before finalization (typically 10GB)
-  - `CompressionLevel`: zstd level (1-19)
-  - `Recipients`: age public keys for encryption
+    - `MaxBlobSize`: Maximum blob size before finalization (typically 10GB)
+    - `CompressionLevel`: zstd level (1-19)
+    - `Recipients`: age public keys for encryption
 
 ```go
 // snapshot/scanner.go: NewScanner()
@@ -220,17 +269,22 @@ packer, err := blob.NewPacker(packerCfg)
 ## Module Responsibilities
 
 ### `internal/cli`
-Entry point for fx application. Combines all modules and handles signal interrupts.
+
+Entry point for fx application. Combines all modules and handles signal
+interrupts.
 
 Key functions:
+
 - `NewApp(AppOptions)` → Creates fx.App with all modules
 - `RunApp(ctx, app)` → Starts app, handles graceful shutdown
 - `RunWithApp(ctx, opts)` → Convenience wrapper
 
 ### `internal/vaultik`
+
 Main orchestrator containing all dependencies and command implementations.
 
 Key methods:
+
 - `New(VaultikParams)` → Constructor (fx DI)
 - `CreateSnapshot(opts)` → Main backup operation
 - `ListSnapshots(jsonOutput)` → List available snapshots
@@ -238,27 +292,34 @@ Key methods:
 - `PurgeSnapshots(...)` → Remove old snapshots
 
 ### `internal/chunker`
+
 Content-defined chunking using FastCDC algorithm.
 
 Key types:
+
 - `Chunk` → Hash, Data, Offset, Size
 - `Chunker` → avgChunkSize, minChunkSize, maxChunkSize
 
 Key methods:
+
 - `NewChunker(avgChunkSize)` → Constructor
-- `ChunkReaderStreaming(reader, callback)` → Stream chunks with callback (preferred)
+- `ChunkReaderStreaming(reader, callback)` → Stream chunks with callback
+  (preferred)
 - `ChunkReader(reader)` → Return all chunks at once (memory-intensive)
 
 ### `internal/blob`
+
 Blob packing: accumulates chunks, compresses, encrypts, tracks metadata.
 
 Key types:
+
 - `Packer` → Thread-safe blob accumulator
 - `ChunkRef` → Hash + Data for adding to packer
 - `FinishedBlob` → Completed blob ready for upload
 - `BlobWithReader` → FinishedBlob + io.Reader for streaming upload
 
 Key methods:
+
 - `NewPacker(PackerConfig)` → Constructor
 - `AddChunk(ChunkRef)` → Add chunk to current blob
 - `FinalizeBlob()` → Compress, encrypt, hash current blob
@@ -268,35 +329,48 @@ Key methods:
 ### `internal/snapshot`
 
 #### Scanner
+
 Orchestrates the backup process for a directory.
 
 Key methods:
+
 - `NewScanner(ScannerConfig)` → Constructor (creates Chunker + Packer)
 - `Scan(ctx, path, snapshotID)` → Main scan operation
 
 Scan phases:
-1. **Phase 0**: Detect deleted files from previous snapshots
-2. **Phase 1**: Walk directory, identify files needing processing
+
+1. **Phase 1**: Walk directory, identify files needing processing
+2. **Phase 1b**: Count as deleted the files that the previous snapshot of the
+   same name lists, the walk did not find, and are gone from disk; a file that
+   is excluded, or inside an excluded directory, is not counted
 3. **Phase 2**: Process files (chunk → pack → upload)
 
 #### SnapshotManager
+
 Manages snapshot lifecycle and metadata export.
 
 Key methods:
+
 - `CreateSnapshot(ctx, hostname, version, commit)` → Create snapshot record
-- `PopulateSnapshotBlobs(ctx, snapshotID)` → Record every blob the snapshot references
+- `PopulateSnapshotBlobs(ctx, snapshotID)` → Record every blob the snapshot
+  references
 - `ExportSnapshotMetadata(ctx, dbPath, snapshotID)` → Export to S3
-- `MarkSnapshotComplete(ctx, snapshotID)` → Record completion, only after a successful export
-- `CompleteSnapshot(ctx, snapshotID)` → Convenience: populate blobs, then mark complete (no export between)
+- `MarkSnapshotComplete(ctx, snapshotID)` → Record completion, only after a
+  successful export
+- `CompleteSnapshot(ctx, snapshotID)` → Convenience: populate blobs, then mark
+  complete (no export between)
 
 ### `internal/database`
+
 SQLite database for local index. Single-writer mode for thread safety.
 
 Key types:
+
 - `DB` → Database connection wrapper
 - `Repositories` → Collection of all repository interfaces
 
 Repository interfaces:
+
 - `FilesRepository` → CRUD for File records
 - `ChunksRepository` → CRUD for Chunk records
 - `BlobsRepository` → CRUD for Blob records
@@ -307,6 +381,8 @@ Repository interfaces:
 
 ```
 CreateSnapshot(opts)
+    │
+    ├─► markDestination()   // Check the marker (see Storage Layout in S3), then write it
     │
     ├─► PruneDatabase()   // Critical: avoid dedup errors
     │
@@ -316,13 +392,13 @@ CreateSnapshot(opts)
     │       │
     │       ├─► scanner.Scan(ctx, path, snapshotID)
     │       │       │
-    │       │       ├─► Phase 0: detectDeletedFiles()
-    │       │       │
     │       │       ├─► Phase 1: scanPhase()
     │       │       │       Walk directory
     │       │       │       Check file metadata changes
     │       │       │       Build list of files to process
     │       │       │
+    │       │       ├─► Phase 1b: detectDeletedFiles()
+    │       │       │
     │       │       └─► Phase 2: processPhase()
     │       │               For each file:
     │       │                   chunker.ChunkReaderStreaming()
@@ -335,10 +411,10 @@ CreateSnapshot(opts)
     │       │
     │       └─► Accumulate statistics
     │
-    ├─► SnapshotManager.UpdateSnapshotStatsExtended()
-    │
     ├─► SnapshotManager.PopulateSnapshotBlobs()   // record referenced blobs
     │
+    ├─► SnapshotManager.UpdateSnapshotStatsExtended()
+    │
     ├─► SnapshotManager.ExportSnapshotMetadata()
     │       │
     │       ├─► Copy database to temp file
@@ -353,16 +429,23 @@ CreateSnapshot(opts)
 
 ## Deduplication Strategy
 
-1. **File-level**: Files unchanged since last backup are skipped (metadata comparison: size, mtime, mode, uid, gid)
+1. **File-level**: Files and symlinks unchanged since last backup are skipped
+   (metadata comparison: size, mtime, mode, uid, gid, and a symlink's target),
+   unless the file lists a chunk that no uploaded blob holds; such a file is
+   re-chunked
 
-2. **Chunk-level**: Chunks are content-addressed by SHA256 hash. If a chunk hash already exists in the database, the chunk data is not re-uploaded.
+2. **Chunk-level**: Chunks are content-addressed by SHA256 hash. If a chunk hash
+   already exists in the database, the chunk data is not re-uploaded.
 
-3. **Blob-level**: Blobs contain only unique chunks. Duplicate chunks within a blob are skipped.
+3. **Blob-level**: Blobs contain only unique chunks. Duplicate chunks within a
+   blob are skipped.
 
 ## Storage Layout in S3
 
 ```
 bucket/
+├── vaultik-destination.txt          # Marker written by every backup
+│
 ├── blobs/
 │   └── {hash[0:2]}/
 │       └── {hash[2:4]}/
@@ -374,15 +457,30 @@ bucket/
         └── manifest.json.zst        # Blob list (for pruning/verification)
 ```
 
+A destination that answers a look-up of `vaultik-destination.txt` with not found
+is treated as not present while the local index holds snapshots stored there
+(`checkDestinationMarker` in `internal/vaultik/destination_marker.go`):
+`snapshot create` refuses to write to it, `prune`, `snapshot purge` and
+`remote nuke` fail, and `snapshot list` and `snapshot remove` behave as for a
+destination that cannot be listed. This keeps those commands from taking an
+unplugged volume's empty mount point for a destination whose snapshots are all
+gone. When the look-up fails for any reason other than not found, for example
+when S3 refuses it to credentials that allow only PUT, the check is skipped, so
+a host with such credentials does not notice an emptied destination; see
+[docs/REPOSTRUCTURE.md](REPOSTRUCTURE.md#marker-file-vaultik-destinationtxt).
+The `file://` backend never creates its destination directory, so a write to a
+missing one fails.
+
 The `{remote-key}` directory name is a one-way double SHA-256 hash of the human
 snapshot ID, so the human ID (hostname, snapshot name, timestamp) is never
 written to the store as a directory name. See
-[docs/REPOSTRUCTURE.md](docs/REPOSTRUCTURE.md#remote-key-derivation) for the
+[docs/REPOSTRUCTURE.md](REPOSTRUCTURE.md#remote-key-derivation) for the
 derivation and a worked example.
 
 ## Thread Safety
 
 - `Packer`: Thread-safe via mutex. Multiple goroutines can call `AddChunk()`.
 - `Scanner`: Uses `packerMu` mutex to coordinate blob finalization.
-- `Database`: Single-writer mode (`MaxOpenConns=1`) ensures SQLite thread safety.
+- `Database`: Single-writer mode (`MaxOpenConns=1`) ensures SQLite thread
+  safety.
 - `Repositories.WithTx()`: Handles transaction lifecycle automatically.

docs/DATAMODEL.md

@@ -2,46 +2,48 @@
 
 ## Overview
 
-Vaultik uses a local SQLite database to track file metadata, chunk mappings, and blob associations during the backup process. This database serves as an index for incremental backups and enables efficient deduplication.
+Vaultik uses a local SQLite database to track file metadata, chunk mappings, and
+blob associations during the backup process. This database serves as an index
+for incremental backups and enables efficient deduplication.
 
 **Important Notes:**
 
-This section is the authoritative explanation of the schema/migration story;
-other documents (the README and `AGENTS.md`) link here.
-
-- **No upgrade path between versions (pre-1.0)**: Vaultik has no supported way to
-  carry an existing local index across a schema change. The index is disposable
-  — if the on-disk schema changes between versions, delete the local SQLite
-  database (`vaultik database delete`) and run a full backup. Remote storage is
-  unaffected; the new index re-deduplicates against existing remote blobs. This
-  is the standing project policy, and it is separate from the schema bootstrap
-  described next.
-- **Schema bootstrap**: a fresh database is populated from numbered SQL files
-  embedded in the binary under `internal/database/schema/`. `000.sql` creates the
-  `schema_migrations` table; `001.sql` creates the application tables. On opening
-  a database the code applies each numbered file that has not yet run and records
-  its version in `schema_migrations`. This bootstraps a new database; it does not
-  upgrade an existing one between released versions.
-- **Changing the schema (pre-1.0)**: edit `internal/database/schema/001.sql` (and
-  the code that touches the affected tables) directly. Do not add new numbered
-  files — there is no installed base to migrate.
-- **Disposability expires at 1.0**: the index is treated as disposable only until
-  1.0 ships and is tagged. Once 1.0 is tagged that clause expires and the
-  question of upgrading existing indexes returns. It is deliberately left open
-  here.
 - **Version Compatibility**: In rare cases, you may need to use the same version
   of Vaultik to restore a backup as was used to create it. This ensures
   compatibility with the metadata format stored in S3.
 
+## Schema Migrations
+
+Vaultik supports schema migrations. They are the numbered SQL files in
+`internal/database/schema/`, embedded in the binary: `000.sql` creates the
+`schema_migrations` table, which records each migration that has run, and
+`001.sql` creates the application tables. `database.New` opens a database and
+applies, in order, every migration that database has not yet recorded.
+
+**Before 1.0** no migrations are added, because nothing is installed anywhere
+yet. A schema change edits `001.sql` (and the code that uses the affected
+tables) directly. A local database created before the change has already
+recorded `001.sql` as run, so it keeps the old schema and can become unusable;
+`vaultik database delete` followed by a full backup rebuilds it.
+
+**After 1.0** each schema change is a new numbered file, so an existing local
+database is migrated the first time an updated vaultik opens it. A file that has
+shipped in a release is never edited.
+
 ## Database Tables
 
 ### 1. `files`
+
 Stores metadata about files in the filesystem being backed up.
 
 **Columns:**
+
 - `id` (TEXT PRIMARY KEY) - UUID for the file record
 - `path` (TEXT NOT NULL UNIQUE) - Absolute file path
-- `mtime` (INTEGER NOT NULL) - Modification time as Unix timestamp
+- `mtime` (INTEGER NOT NULL) - Modification time, whole seconds since the Unix
+  epoch
+- `mtime_nsec` (INTEGER NOT NULL) - Nanoseconds within that second, 0 to
+  999999999
 - `size` (INTEGER NOT NULL) - File size in bytes
 - `mode` (INTEGER NOT NULL) - Unix file permissions and type
 - `uid` (INTEGER NOT NULL) - User ID of file owner
@@ -49,23 +51,30 @@ Stores metadata about files in the filesystem being backed up.
 - `link_target` (TEXT) - Symlink target path (NULL for regular files)
 
 **Indexes:**
+
 - `idx_files_path` on `path` for efficient lookups
 
-**Purpose:** Tracks file metadata to detect changes between backup runs. Used for incremental backup decisions. The UUID primary key provides stable references that don't change if files are moved.
+**Purpose:** Tracks file metadata to detect changes between backup runs. Used
+for incremental backup decisions. The UUID primary key provides stable
+references that don't change if files are moved.
 
 ### 2. `chunks`
+
 Stores information about content-defined chunks created from files.
 
 **Columns:**
+
 - `chunk_hash` (TEXT PRIMARY KEY) - SHA256 hash of chunk content
 - `size` (INTEGER NOT NULL) - Chunk size in bytes
 
 **Purpose:** Enables deduplication by tracking unique chunks across all files.
 
 ### 3. `file_chunks`
+
 Maps files to their constituent chunks in order.
 
 **Columns:**
+
 - `file_id` (TEXT) - File ID (FK to files.id)
 - `idx` (INTEGER) - Chunk index within file (0-based)
 - `chunk_hash` (TEXT) - Chunk hash (FK to chunks.chunk_hash)
@@ -74,35 +83,49 @@ Maps files to their constituent chunks in order.
 **Purpose:** Allows reconstruction of files from chunks during restore.
 
 ### 4. `chunk_files`
+
 Reverse mapping showing which files contain each chunk.
 
 **Columns:**
+
 - `chunk_hash` (TEXT) - Chunk hash (FK to chunks.chunk_hash)
 - `file_id` (TEXT) - File ID (FK to files.id)
 - `file_offset` (INTEGER) - Byte offset of chunk within file
 - `length` (INTEGER) - Length of chunk in bytes
 - PRIMARY KEY (`chunk_hash`, `file_id`)
 
-**Purpose:** Supports efficient queries for chunk usage and deduplication statistics.
+**Purpose:** Supports efficient queries for chunk usage and deduplication
+statistics.
 
 ### 5. `blobs`
+
 Stores information about packed, compressed, and encrypted blob files.
 
 **Columns:**
+
 - `id` (TEXT PRIMARY KEY) - UUID assigned when blob creation starts
-- `blob_hash` (TEXT UNIQUE) - `hex(SHA256(SHA256(uncompressed blob contents)))`, computed before compression and encryption (NULL until finalized); see [REPOSTRUCTURE.md](REPOSTRUCTURE.md#blobs-directory-blobs)
+- `blob_hash` (TEXT UNIQUE) - `hex(SHA256(SHA256(uncompressed blob contents)))`,
+  computed before compression and encryption (NULL until finalized); see
+  [REPOSTRUCTURE.md](REPOSTRUCTURE.md#blobs-directory-blobs)
 - `created_ts` (INTEGER NOT NULL) - Creation timestamp
 - `finished_ts` (INTEGER) - Finalization timestamp (NULL if in progress)
-- `uncompressed_size` (INTEGER NOT NULL DEFAULT 0) - Total size of chunks before compression
-- `compressed_size` (INTEGER NOT NULL DEFAULT 0) - Size after compression and encryption
+- `uncompressed_size` (INTEGER NOT NULL DEFAULT 0) - Total size of chunks before
+  compression
+- `compressed_size` (INTEGER NOT NULL DEFAULT 0) - Size after compression and
+  encryption. Once uploaded, the size of the object in storage: when the blob
+  was already stored, under another compression level or other recipients, this
+  can differ from the size of the local copy
 - `uploaded_ts` (INTEGER) - Upload completion timestamp (NULL if not uploaded)
 
-**Purpose:** Tracks blob lifecycle from creation through upload. The UUID primary key allows immediate association of chunks with blobs.
+**Purpose:** Tracks blob lifecycle from creation through upload. The UUID
+primary key allows immediate association of chunks with blobs.
 
 ### 6. `blob_chunks`
+
 Maps chunks to the blobs that contain them.
 
 **Columns:**
+
 - `blob_id` (TEXT) - Blob ID (FK to blobs.id)
 - `chunk_hash` (TEXT) - Chunk hash (FK to chunks.chunk_hash)
 - `offset` (INTEGER) - Byte offset of chunk within blob (before compression)
@@ -112,32 +135,46 @@ Maps chunks to the blobs that contain them.
 **Purpose:** Enables chunk retrieval from blobs during restore operations.
 
 ### 7. `snapshots`
+
 Tracks backup snapshots.
 
 **Columns:**
-- `id` (TEXT PRIMARY KEY) - Snapshot ID (format: hostname-YYYYMMDD-HHMMSSZ)
+
+- `id` (TEXT PRIMARY KEY) - Snapshot ID (format: `hostname_name_timestamp`, e.g.
+  `server1_home_2025-06-01T12:00:00Z`: the hostname up to its first `.`, the
+  snapshot name, and an RFC 3339 UTC timestamp)
 - `hostname` (TEXT) - Hostname where backup was created
 - `vaultik_version` (TEXT) - Version of Vaultik used
 - `vaultik_git_revision` (TEXT) - Git revision of Vaultik used
 - `started_at` (INTEGER) - Start timestamp
 - `completed_at` (INTEGER) - Completion timestamp (NULL if in progress)
 - `file_count` (INTEGER) - Number of files in snapshot
-- `chunk_count` (INTEGER) - Number of unique chunks
-- `blob_count` (INTEGER) - Number of blobs referenced
+- `chunk_count` (INTEGER) - Number of chunks this snapshot stored that were not
+  stored before
+- `new_chunk_size` (INTEGER) - Total size of those chunks before compression;
+  `snapshot list` shows it as NEW CHUNK SIZE
+- `blob_count` (INTEGER) - Number of blobs this snapshot created
 - `total_size` (INTEGER) - Total size of all files
-- `blob_size` (INTEGER) - Total size of all blobs (compressed)
-- `blob_uncompressed_size` (INTEGER) - Total uncompressed size of all referenced blobs
+- `uncompressed_chunk_size` (INTEGER) - Total size before compression of the
+  distinct chunks the snapshot's files reference, recorded when its backup ran;
+  `snapshot list` shows it as UNCOMPRESSED SIZE
+- `blob_size` (INTEGER) - Total compressed size of all referenced blobs
+- `blob_uncompressed_size` (INTEGER) - Total uncompressed size of all referenced
+  blobs
 - `compression_ratio` (REAL) - Compression ratio achieved
 - `compression_level` (INTEGER) - Compression level used for this snapshot
 - `upload_bytes` (INTEGER) - Total bytes uploaded during this snapshot
 - `upload_duration_ms` (INTEGER) - Total milliseconds spent uploading to S3
 
-**Purpose:** Provides snapshot metadata and statistics including version tracking for compatibility.
+**Purpose:** Provides snapshot metadata and statistics including version
+tracking for compatibility.
 
 ### 8. `snapshot_files`
+
 Maps snapshots to the files they contain.
 
 **Columns:**
+
 - `snapshot_id` (TEXT) - Snapshot ID (FK to snapshots.id)
 - `file_id` (TEXT) - File ID (FK to files.id)
 - PRIMARY KEY (`snapshot_id`, `file_id`)
@@ -145,99 +182,147 @@ Maps snapshots to the files they contain.
 **Purpose:** Records which files are included in each snapshot.
 
 ### 9. `snapshot_blobs`
+
 Maps snapshots to the blobs they reference.
 
 **Columns:**
+
 - `snapshot_id` (TEXT) - Snapshot ID (FK to snapshots.id)
 - `blob_id` (TEXT) - Blob ID (FK to blobs.id)
 - `blob_hash` (TEXT) - Denormalized blob hash for manifest generation
 - PRIMARY KEY (`snapshot_id`, `blob_id`)
 
-**Purpose:** Tracks blob dependencies for snapshots and enables manifest generation.
+**Purpose:** Tracks blob dependencies for snapshots and enables manifest
+generation.
+
+### 10. `snapshot_mount_points`
+
+The directories at or below a configured path, the configured path itself
+included, that a snapshot's backup found an entry in and, once it had listed
+them, found on a different filesystem from their parent directory (a different
+device number), as a mount point is. An automount point whose volume cannot be
+mounted is not recorded, because the backup finds no entry in it.
+
+**Columns:**
+
+- `snapshot_id` (TEXT) - Snapshot ID (FK to snapshots.id)
+- `path` (TEXT) - Absolute path of the directory
+- PRIMARY KEY (`snapshot_id`, `path`)
+
+**Purpose:** Lets `snapshot create --prune` skip the prune when a backup does
+not find a mount point that an earlier snapshot of the same name found, as when
+the volume is not mounted.
+
+### 11. `uploads`
 
-### 10. `uploads`
 Tracks blob upload metrics.
 
 **Columns:**
+
 - `blob_hash` (TEXT PRIMARY KEY) - Hash of uploaded blob
-- `snapshot_id` (TEXT NOT NULL) - The snapshot that triggered this upload (FK to snapshots.id)
+- `snapshot_id` (TEXT NOT NULL) - The snapshot that triggered this upload (FK to
+  snapshots.id)
 - `uploaded_at` (INTEGER) - Upload timestamp
 - `size` (INTEGER) - Size of uploaded blob
 - `duration_ms` (INTEGER) - Upload duration in milliseconds
 
-**Purpose:** Performance monitoring and tracking which blobs were newly created (uploaded) during each snapshot.
+**Purpose:** Performance monitoring and tracking which blobs were newly created
+(uploaded) during each snapshot.
 
 ## Data Flow and Operations
 
 ### 1. Backup Process
 
 1. **File Scanning**
-   - `INSERT OR REPLACE INTO files` - Update file metadata
-   - `SELECT * FROM files WHERE path = ?` - Check if file has changed
-   - `INSERT INTO snapshot_files` - Add file to current snapshot
+    - `INSERT OR REPLACE INTO files` - Update file metadata
+    - `SELECT * FROM files WHERE path = ?` - Check if file has changed
+    - `INSERT INTO snapshot_files` - Add file to current snapshot
+    - `INSERT OR IGNORE INTO snapshot_mount_points` - Record each directory on a
+      different filesystem from its parent that the walk found an entry in
 
 2. **Chunking** (for changed files)
-   - `INSERT OR IGNORE INTO chunks` - Store new chunks
-   - `INSERT INTO file_chunks` - Map chunks to file
-   - `INSERT INTO chunk_files` - Create reverse mapping
+    - `INSERT OR IGNORE INTO chunks` - Store new chunks
+    - `INSERT INTO file_chunks` - Map chunks to file
+    - `INSERT INTO chunk_files` - Create reverse mapping
 
 3. **Blob Packing**
-   - `INSERT INTO blobs` - Create blob record with UUID (blob_hash NULL)
-   - `INSERT INTO blob_chunks` - Associate chunks with blob immediately
-   - `UPDATE blobs SET blob_hash = ?, finished_ts = ?` - Finalize blob after packing
+    - `INSERT INTO blobs` - Create blob record with UUID (blob_hash NULL)
+    - `INSERT INTO blob_chunks` - Associate chunks with blob immediately
+    - `UPDATE blobs SET blob_hash = ?, finished_ts = ?` - Finalize blob after
+      packing
 
 4. **Upload**
-   - `UPDATE blobs SET uploaded_ts = ?` - Mark blob as uploaded
-   - `INSERT INTO uploads` - Record upload metrics with snapshot_id
-   - `INSERT INTO snapshot_blobs` - Associate blob with snapshot
+    - `UPDATE blobs SET uploaded_ts = ?, compressed_size = ?` - Mark blob as
+      uploaded, with the size of the object in storage (the upload is skipped
+      when the blob is already stored)
+    - `INSERT INTO uploads` - Record upload metrics with snapshot_id
+    - `INSERT INTO snapshot_blobs` - Associate blob with snapshot
 
 5. **Snapshot Completion**
-   - `UPDATE snapshots SET completed_at = ?, stats...` - Finalize snapshot
-   - Generate and upload blob manifest from `snapshot_blobs`
+    - `UPDATE snapshots SET completed_at = ?, stats...` - Finalize snapshot
+    - Generate and upload blob manifest from `snapshot_blobs`
 
 ### 2. Incremental Backup
 
 1. **Change Detection**
-   - `SELECT * FROM files WHERE path = ?` - Get previous file metadata
-   - Compare mtime, size, mode to detect changes
-   - Skip unchanged files but still add to `snapshot_files`
+    - `SELECT * FROM files WHERE path = ?` - Get previous file metadata
+    - Compare mtime, size, mode to detect changes
+    - Re-chunk a file that lists a chunk no uploaded blob holds, even when its
+      metadata is unchanged
+    - Skip unchanged files but still add to `snapshot_files`
 
 2. **Chunk Reuse**
-   - `SELECT * FROM blob_chunks WHERE chunk_hash = ?` - Find existing chunks
-   - `INSERT INTO snapshot_blobs` - Reference existing blobs for unchanged files
+    - `SELECT * FROM blob_chunks WHERE chunk_hash = ?` - Find existing chunks
+    - `INSERT INTO snapshot_blobs` - Reference existing blobs for unchanged
+      files
 
 ### 3. Snapshot Metadata Export
 
 After a snapshot is completed:
+
 1. Copy database to temporary file
 2. Clean temporary database to contain only current snapshot data
 3. VACUUM the trimmed database so deleted rows leave no pages behind
 4. Compress with zstd and encrypt with age
 5. Upload to S3 as `metadata/{remote-key}/db.zst.age`
-6. Generate blob manifest and upload as `metadata/{remote-key}/manifest.json.zst`
+6. Generate blob manifest and upload as
+   `metadata/{remote-key}/manifest.json.zst`
 
-The `{remote-key}` directory name is a one-way hash of the human snapshot ID, so the ID is never written to the store in plaintext. The hash uses no secret, so a guessed hostname and snapshot name can still be confirmed against a listing; see [REPOSTRUCTURE.md](REPOSTRUCTURE.md#remote-key-derivation) and its [Accepted Risks](REPOSTRUCTURE.md#accepted-risks).
+The `{remote-key}` directory name is a one-way hash of the human snapshot ID, so
+the ID is never written to the store in plaintext. The hash uses no secret, so a
+guessed hostname and snapshot name can still be confirmed against a listing; see
+[REPOSTRUCTURE.md](REPOSTRUCTURE.md#remote-key-derivation) and its
+[Accepted Risks](REPOSTRUCTURE.md#accepted-risks).
 
 ### 4. Restore Process
 
 The restore process doesn't use the local database. Instead:
-1. Downloads snapshot metadata from S3
-2. Downloads required blobs based on manifest
+
+1. Downloads and decrypts the snapshot's metadata database (`db.zst.age`) from
+   S3
+2. Downloads the blobs holding the chunks of the files being restored, found
+   through that database's `blob_chunks` table; the manifest is not read
 3. Reconstructs files from decrypted and decompressed chunks
 
 ### 5. Pruning
 
 1. **Identify Unreferenced Blobs**
-   - Query blobs not referenced by any remaining snapshot
-   - Delete from S3 and local database
+    - Query blobs not referenced by any remaining snapshot
+    - Delete from S3, once older than `prune_grace_period`, and from the local
+      database
+2. **Delete Metadata Directories With No Manifest**
+    - Delete each directory under `metadata/` that has no `manifest.json.zst`
+      and no local snapshot record, once every object in it is older than
+      `prune_grace_period`
 
 ### 6. Incomplete Snapshot Cleanup
 
 Before each backup:
+
 1. Query incomplete snapshots (where `completed_at IS NULL`)
-2. Check if metadata exists in S3
-3. If no metadata, delete snapshot and all associations
-4. Clean up orphaned files, chunks, and blobs
+2. Delete each one and all its associations, without checking S3 for its
+   metadata
+3. Clean up orphaned files, chunks, and blobs
 
 ## Repository Pattern
 
@@ -246,18 +331,20 @@ Vaultik uses a repository pattern for database access:
 - `FileRepository` - CRUD operations for files and file metadata
 - `ChunkRepository` - CRUD operations for content chunks
 - `FileChunkRepository` - Manage file-to-chunk mappings
-- `ChunkFileRepository` - Manage chunk-to-file reverse mappings  
+- `ChunkFileRepository` - Manage chunk-to-file reverse mappings
 - `BlobRepository` - Manage blob lifecycle (creation, finalization, upload)
 - `BlobChunkRepository` - Manage blob-to-chunk associations
 - `SnapshotRepository` - Manage snapshots and their relationships
 - `UploadRepository` - Track blob upload metrics
 
 Each repository provides methods like:
+
 - `Create()` - Insert new record
 - `GetByID()` / `GetByPath()` / `GetByHash()` - Retrieve records
 - `Update()` - Update existing records
 - `Delete()` - Remove records
-- Specialized queries for each entity type (e.g., `DeleteOrphaned()`, `GetIncompleteByHostname()`)
+- Specialized queries for each entity type (e.g., `DeleteOrphaned()`,
+  `GetIncompleteByHostname()`)
 
 ## Transaction Management
 
@@ -270,25 +357,33 @@ err := repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
 ```
 
 This ensures consistency, especially important for operations like:
+
 - Creating file-chunk mappings
 - Associating chunks with blobs
 - Updating snapshot statistics
 
 ## Performance Considerations
 
-1. **Indexes**: 
-   - Primary keys are automatically indexed
-   - `idx_files_path` on `files(path)` for efficient file lookups
+1. **Indexes**:
+    - Primary keys are automatically indexed
+    - `idx_files_path` on `files(path)` for efficient file lookups
 
-2. **Prepared Statements**: All queries use prepared statements for performance and security
+2. **Prepared Statements**: All queries use prepared statements for performance
+   and security
 
-3. **Batch Operations**: Where possible, operations are batched within transactions
+3. **Batch Operations**: Where possible, operations are batched within
+   transactions
 
-4. **Write-Ahead Logging**: SQLite WAL mode is enabled for better concurrency
+4. **Write-Ahead Logging**: The local index runs in SQLite WAL mode with a
+   10-second busy timeout, so a read-only command such as `snapshot list` can
+   read it while a backup writes to it. Committed rows can sit in the `-wal`
+   file beside the index until a checkpoint, so the metadata export copies the
+   index through SQLite (`VACUUM INTO`), not as a file
 
 ## Data Integrity
 
-1. **Foreign Keys**: Enforced through CASCADE DELETE and application-level repository methods
+1. **Foreign Keys**: Enforced through CASCADE DELETE and application-level
+   repository methods
 2. **Unique Constraints**: Chunk hashes, file paths, and blob hashes are unique
 3. **Null Handling**: Nullable fields clearly indicate in-progress operations
-4. **Timestamp Tracking**: All major operations record timestamps for auditing
+4. **Timestamp Tracking**: All major operations record timestamps for auditing

docs/REPOSTRUCTURE.md

@@ -1,10 +1,13 @@
 # Vaultik S3 Repository Structure
 
-This document describes the structure and organization of data stored in the S3 bucket by Vaultik.
+This document describes the structure and organization of data stored in the S3
+bucket by Vaultik.
 
 ## Overview
 
-Vaultik stores all backup data in an S3-compatible object store. The repository consists of two main components:
+Vaultik stores all backup data in an S3-compatible object store. The repository
+consists of two main components, plus a marker file:
+
 1. **Blobs** - The actual backup data (content-addressed, encrypted)
 2. **Metadata** - Snapshot information and manifests (partially encrypted)
 
@@ -12,6 +15,7 @@ Vaultik stores all backup data in an S3-compatible object store. The repository
 
 ```
 <bucket>/<prefix>/
+├── vaultik-destination.txt
 ├── blobs/
 │   └── <hash[0:2]>/
 │       └── <hash[2:4]>/
@@ -22,114 +26,208 @@ Vaultik stores all backup data in an S3-compatible object store. The repository
         └── manifest.json.zst
 ```
 
-The metadata subdirectory is named with the **remote key**, a one-way hash of the snapshot ID, not with the human-readable snapshot ID itself. See [Remote Key Derivation](#remote-key-derivation).
+The metadata subdirectory is named with the **remote key**, a one-way hash of
+the snapshot ID, not with the human-readable snapshot ID itself. See
+[Remote Key Derivation](#remote-key-derivation).
+
+## Marker File (`vaultik-destination.txt`)
+
+Every backup writes this file at the root before it uploads anything. It holds
+one fixed line of plain text and names no host, path or time. While the local
+index holds snapshots stored there, a destination that answers a look-up of the
+marker with not found is treated as not present: `snapshot create` refuses to
+write to it, and `prune`, `snapshot purge`, `snapshot remove`, `snapshot list`
+and `remote nuke` behave as for a destination that cannot be listed. An
+unplugged volume leaves an empty directory at its mount point, and without the
+marker that directory would read as a destination whose snapshots are all gone.
+`remote nuke` deletes the marker last.
+
+When the look-up of the marker fails for any reason other than not found, the
+check is skipped and the command goes on. S3 refuses it on every run to
+credentials that allow only PUT, so a host with such credentials does not notice
+a destination emptied since its last backup, and its next backup would leave out
+the data its local index records as already stored. After the destination is
+emptied, run `vaultik database delete` on such a host before it backs up again.
 
 ## Blobs Directory (`blobs/`)
 
 ### Structure
+
 - **Path format**: `blobs/<first-2-chars>/<next-2-chars>/<full-hash>`
-- **Example**: `blobs/ca/fe/cafebabe1234567890abcdef1234567890abcdef1234567890abcdef12345678`
-- **Sharding**: The two-level directory structure (using the first 4 characters of the hash) prevents any single directory from containing too many objects
+- **Example**:
+  `blobs/ca/fe/cafebabe1234567890abcdef1234567890abcdef1234567890abcdef12345678`
+- **Sharding**: The two-level directory structure (using the first 4 characters
+  of the hash) prevents any single directory from containing too many objects
 
 ### Content
+
 - **What it contains**: Packed collections of content-defined chunks from files
 - **Format**: Zstandard compressed, then Age encrypted
 - **Encryption**: Always encrypted with Age using the configured recipients
-- **Naming**: Content-addressed. The blob's name is `hex(SHA256(SHA256(uncompressed blob contents)))` — the double SHA-256 of the concatenated chunk data, computed before compression and encryption, not a hash of the stored (compressed, encrypted) bytes. One consequence: only a holder of the age private key can check a stored blob's integrity, because matching a blob to its name means decrypting and decompressing it first — which is what `restore` and `verify --deep` do. Implemented in `internal/blobgen` (`DoubleSHA256`). This is the canonical description of blob naming; other documents and comments point here.
+- **Naming**: Content-addressed. The blob's name is
+  `hex(SHA256(SHA256(uncompressed blob contents)))` — the double SHA-256 of the
+  concatenated chunk data, computed before compression and encryption, not a
+  hash of the stored (compressed, encrypted) bytes. One consequence: only a
+  holder of the age private key can check a stored blob's integrity, because
+  matching a blob to its name means decrypting and decompressing it first —
+  which is what `restore` and `verify --deep` do. Implemented in
+  `internal/blobgen` (`DoubleSHA256`). This is the canonical description of blob
+  naming; other documents and comments point here.
 
 ### Why Encrypted
-Blobs contain the actual file data from backups and must be encrypted for security. The content-addressing ensures deduplication while the encryption ensures privacy.
+
+Blobs contain the actual file data from backups and must be encrypted for
+security. The content-addressing ensures deduplication while the encryption
+ensures privacy.
 
 ## Metadata Directory (`metadata/`)
 
-Each snapshot has its own subdirectory. The directory is **not** named with the human-readable snapshot ID; it is named with the remote key — a one-way hash of that ID. The human ID is never written to the destination store as a directory name (see [Remote Key Derivation](#remote-key-derivation)).
+Each snapshot has its own subdirectory. The directory is **not** named with the
+human-readable snapshot ID; it is named with the remote key — a one-way hash of
+that ID. The human ID is never written to the destination store as a directory
+name (see [Remote Key Derivation](#remote-key-derivation)).
 
 ### Snapshot ID Format
 
-The human-readable snapshot ID is used in CLI arguments, log lines, and the local database. It is not written to the destination store.
-- **Format**: `<hostname>_<snapshot-name>_<RFC3339>` (or `<hostname>_<RFC3339>` if no
-  name was specified)
+The human-readable snapshot ID is used in CLI arguments, log lines, and the
+local database. It is not written to the destination store.
+
+- **Format**: `<hostname>_<snapshot-name>_<RFC3339>` (or `<hostname>_<RFC3339>`
+  if no name was specified)
 - **Example**: `laptop_home_2024-01-15T14:30:52Z`
 - **Components**:
-  - Short hostname (everything before the first dot is stripped from the FQDN)
-  - Snapshot name from the configured `snapshots:` map (optional)
-  - RFC3339 UTC timestamp
+    - Short hostname (everything before the first dot is stripped from the FQDN)
+    - Snapshot name from the configured `snapshots:` map (optional)
+    - RFC3339 UTC timestamp
 
-This ID reveals the hostname, the configured snapshot name, and the backup time, so it is never used as the on-disk directory name — the remote key is used instead.
+This ID reveals the hostname, the configured snapshot name, and the backup time,
+so it is never used as the on-disk directory name — the remote key is used
+instead.
 
 ### Remote Key Derivation
 
-The remote key is `hex(SHA256(SHA256("vaultik|" + snapshot-id)))`: a double SHA-256 over the snapshot ID, with a `vaultik|` domain-separation prefix. The result is a 64-character hex string. The hash is not reversible, but it uses no secret: an observer who guesses a candidate hostname and snapshot name can hash it the same way and confirm whether that snapshot is present. The remote key keeps names out of a plain listing; it does not hide them from a guess. Implemented in `internal/snapshot/remotekey.go`.
+The remote key is `hex(SHA256(SHA256("vaultik|" + snapshot-id)))`: a double
+SHA-256 over the snapshot ID, with a `vaultik|` domain-separation prefix. The
+result is a 64-character hex string. The hash is not reversible, but it uses no
+secret: an observer who guesses a candidate hostname and snapshot name can hash
+it the same way and confirm whether that snapshot is present. The remote key
+keeps names out of a plain listing; it does not hide them from a guess.
+Implemented in `internal/snapshot/remotekey.go`.
 
 Worked example:
+
 - Snapshot ID: `server1_home_2025-06-01T12:00:00Z`
 - Remote key: `17f97bcde958748af076b926af59823943db59e80ce7170b40f124dfa28f64aa`
-- Directory: `metadata/17f97bcde958748af076b926af59823943db59e80ce7170b40f124dfa28f64aa/`
+- Directory:
+  `metadata/17f97bcde958748af076b926af59823943db59e80ce7170b40f124dfa28f64aa/`
 
-A plain listing of the destination store therefore shows only these hashes, not the hostname or snapshot name of any backup — but because the hash uses no secret, a guessed hostname and snapshot name can be hashed and confirmed against the listing. The same remote key is stored in the manifest's `snapshot_id` field.
+A plain listing of the destination store therefore shows only these hashes, not
+the hostname or snapshot name of any backup — but because the hash uses no
+secret, a guessed hostname and snapshot name can be hashed and confirmed against
+the listing. The same remote key is stored in the manifest's `snapshot_id`
+field.
 
 ### Files in Each Snapshot Directory
 
 #### `db.zst.age` - Encrypted Database
+
 - **What it contains**: Pruned binary SQLite database for this snapshot
 - **Format**: Binary SQLite → Zstandard compressed → Age encrypted
 - **Encryption**: Encrypted with Age
-- **Purpose**: Contains full file metadata, chunk mappings, and all relationships
-- **Why encrypted**: Contains sensitive metadata like file paths, permissions, and ownership
+- **Purpose**: Contains full file metadata, chunk mappings, and all
+  relationships
+- **Why encrypted**: Contains sensitive metadata like file paths, permissions,
+  and ownership
 
 #### `manifest.json.zst` - Unencrypted Blob Manifest
+
 - **What it contains**: JSON list of all blob hashes referenced by this snapshot
 - **Format**: JSON → Zstandard compressed (NOT encrypted)
 - **Encryption**: NOT encrypted
 - **Purpose**: Enables pruning operations without requiring decryption keys
 - **Structure**:
+
 ```json
 {
-  "snapshot_id": "17f97bcde958748af076b926af59823943db59e80ce7170b40f124dfa28f64aa",
-  "timestamp": "2025-06-01T12:00:00Z",
-  "blob_count": 42,
-  "total_compressed_size": 1048576,
-  "blobs": [
-    { "hash": "cafebabe1234567890abcdef1234567890abcdef1234567890abcdef12345678", "compressed_size": 24576 },
-    { "hash": "deadbeef1234567890abcdef1234567890abcdef1234567890abcdef12345678", "compressed_size": 32768 }
-  ]
+    "snapshot_id": "17f97bcde958748af076b926af59823943db59e80ce7170b40f124dfa28f64aa",
+    "timestamp": "2025-06-01T12:00:00Z",
+    "blob_count": 42,
+    "total_compressed_size": 1048576,
+    "blobs": [
+        {
+            "hash": "cafebabe1234567890abcdef1234567890abcdef1234567890abcdef12345678",
+            "compressed_size": 24576
+        },
+        {
+            "hash": "deadbeef1234567890abcdef1234567890abcdef1234567890abcdef12345678",
+            "compressed_size": 32768
+        }
+    ]
 }
 ```
-`snapshot_id` is the remote key (a hash), not the human ID; `timestamp` is written in the clear.
+
+`snapshot_id` is the remote key (a hash), not the human ID; `timestamp` is
+written in the clear.
 
 ### Why Manifest is Unencrypted
+
 The manifest must be readable without the private key to enable:
+
 1. **Pruning operations** - Identifying unreferenced blobs for deletion
 2. **Storage analysis** - Understanding space usage without decryption
 3. **Verification** - Checking blob existence without decryption
-4. **Cross-snapshot deduplication analysis** - Finding shared blobs between snapshots
+4. **Cross-snapshot deduplication analysis** - Finding shared blobs between
+   snapshots
 
-The manifest contains the remote key, the backup timestamp, the blob count and total compressed size, and each blob's hash and compressed size. It contains no file names, paths, or other decrypted metadata.
+The manifest contains the remote key, the backup timestamp, the blob count and
+total compressed size, and each blob's hash and compressed size. It contains no
+file names, paths, or other decrypted metadata.
 
 ## Security Considerations
 
 ### What's Encrypted
+
 - **All file content** (in blobs)
-- **All file metadata** (paths, permissions, timestamps, ownership in db.zst.age)
+- **All file metadata** (paths, permissions, timestamps, ownership in
+  db.zst.age)
 - **File-to-chunk mappings** (in db.zst.age)
 
 ### What's Not Encrypted
-- **The remote key** — directory names and the manifest `snapshot_id`, a one-way hash of the snapshot ID (see [Remote Key Derivation](#remote-key-derivation))
+
+- **The remote key** — directory names and the manifest `snapshot_id`, a one-way
+  hash of the snapshot ID (see [Remote Key Derivation](#remote-key-derivation))
 - **The backup timestamp** (in manifest.json.zst)
 - **Blob hashes and their compressed sizes** (in manifest.json.zst)
 - **Blob count and total compressed size per snapshot** (in manifest.json.zst)
 
 ### Privacy Implications
-From the unencrypted data, an observer of the destination store can determine:
-- **When each backup was taken** — not from the directory name, which is a one-way hash, but from the plaintext `timestamp` field in manifest.json.zst, which is published in the clear
-- How many blobs each snapshot references, and the total compressed size
-- The compressed size of each blob, and which blobs are shared between snapshots (deduplication patterns)
-- **Whether a guessed hostname and snapshot name are present** — the remote key is an unkeyed hash, so an observer holding candidate names can hash each one and match it against the directory listing. The human ID is never published, so it cannot be read off directly, but it can be confirmed by guessing.
 
-Together these give an observer a timing-and-size profile of every snapshot. This is an accepted, documented property of the format, not a defect: the manifest is unencrypted so that pruning can run without the private key, and the timing channel could not be closed by encrypting it anyway — object creation times and per-object sizes stay visible at the storage layer on both `s3://` and `file://` destinations regardless.
+From the unencrypted data, an observer of the destination store can determine:
+
+- **When each backup was taken** — not from the directory name, which is a
+  one-way hash, but from the plaintext `timestamp` field in manifest.json.zst,
+  which is published in the clear
+- How many blobs each snapshot references, and the total compressed size
+- The compressed size of each blob, and which blobs are shared between snapshots
+  (deduplication patterns)
+- **Whether a guessed hostname and snapshot name are present** — the remote key
+  is an unkeyed hash, so an observer holding candidate names can hash each one
+  and match it against the directory listing. The human ID is never published,
+  so it cannot be read off directly, but it can be confirmed by guessing.
+
+Together these give an observer a timing-and-size profile of every snapshot.
+This is an accepted, documented property of the format, not a defect: the
+manifest is unencrypted so that pruning can run without the private key, and the
+timing channel could not be closed by encrypting it anyway — object creation
+times and per-object sizes stay visible at the storage layer on both `s3://` and
+`file://` destinations regardless.
 
 An observer cannot determine:
-- The hostname or snapshot name of any backup by reading it off the store — the directory name and the manifest `snapshot_id` are unkeyed hashes of the human ID, so the text is never published (though a guessed name can be confirmed, as above)
+
+- The hostname or snapshot name of any backup by reading it off the store — the
+  directory name and the manifest `snapshot_id` are unkeyed hashes of the human
+  ID, so the text is never published (though a guessed name can be confirmed, as
+  above)
 - File names or paths
 - File contents
 - File permissions or ownership
@@ -138,40 +236,119 @@ An observer cannot determine:
 
 ### Accepted Risks
 
-These are known, deliberate properties of the format and the tooling, recorded so an operator can weigh them rather than discover them.
+These are known, deliberate properties of the format and the tooling, recorded
+so an operator can weigh them rather than discover them.
 
-1. **No proof of authorship.** Restore and `verify --deep` prove that data decrypts with the age private key and matches its unkeyed content hashes. They do not prove who wrote it: anyone who knows a recipient public key and can replace objects on the destination can substitute a snapshot they built. The recipient string is not stored at the destination, but a compromised backed-up host has it. Defences live on the destination side — bucket versioning or object lock, credentials for the source host that cannot delete or overwrite existing versions, and pruning from a trusted host. Note that S3 `PutObject` overwrites an existing key, so PUT permission alone is not append-only.
-2. **Compression reveals sizes.** Blobs and `db.zst.age` are zstd-compressed then age-encrypted; the manifest is compressed only. age does not pad, so an object's size is the exact compressed length of its contents. All new chunks packed into one blob share a single zstd stream (8 MiB window, 4 MiB at compression levels 1-2), and a blob is closed at `blob_size_limit` and at the end of each configured path. Because a stored chunk is never packed again, someone who can write into a backed-up file and watch blob sizes learns something only when their controlled data and a secret land in the same chunk of a file that keeps changing. Advice: back up any outsider-writable directory as its own snapshot.
-3. **Chunking uses no secret.** The FastCDC parameters are fixed and public. The default 10 MB average yields chunks between 2.5 MB and 40 MB, and any file of 2.5 MB or less is a single chunk. At the default 10 GB `blob_size_limit` a blob holds hundreds of chunks, so individual chunk lengths are not visible in the blob's size; lowering the limit toward the chunk size begins to expose them.
-4. **Decrypted data on local disk.** Several commands stage plaintext under `$TMPDIR`: `snapshot restore` writes decrypted blobs under `vaultik-blobcache-*/` (no size cap) and the decrypted metadata database at `vaultik-restore-*/snapshot.db`; `verify --deep` writes that database at `vaultik-verify-*/snapshot.db`; `snapshot create` keeps a plaintext copy of the index at `vaultik-snapshot-*/snapshot.db`. These files are created `0600` and removed on success, but a `kill -9` or a power loss leaves them behind — delete any leftover `vaultik-*` directory under `$TMPDIR` by hand. `$TMPDIR` should be trusted to the same degree as the restore target.
-5. **Store permissions differ per command.** The backed-up host needs only PUT to run `snapshot create`: it writes blobs and metadata and neither reads nor deletes them. Other commands need more — `snapshot verify`, `snapshot restore`, and `prune` list and read; `prune`, `snapshot purge`, `snapshot remove`, and `remote nuke` also delete. The recommended cron line uses `--prune`, which runs `prune` on the backed-up host, so granting that host `--prune` gives it credentials that can delete its own backups. To keep the source host to PUT only, prune from a separate trusted host instead.
-6. **Changing recipients does not re-encrypt existing data.** Deduplicated chunks and same-named blobs already on the destination stay encrypted to the recipients in force when they were written. A new snapshot that reuses them cannot be restored with a newly added recipient's key alone, because those reused objects were never encrypted to it. To make everything readable by a new key, run `vaultik database delete` and take a full backup to a fresh destination or prefix.
-7. **X25519 recipients only.** vaultik rejects age ssh and plugin recipients. Long-lived ciphertext held by a third party (the destination operator) has no fallback if X25519 is ever broken: there is no second recipient type and no post-quantum option.
+1. **No proof of authorship.** Restore and `verify --deep` prove that data
+   decrypts with the age private key and matches its unkeyed content hashes.
+   They do not prove who wrote it: anyone who knows a recipient public key and
+   can replace objects on the destination can substitute a snapshot they built.
+   The recipient string is not stored at the destination, but a compromised
+   backed-up host has it. Defences live on the destination side — bucket
+   versioning or object lock, credentials for the source host that cannot delete
+   or overwrite existing versions, and pruning from a trusted host. Note that S3
+   `PutObject` overwrites an existing key, so PUT permission alone is not
+   append-only.
+2. **Compression reveals sizes.** Blobs and `db.zst.age` are zstd-compressed
+   then age-encrypted; the manifest is compressed only. age does not pad, so an
+   object's size is the exact compressed length of its contents. All new chunks
+   packed into one blob share a single zstd stream (8 MiB window, 4 MiB at
+   compression levels 1-2), and a blob is closed at `blob_size_limit` and at the
+   end of each configured path. Because a stored chunk is never packed again,
+   someone who can write into a backed-up file and watch blob sizes learns
+   something only when their controlled data and a secret land in the same chunk
+   of a file that keeps changing. Advice: back up any outsider-writable
+   directory as its own snapshot.
+3. **Chunking uses no secret.** The FastCDC parameters are fixed and public. The
+   default 10 MB average yields chunks between 2.5 MB and 40 MB, and any file of
+   2.5 MB or less is a single chunk. At the default 10 GB `blob_size_limit` a
+   blob holds hundreds of chunks, so individual chunk lengths are not visible in
+   the blob's size; lowering the limit toward the chunk size begins to expose
+   them.
+4. **Decrypted data on local disk.** Several commands stage plaintext under
+   `$TMPDIR`: `snapshot restore` writes decrypted blobs under
+   `vaultik-blobcache-*/` (no size cap) and the decrypted metadata database at
+   `vaultik-restore-*/snapshot.db`; `verify --deep` writes that database at
+   `vaultik-verify-*/snapshot.db`; `snapshot create` keeps a plaintext copy of
+   the index at `vaultik-snapshot-*/snapshot.db`. These files are created `0600`
+   and removed on success, but a `kill -9` or a power loss leaves them behind —
+   delete any leftover `vaultik-*` directory under `$TMPDIR` by hand. `$TMPDIR`
+   should be trusted to the same degree as the restore target.
+5. **Store permissions differ per command.** The backed-up host needs only PUT
+   to run `snapshot create`: it writes blobs and metadata and neither reads nor
+   deletes them. Other commands need more — `snapshot verify`,
+   `snapshot restore`, and `prune` list and read; `prune`, `snapshot purge`,
+   `snapshot remove`, and `remote nuke` also delete. The recommended cron line
+   uses `--prune`, which runs `prune` on the backed-up host, so granting that
+   host `--prune` gives it credentials that can delete its own backups. To keep
+   the source host to PUT only, prune from a separate trusted host instead.
+6. **Changing recipients does not re-encrypt existing data.** Deduplicated
+   chunks and same-named blobs already on the destination stay encrypted to the
+   recipients in force when they were written. A new snapshot that reuses them
+   cannot be restored with a newly added recipient's key alone, because those
+   reused objects were never encrypted to it. To make everything readable by a
+   new key, run `vaultik database delete` and take a full backup to a fresh
+   destination or prefix.
+7. **X25519 recipients only.** vaultik rejects age ssh and plugin recipients.
+   Long-lived ciphertext held by a third party (the destination operator) has no
+   fallback if X25519 is ever broken: there is no second recipient type and no
+   post-quantum option.
 
 ## Consistency Guarantees
 
 1. **Blobs are immutable** - Once written, a blob is never modified
-2. **Blobs are written before metadata** - A snapshot's metadata is only written after all its blobs are successfully uploaded
-3. **Metadata is written atomically** - Both db.zst.age and manifest.json.zst are written as complete files
-4. **A snapshot is marked complete in the local DB only after its metadata is uploaded** - `finalizeSnapshotMetadata` runs `ExportSnapshotMetadata` first and records completion (`MarkSnapshotComplete`) only once the export succeeds (see the backup data flow in [ARCHITECTURE.md](../ARCHITECTURE.md)). A crash during the export therefore leaves the snapshot incomplete, so the next backup's `PruneDatabase` drops it and re-backs-up its data, rather than leaving a completed-looking row in the local index with no matching metadata on the destination store. (A crash in the brief moment after the export succeeds but before completion is recorded leaves a fully-restorable snapshot on the destination that the local index drops as incomplete on the next run; `snapshot list` then reports it honestly as remote-only, which is the safe direction: the destination copy stays restorable.)
+2. **Blobs are written before metadata** - A snapshot's metadata is only written
+   after all its blobs are successfully uploaded
+3. **Metadata is written atomically** - Both db.zst.age and manifest.json.zst
+   are written as complete files
+4. **A snapshot is marked complete in the local DB only after its metadata is
+   uploaded** - `finalizeSnapshotMetadata` runs `ExportSnapshotMetadata` first
+   and records completion (`MarkSnapshotComplete`) only once the export succeeds
+   (see the backup data flow in [ARCHITECTURE.md](ARCHITECTURE.md)). A crash
+   during the export therefore leaves the snapshot incomplete, so the next
+   backup's `PruneDatabase` drops it and re-backs-up its data, rather than
+   leaving a completed-looking row in the local index with no matching metadata
+   on the destination store. (A crash in the brief moment after the export
+   succeeds but before completion is recorded leaves a fully-restorable snapshot
+   on the destination that the local index drops as incomplete on the next run;
+   `snapshot list` then reports it honestly as remote-only, which is the safe
+   direction: the destination copy stays restorable.)
 
 ## Pruning Safety
 
 The prune operation is safe because:
-1. It keeps every blob listed in any snapshot's manifest and deletes only blobs that no manifest references
+
+1. It keeps every blob listed in any snapshot's manifest and deletes only blobs
+   that no manifest references. Under `metadata/` it deletes only a directory
+   that has no `manifest.json.zst`, whose name is a remote key, and that no
+   local snapshot record claims
 2. Manifests are unencrypted and can be read without keys
-3. If any manifest cannot be downloaded or decoded, prune deletes nothing and exits with an error, rather than treating that snapshot's blobs as unreferenced
-4. Prune requires exclusive access to the destination: running it during a concurrent backup can race a snapshot whose manifest is not yet written, so do not prune while a backup is in progress
+3. If any manifest cannot be downloaded or decoded, prune deletes nothing and
+   exits with an error, rather than treating that snapshot's blobs as
+   unreferenced
+4. It deletes an unreferenced blob only once the blob is older than
+   `prune_grace_period` (default `24h`), and a directory under `metadata/` with
+   no manifest only once every object in it is. A backup uploads its blobs and
+   its `db.zst.age` before its manifest, so a prune running meanwhile, on this
+   host or another sharing the destination, leaves them as long as no backup run
+   takes longer than the grace period. The grace period does not protect a blob
+   a backup reuses from an earlier run that stopped before its manifest: the
+   backup skips the upload of a blob already on the destination, so that blob
+   keeps the time of its first upload, and a prune before the backup's manifest
+   is written can delete it; the new snapshot then cannot be restored.
+   `remote nuke` has no such check
 
 ## Restoration Requirements
 
 To restore from a backup, you need:
+
 1. **The Age private key** - To decrypt blobs and database
 2. **The snapshot metadata** - Both files from the snapshot's metadata directory
 3. **All referenced blobs** - As listed in the manifest
 
 The restoration process:
+
 1. Download and decrypt the database dump to understand file structure
 2. Download and decrypt the required blobs
 3. Reconstruct files from their chunks
-4. Restore file metadata (permissions, timestamps, etc.)
+4. Restore file metadata (permissions, timestamps, etc.)

docs/TODO.md

@@ -0,0 +1,22 @@
+# Workflow
+
+- branch (from `main`)
+- do the work in Next Step
+- delete Next Step from `TODO.md`
+- move the top item of Future Steps into Next Step
+- commit (`TODO.md` changes in the same commit as the work)
+- merge to `main` if the branch is not protected, otherwise open a PR
+- push
+
+# Status
+
+pre-1.0
+
+# Next Step
+
+Fix the open bugs on the issue tracker that block the release. Merging `next`
+into `main` is the owner's.
+
+# Future Steps
+
+Work planned after 1.0 is listed in the README [roadmap](../README.md#roadmap).

go.mod

@@ -20,9 +20,13 @@ require (
 	github.com/rclone/rclone v1.72.1
 	github.com/spf13/afero v1.15.0
 	github.com/spf13/cobra v1.10.1
+	github.com/spf13/pflag v1.0.10
 	github.com/stretchr/testify v1.11.1
+	go.uber.org/dig v1.19.0
 	go.uber.org/fx v1.24.0
+	golang.org/x/mod v0.29.0
 	golang.org/x/sync v0.18.0
+	golang.org/x/sys v0.38.0
 	golang.org/x/term v0.37.0
 	gopkg.in/yaml.v3 v3.0.1
 	modernc.org/sqlite v1.38.0
@@ -225,7 +229,6 @@ require (
 	github.com/smarty/assertions v1.16.0 // indirect
 	github.com/sony/gobreaker v1.0.0 // indirect
 	github.com/spacemonkeygo/monkit/v3 v3.0.25-0.20251022131615-eb24eb109368 // indirect
-	github.com/spf13/pflag v1.0.10 // indirect
 	github.com/t3rm1n4l/go-mega v0.0.0-20251031123324-a804aaa87491 // indirect
 	github.com/tidwall/gjson v1.18.0 // indirect
 	github.com/tidwall/match v1.1.1 // indirect
@@ -255,7 +258,6 @@ require (
 	go.opentelemetry.io/otel/metric v1.38.0 // indirect
 	go.opentelemetry.io/otel/trace v1.38.0 // indirect
 	go.shabbyrobe.org/gocovmerge v0.0.0-20230507111327-fa4f82cfbf4d // indirect
-	go.uber.org/dig v1.19.0 // indirect
 	go.uber.org/multierr v1.11.0 // indirect
 	go.uber.org/zap v1.27.0 // indirect
 	go.yaml.in/yaml/v2 v2.4.3 // indirect
@@ -263,7 +265,6 @@ require (
 	golang.org/x/exp v0.0.0-20251023183803-a4bb9ffd2546 // indirect
 	golang.org/x/net v0.47.0 // indirect
 	golang.org/x/oauth2 v0.33.0 // indirect
-	golang.org/x/sys v0.38.0 // indirect
 	golang.org/x/text v0.31.0 // indirect
 	golang.org/x/time v0.14.0 // indirect
 	golang.org/x/tools v0.38.0 // indirect

internal/blob/packer.go

@@ -245,6 +245,17 @@ func (p *Packer) Flush(ctx context.Context) error {
 	return nil
 }
 
+// DiscardCurrentBlob closes and removes the temp file of the blob being
+// packed, if any, and drops that blob. A caller that stops before Flush
+// must call it, or the temp file is left behind. Thread-safe.
+func (p *Packer) DiscardCurrentBlob() {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+
+	p.cleanupTempFile()
+	p.currentBlob = nil
+}
+
 // FinalizeBlob finalizes the current blob being assembled.
 // This compresses the accumulated chunks, encrypts the result, and computes
 // the content-addressed hash. The finalized blob is either passed to the

internal/blobgen/negative_test.go

@@ -2,8 +2,10 @@ package blobgen_test
 
 import (
 	"bytes"
+	"errors"
 	"fmt"
 	"testing"
+	"testing/iotest"
 
 	"filippo.io/age"
 	"github.com/stretchr/testify/assert"
@@ -120,6 +122,38 @@ func TestNewReaderTrailingAndGarbage(t *testing.T) {
 	})
 }
 
+// TestNewReaderGarbageErrorQuotesNoBytes checks that lines of text in
+// place of a blob fail with ErrNotAgeStream, and that the error carries
+// none of their bytes. The lines end in a newline because age's own error
+// quotes the data's first line only when it reads a whole line.
+func TestNewReaderGarbageErrorQuotesNoBytes(t *testing.T) {
+	t.Parallel()
+
+	id, _ := makeIdentity(t)
+	garbage := bytes.Repeat([]byte("garbage-bytes-\n"), 100)
+
+	_, err := blobgen.NewReader(bytes.NewReader(garbage), id)
+
+	require.ErrorIs(t, err, blobgen.ErrNotAgeStream)
+	assert.NotContains(t, err.Error(), "garbage-bytes-")
+}
+
+// errConnectionReset stands for a storage read that fails.
+var errConnectionReset = errors.New("connection reset")
+
+// TestNewReaderReadFailure checks that a failed read of the age header is
+// returned as that read error, not as ErrNotAgeStream.
+func TestNewReaderReadFailure(t *testing.T) {
+	t.Parallel()
+
+	id, _ := makeIdentity(t)
+
+	_, err := blobgen.NewReader(iotest.ErrReader(errConnectionReset), id)
+
+	require.ErrorIs(t, err, errConnectionReset)
+	assert.NotErrorIs(t, err, blobgen.ErrNotAgeStream)
+}
+
 // TestNewWriterInvalidLevel covers the rejected end of issue case 9: an
 // out-of-range compression level errors and writes nothing to the destination.
 func TestNewWriterInvalidLevel(t *testing.T) {

internal/blobgen/reader.go

@@ -11,6 +11,11 @@ import (
 	"github.com/klauspost/compress/zstd"
 )
 
+// ErrNotAgeStream is returned by NewReader when the data cannot be opened
+// as an age stream, for example because its header is malformed. It
+// replaces age's own error, which can quote the data's first bytes.
+var ErrNotAgeStream = errors.New("not a valid age stream")
+
 // Reader wraps decompression and decryption with SHA256 verification
 type Reader struct {
 	reader       io.Reader
@@ -24,11 +29,25 @@ type Reader struct {
 // NewReader creates a new Reader that decrypts, decompresses, and verifies
 // data. Every supplied identity is offered to age.Decrypt, so a blob
 // encrypted to any one of them can be read.
+//
+// A failure to read r is returned as that error, and a blob encrypted to
+// none of the identities as age.NoIdentityMatchError. Any other failure
+// to open the age stream is ErrNotAgeStream.
 func NewReader(r io.Reader, identities ...age.Identity) (*Reader, error) {
-	// Create decryption reader
-	decReader, err := age.Decrypt(r, identities...)
+	src := &readErrorRecorder{reader: r}
+
+	decReader, err := age.Decrypt(src, identities...)
 	if err != nil {
-		return nil, fmt.Errorf("creating decryption reader: %w", err)
+		var noMatch *age.NoIdentityMatchError
+
+		switch {
+		case src.err != nil:
+			return nil, fmt.Errorf("reading age header: %w", src.err)
+		case errors.As(err, &noMatch):
+			return nil, fmt.Errorf("creating decryption reader: %w", err)
+		default:
+			return nil, ErrNotAgeStream
+		}
 	}
 
 	// Create decompression reader
@@ -52,6 +71,23 @@ func NewReader(r io.Reader, identities ...age.Identity) (*Reader, error) {
 	}, nil
 }
 
+// readErrorRecorder passes reads through to reader and keeps the first
+// error other than io.EOF, so NewReader can tell a failed read from bad
+// data.
+type readErrorRecorder struct {
+	reader io.Reader
+	err    error
+}
+
+func (r *readErrorRecorder) Read(p []byte) (int, error) {
+	n, err := r.reader.Read(p)
+	if err != nil && !errors.Is(err, io.EOF) && r.err == nil {
+		r.err = err
+	}
+
+	return n, err
+}
+
 // Read implements io.Reader
 func (r *Reader) Read(p []byte) (int, error) {
 	n, err := r.teeReader.Read(p)

internal/cli/app.go

@@ -7,13 +7,11 @@ import (
 	"context"
 	"errors"
 	"fmt"
-	"path/filepath"
-	"strings"
 	"sync"
 	"time"
 
-	"github.com/adrg/xdg"
 	"github.com/spf13/cobra"
+	"go.uber.org/dig"
 	"go.uber.org/fx"
 	"sneak.berlin/go/vaultik/internal/config"
 	"sneak.berlin/go/vaultik/internal/database"
@@ -90,7 +88,7 @@ func setupGlobals(
 // blank line. Used both from the fx hook (for subcommand invocations) and
 // from the root cobra Run handler (for `vaultik` with no subcommand).
 func writeStartupBanner(w *ui.Writer, startTime time.Time, shortCommit string) {
-	w.Bannerf("%s %s by %s (commit %s, built on %s) starting up at %s.",
+	w.Bannerf("%s %s by %s (commit %s, commit date %s) starting up at %s.",
 		globals.Appname, globals.Version, globals.Author,
 		shortCommit, globals.CommitDate,
 		startTime.Format(time.RFC3339))
@@ -146,16 +144,12 @@ func (e *startupError) Error() string {
 //	could not build arguments for function "Y" (file:line): failed to build U:
 //	received non-nil error from function "Z" (file:line): <real error>
 //
-// Users care about the real error, not the DI plumbing. We strip everything
-// up through the last "): " (which is always the close-paren of an fx
-// function-location annotation followed by the wrapped error).
+// Users care about the real error, not the DI plumbing. dig.RootCause,
+// from the library fx is built on, unwraps those layers and returns the
+// real error whole. Cutting the text at "): " instead would also cut a
+// real error that contains it, such as the blob_size_limit one.
 func cleanStartupError(err error) error {
-	msg := err.Error()
-	if idx := strings.LastIndex(msg, "): "); idx >= 0 {
-		msg = msg[idx+3:]
-	}
-
-	return &startupError{msg: msg}
+	return &startupError{msg: dig.RootCause(err).Error()}
 }
 
 // RunApp starts the fx application, blocks until it is asked to stop, and
@@ -200,12 +194,16 @@ func RunApp(ctx context.Context, app *fx.App) error {
 }
 
 // errReported marks a failure the operation has already shown the user
-// (and deliberately withheld under --json). Entry turns it into a
-// non-zero exit status without printing anything further, so the error
-// line is not doubled. It flows up from RunOperation through cobra to
-// Entry.
+// (or, under `snapshot verify --json`, put in its document). Entry
+// turns it into a non-zero exit status without printing anything
+// further, so the error line is not doubled. It flows up from
+// RunOperation through cobra to Entry.
 var errReported = errors.New("operation failed")
 
+// errInterrupted marks an operation that SIGINT or SIGTERM stopped
+// before it finished. Entry shows it and returns exitCodeInterrupted.
+var errInterrupted = errors.New("interrupted before the command finished")
+
 // RunOperation runs op against the Vaultik instance inside the fx app
 // and turns a failure into a returned error rather than an os.Exit from
 // within the goroutine. An os.Exit there skipped main's deferred
@@ -220,17 +218,21 @@ var errReported = errors.New("operation failed")
 // interrupt OnStop cancels op and waits for the goroutine to return, so
 // op's cleanup (removing decrypted scratch files) runs before the
 // process exits; the wait is bounded by shutdownTimeout. report is
-// called with a non-canceled failure so the caller can log it (and
-// suppress it under --json) before it becomes errReported. A context
-// cancellation is the interrupt path, not a failure: it is neither
-// reported nor counted as one.
+// called with a failure so the caller can show it to the user before
+// it becomes errReported.
+//
+// The run counts as interrupted unless op returned, without an
+// interrupt having cancelled it, before RunWithApp returned. An
+// interrupted op is not reported, whatever it returned; RunOperation
+// returns errInterrupted instead.
 func RunOperation(
 	ctx context.Context, opts AppOptions,
 	op func(v *vaultik.Vaultik) error, report func(err error),
 ) error {
 	var (
-		mu     sync.Mutex
-		failed bool
+		mu       sync.Mutex
+		finished bool // op returned before any interrupt cancelled it
+		failed   bool // op finished with an error
 	)
 
 	opts.Invokes = append(opts.Invokes,
@@ -241,11 +243,21 @@ func RunOperation(
 				OnStart: func(_ context.Context) error {
 					stop = v.StartOperation(func() {
 						err := op(v)
-						if err != nil && !errors.Is(err, context.Canceled) {
-							report(err)
+
+						// Only stop, called from OnStop below, cancels the
+						// Vaultik context, so a live context means no
+						// interrupt cancelled op. Check the context, not
+						// err: an interrupted op need not return
+						// context.Canceled (`snapshot verify --json`
+						// returns a verification failure).
+						if v.Context().Err() == nil {
+							if err != nil {
+								report(err)
+							}
 
 							mu.Lock()
-							failed = true
+							finished = true
+							failed = err != nil
 							mu.Unlock()
 						}
 
@@ -278,28 +290,34 @@ func RunOperation(
 		return err
 	}
 
-	// The goroutine sets failed before triggering the shutdown that lets
-	// RunWithApp return, so the write is in place by the time we read it.
+	// RunWithApp returns only after the app was asked to stop, either by
+	// an interrupt or by the goroutine's Shutdown call. When op finished
+	// without being cancelled, the goroutine set finished before that
+	// call. So if finished is unset here, an interrupt stopped the app,
+	// and op either returned after it was cancelled or is still running
+	// because the shutdown timed out.
 	mu.Lock()
 	defer mu.Unlock()
 
-	if failed {
+	switch {
+	case !finished:
+		return errInterrupted
+	case failed:
 		return errReported
+	default:
+		return nil
 	}
-
-	return nil
 }
 
 // runVaultikApp runs the standard single-operation command lifecycle
 // shared by the snapshot list/purge/remove and remote nuke subcommands:
 // resolve the config, then run op against the Vaultik instance through
-// RunOperation, reporting a failure prefixed with failMsg (suppressed
-// while suppressErrors is true, e.g. under --json). mode says whether the
-// command takes the PID lock. jsonOutput marks a command whose stdout is a
-// JSON document: it quiets the UI but, unlike Quiet, leaves the stderr log
-// level alone.
+// RunOperation, reporting a failure prefixed with failMsg on stderr. mode
+// says whether the command takes the PID lock. jsonOutput marks a command
+// whose stdout is a JSON document: it quiets the UI but, unlike Quiet,
+// leaves the stderr log level alone.
 func runVaultikApp(
-	cmd *cobra.Command, mode lockMode, jsonOutput, suppressErrors bool,
+	cmd *cobra.Command, mode lockMode, jsonOutput bool,
 	failMsg string, op func(v *vaultik.Vaultik) error,
 ) error {
 	configPath, err := ResolveConfigPath()
@@ -319,10 +337,6 @@ func runVaultikApp(
 		},
 		Mode: mode,
 	}, op, func(err error) {
-		if suppressErrors {
-			return
-		}
-
 		log.Error(failMsg, "error", err)
 		ReportErrorf("%s: %v", failMsg, err)
 	})
@@ -335,8 +349,7 @@ func runVaultikApp(
 // only one runs at a time; a read-only command runs without it and is not
 // blocked while a mutator holds the lock (opts.Mode).
 func RunWithApp(ctx context.Context, opts AppOptions) error {
-	release, err := acquireLockIfMutating(opts.Mode,
-		filepath.Join(xdg.DataHome, "vaultik"))
+	release, err := acquireLockIfMutating(opts.Mode, pidlock.Dir())
 	if err != nil {
 		return err
 	}

internal/cli/app_test.go

@@ -2,58 +2,65 @@ package cli //nolint:testpackage // needs access to unexported cleanStartupError
 
 import (
 	"errors"
+	"os"
 	"path/filepath"
 	"testing"
 
+	"go.uber.org/fx"
+	"sneak.berlin/go/vaultik/internal/config"
 	"sneak.berlin/go/vaultik/internal/pidlock"
 )
 
+// TestCleanStartupError starts a real fx app whose config has a
+// blob_size_limit below the minimum. fx puts annotations naming the
+// functions it was building in front of the config error; only those
+// may be removed, although the config error itself contains "): ".
 func TestCleanStartupError(t *testing.T) {
 	t.Parallel()
 
-	tests := []struct {
-		name string
-		in   string
-		want string
-	}{
-		{
-			name: "real fx error chain",
-			in: `could not build arguments for function ` +
-				`"sneak.berlin/go/vaultik/internal/cli".newSnapshotCreateCommand.func1.1 ` +
-				`(/Users/user/dev/vaultik/internal/cli/snapshot.go:71): ` +
-				`failed to build *vaultik.Vaultik: ` +
-				`could not build arguments for function ` +
-				`"sneak.berlin/go/vaultik/internal/vaultik".New ` +
-				`(/Users/user/dev/vaultik/internal/vaultik/vaultik.go:59): ` +
-				`failed to build storage.Storer: ` +
-				`received non-nil error from function ` +
-				`"sneak.berlin/go/vaultik/internal/storage".NewStorer ` +
-				`(/Users/user/dev/vaultik/internal/storage/module.go:23): ` +
-				`creating base path: mkdir /Volumes/BACKUPS: permission denied`,
-			want: `creating base path: mkdir /Volumes/BACKUPS: permission denied`,
-		},
-		{
-			name: "no fx wrapping",
-			in:   "plain error",
-			want: "plain error",
-		},
-		{
-			name: "single fx wrapping",
-			in:   `received non-nil error from function "foo" (file.go:1): underlying problem`,
-			want: "underlying problem",
-		},
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte("snapshots:\n"+
+		"  test:\n"+
+		"    paths: [/tmp/vaultik-test-source]\n"+
+		"storage_url: file:///tmp/vaultik-test-storage\n"+
+		"chunk_size: 10MB\n"+
+		"blob_size_limit: 30MB\n"), 0o600)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
 	}
 
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			t.Parallel()
+	app := fx.New(
+		fx.NopLogger,
+		fx.Supply(config.Path(path)),
+		config.Module,
+		fx.Invoke(func(*config.Config) {}),
+	)
 
-			//nolint:err113 // test constructs errors from table input
-			got := cleanStartupError(errors.New(tt.in)).Error()
-			if got != tt.want {
-				t.Errorf("got %q, want %q", got, tt.want)
-			}
-		})
+	err = app.Start(t.Context())
+	if err == nil {
+		t.Fatal("app started; want a blob_size_limit error")
+	}
+
+	want := "failed to load config: invalid config: blob_size_limit must " +
+		"be at least the largest chunk the chunker can emit (chunk_size " +
+		"times the FastCDC size spread): need at least 40000000 bytes"
+
+	got := cleanStartupError(err).Error()
+	if got != want {
+		t.Errorf("got %q, want %q", got, want)
+	}
+}
+
+// TestCleanStartupErrorNotFromFx checks that an error fx did not wrap is
+// shown unchanged.
+func TestCleanStartupErrorNotFromFx(t *testing.T) {
+	t.Parallel()
+
+	//nolint:err113 // a one-off error standing in for a startup failure
+	got := cleanStartupError(errors.New("plain error")).Error()
+	if got != "plain error" {
+		t.Errorf("got %q, want %q", got, "plain error")
 	}
 }
 

internal/cli/config.go

@@ -7,11 +7,16 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
+	"reflect"
+	"runtime"
+	"slices"
 	"strconv"
 	"strings"
+	"unicode/utf8"
 
 	"github.com/spf13/cobra"
 	"gopkg.in/yaml.v3"
+	"sneak.berlin/go/vaultik/internal/config"
 	"sneak.berlin/go/vaultik/internal/ui"
 )
 
@@ -26,11 +31,14 @@ const configSetArgs = 2
 // parent config dirs (e.g. ~/.config) are conventionally traversable.
 const configDirMode = 0o755
 
-// configYAMLIndent matches the 2-space indentation of defaultConfigTemplate,
+// configYAMLIndent matches the 2-space indentation of the default config,
 // so `config set` writes the file back with the same indentation rather than
 // yaml.Marshal's 4-space default.
 const configYAMLIndent = 2
 
+// yamlStringTag is YAML's tag for a string scalar.
+const yamlStringTag = "!!str"
+
 var (
 	errConfigExists     = errors.New("config file already exists")
 	errEmptyConfig      = errors.New("empty config file")
@@ -38,23 +46,44 @@ var (
 	errNeedNumericIndex = errors.New("key is a list; use a numeric index")
 	errIndexOutOfRange  = errors.New("index out of range")
 	errNotMapOrList     = errors.New("key is not a map or list")
+	errWouldNotLoad     = errors.New(
+		"vaultik would not load the config after this change, " +
+			"so the file was not changed")
 )
 
-const defaultConfigTemplate = `# vaultik configuration
+// macOS is the value of runtime.GOOS on macOS.
+const macOS = "darwin"
+
+// defaultConfigText returns the file `config init` writes on the operating
+// system goos, named as runtime.GOOS names it. Only macOS gets the apps
+// snapshot, since /Applications exists nowhere else.
+func defaultConfigText(goos string) string {
+	if goos == macOS {
+		return defaultConfigHead + appsSnapshotConfig + defaultConfigTail
+	}
+
+	return defaultConfigHead + defaultConfigTail
+}
+
+// defaultConfigHead is the default config up to the end of the home snapshot.
+const defaultConfigHead = `# vaultik configuration
 # Documentation: https://sneak.berlin/go/vaultik
 
 # ─── REQUIRED ────────────────────────────────────────────────────────────────
 
-# Age recipient public keys for encryption.
+# Age recipient public keys for encryption. snapshot create needs at least
+# one; listing, verifying and restoring do not, so a machine that only
+# restores can leave this empty.
 # Backups are encrypted to ALL listed recipients; any one of the corresponding
 # private keys can decrypt. Adding a recipient later does not re-encrypt data
 # already stored: deduplicated chunks and existing blobs stay encrypted to the
 # earlier recipients, so a newly added key cannot restore them on its own (see
-# docs/REPOSTRUCTURE.md, Accepted Risks). Generate a keypair with:
+# docs/REPOSTRUCTURE.md, Accepted Risks). Generate a keypair and add its
+# public key with:
 #   age-keygen -o vaultik_backup_private_key.txt
 #   grep 'public key' vaultik_backup_private_key.txt
-age_recipients:
-  - age1REPLACE_WITH_YOUR_PUBLIC_KEY
+#   vaultik config set age_recipients.0 age1...
+age_recipients: []
 
 # Named snapshots. Each snapshot backs up one or more paths and can have its
 # own exclude patterns in addition to the global excludes below.
@@ -155,7 +184,10 @@ snapshots:
       - "/Movies/ProxyMedia"
       - "/Music/iTunes/Album Artwork"
       - "/Pictures/iPod Photo Cache"
+`
 
+// appsSnapshotConfig is the apps snapshot of the default config.
+const appsSnapshotConfig = `
   # Third-party applications. OS-provided apps live in /System/Applications
   # on modern macOS and are never in /Applications, but Apple-installed
   # App Store apps (Safari, GarageBand, iWork, iMovie) are excluded since
@@ -176,7 +208,10 @@ snapshots:
       - "/Steam.app"
       - "/VirtualBox.app"
       - "/Utilities/Adobe Installers"
+`
 
+// defaultConfigTail is the default config after the snapshots.
+const defaultConfigTail = `
 # Storage backend (pick ONE of the three forms below).
 #
 # S3-compatible:
@@ -196,7 +231,7 @@ storage_url: ""
 #   access_key_id: YOUR_ACCESS_KEY
 #   secret_access_key: YOUR_SECRET_KEY
 #   # region: us-east-1        # Default: us-east-1
-#   # part_size: 5MB           # Multipart upload part size. Default: 5MB
+#   # part_size: 5MiB          # Upload part size, 5MiB to 5GiB. Default: 5MiB
 #   # For the s3:// form, disable TLS with ?ssl=false in the URL, not use_ssl.
 
 # ─── OPTIONAL ────────────────────────────────────────────────────────────────
@@ -211,7 +246,7 @@ storage_url: ""
 
 # Average chunk size for content-defined chunking (FastCDC).
 # Smaller = better deduplication but more metadata overhead.
-# Accepts: 1MB, 10M, 64KB, etc.
+# Minimum 1MB. Accepts: 1MB, 10M, 64MiB, etc.
 # Default: 10MB
 # chunk_size: 10MB
 
@@ -226,10 +261,23 @@ storage_url: ""
 # Default: 3
 # compression_level: 3
 
+# How old a blob that no snapshot lists must be before prune deletes it,
+# and how old every object in a directory under metadata/ with no manifest
+# must be before prune deletes that directory. A backup uploads its blobs
+# and its db.zst.age before its manifest, so on a destination shared
+# between hosts or users this must be longer than the longest backup run.
+# It does not protect a blob a backup reuses from an earlier run that
+# stopped before its manifest: that blob keeps the time of its first
+# upload, so a prune can delete it.
+# Accepts: 36h, 90m, etc. 0s deletes such blobs and directories at once.
+# Default: 24h
+# prune_grace_period: 24h
+
 # Hostname used in snapshot IDs. Default: system hostname.
 # hostname: myserver
 
-# Path to the local SQLite index database.
+# Path to the local SQLite index database. It must be an absolute path;
+# ~ is expanded.
 # Default: the platform data directory, e.g.
 #   macOS: ~/Library/Application Support/vaultik/index.sqlite
 #   Linux: ~/.local/share/vaultik/index.sqlite
@@ -242,6 +290,12 @@ func NewConfigCommand() *cobra.Command {
 		Use:   "config",
 		Short: "Manage the configuration file",
 		Long:  "Commands for creating, editing, and querying the vaultik config file.",
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	cmd.AddCommand(newConfigInitCommand())
@@ -280,7 +334,8 @@ on macOS, ~/.config/ on Linux, /etc/vaultik/ as root).`,
 				return fmt.Errorf("creating config directory %s: %w", dir, err)
 			}
 
-			err = os.WriteFile(path, []byte(defaultConfigTemplate), configFileMode)
+			err = os.WriteFile(path, []byte(defaultConfigText(runtime.GOOS)),
+				configFileMode)
 			if err != nil {
 				return fmt.Errorf("writing config file: %w", err)
 			}
@@ -314,8 +369,15 @@ func newConfigEditCommand() *cobra.Command {
 				editor = "vi"
 			}
 
+			// Run $EDITOR through the shell, as git does, so a value
+			// with arguments such as "code --wait" works. The path
+			// goes in as an argument ("$@"), not into the command
+			// text, so a path with spaces or quotes arrives whole.
+			// sh -c takes the word after the command as $0, which
+			// is why editor appears twice.
 			//nolint:gosec // G204: launching the operator's own $EDITOR is the point
-			ed := exec.CommandContext(cmd.Context(), editor, path)
+			ed := exec.CommandContext(cmd.Context(),
+				"sh", "-c", editor+` "$@"`, editor, path)
 			ed.Stdin = os.Stdin
 			ed.Stdout = os.Stdout
 			ed.Stderr = os.Stderr
@@ -376,8 +438,16 @@ func newConfigSetCommand() *cobra.Command {
 		Use:   "set <key> <value>",
 		Short: "Set a config value by dotted path (e.g. compression_level 5)",
 		Long: `Sets a scalar config value addressed by dotted YAML path and writes
-the file back, preserving comments and formatting. Intermediate maps
-are created as needed.
+the file back, preserving comments and formatting. Missing keys along
+the path are created: as a list when the next path element is a
+number, otherwise as a map.
+
+If the config loads, a change after which it would not, such as
+compression_level 99, is refused and the file is left unchanged. A
+config that does not load is checked one setting at a time. Examples
+are the file config init writes before storage_url is set, and one
+with a ${ENV:...} reference to a variable not set in this shell. In
+such a config, an unknown key or a value of the wrong type is refused.
 
 Examples:
   vaultik config set storage_url "file:///mnt/backups"
@@ -407,7 +477,21 @@ func writeConfigSet(out *ui.Writer, path, key, value string) error {
 		return err
 	}
 
-	err = yamlPathSet(root, strings.Split(key, "."), value)
+	oldData, err := marshalConfigYAML(root)
+	if err != nil {
+		return fmt.Errorf("marshaling config: %w", err)
+	}
+
+	keys := strings.Split(key, ".")
+
+	if keys[0] != config.EnvSection {
+		_, err = config.SettingType(keys)
+		if err != nil {
+			return err
+		}
+	}
+
+	err = yamlPathSet(root, keys, value)
 	if err != nil {
 		return err
 	}
@@ -417,6 +501,21 @@ func writeConfigSet(out *ui.Writer, path, key, value string) error {
 		return fmt.Errorf("marshaling config: %w", err)
 	}
 
+	setting, err := yamlPathOnly(root, keys)
+	if err != nil {
+		return err
+	}
+
+	settingData, err := marshalConfigYAML(setting)
+	if err != nil {
+		return fmt.Errorf("marshaling config: %w", err)
+	}
+
+	err = checkConfigChange(oldData, data, settingData)
+	if err != nil {
+		return fmt.Errorf("%w: %w", errWouldNotLoad, err)
+	}
+
 	err = os.WriteFile(path, data, configFileMode)
 	if err != nil {
 		return fmt.Errorf("writing config file: %w", err)
@@ -438,8 +537,60 @@ func writeConfigSet(out *ui.Writer, path, key, value string) error {
 	return nil
 }
 
+// checkConfigChange returns the reason vaultik would not load newData, the
+// config after a change, if the change brings that failure in. If oldData,
+// the config before it, loads, the new one must load too. A config that
+// does not load, such as one still being filled in or one with a ${...}
+// reference that cannot be filled in here, is checked one setting at a
+// time: settingData, the changed setting on its own, must decode into the
+// config type.
+func checkConfigChange(oldData, newData, settingData []byte) error {
+	oldCfg, err := config.Parse(oldData)
+	if err == nil && oldCfg.Validate() == nil {
+		newCfg, err := config.Parse(newData)
+		if err != nil {
+			return err
+		}
+
+		return newCfg.Validate()
+	}
+
+	_, err = config.Parse(settingData)
+
+	return err
+}
+
+// yamlPathOnly returns a document holding only the value at the dotted key
+// path in root, inside the maps and lists that lead to it there.
+func yamlPathOnly(root *yaml.Node, keys []string) (*yaml.Node, error) {
+	node, err := yamlPathGet(root, keys)
+	if err != nil {
+		return nil, err
+	}
+
+	for i, key := range slices.Backward(keys) {
+		parent, err := yamlPathGet(root, keys[:i])
+		if err != nil {
+			return nil, err
+		}
+
+		if parent.Kind == yaml.SequenceNode {
+			node = &yaml.Node{
+				Kind: yaml.SequenceNode, Content: []*yaml.Node{node},
+			}
+		} else {
+			keyNode := &yaml.Node{Kind: yaml.ScalarNode, Value: key}
+			node = &yaml.Node{
+				Kind: yaml.MappingNode, Content: []*yaml.Node{keyNode, node},
+			}
+		}
+	}
+
+	return &yaml.Node{Kind: yaml.DocumentNode, Content: []*yaml.Node{node}}, nil
+}
+
 // marshalConfigYAML renders a config document tree with 2-space indentation,
-// matching defaultConfigTemplate. yaml.Marshal defaults to 4 spaces, which
+// matching the default config. yaml.Marshal defaults to 4 spaces, which
 // would reindent the whole file on the first `config set` despite the promise
 // to preserve formatting.
 func marshalConfigYAML(root *yaml.Node) ([]byte, error) {
@@ -544,8 +695,8 @@ func yamlPathGet(root *yaml.Node, keys []string) (*yaml.Node, error) {
 	return node, nil
 }
 
-// yamlPathSet navigates a dotted key path, creating intermediate maps as
-// needed, and sets the final key to the given scalar value. Numeric path
+// yamlPathSet navigates a dotted key path, creating missing keys as needed,
+// and sets the final key to the given scalar value. Numeric path
 // components index into sequences; an index equal to the sequence length
 // appends a new element (e.g. "age_recipients.1" on a 1-element list).
 func yamlPathSet(root *yaml.Node, keys []string, value string) error {
@@ -558,12 +709,12 @@ func yamlPathSet(root *yaml.Node, keys []string, value string) error {
 		node = node.Content[0]
 	}
 
-	for i, key := range keys {
+	for i := range keys {
 		last := i == len(keys)-1
 
 		switch node.Kind {
 		case yaml.MappingNode:
-			node = yamlSetInMapping(node, key, value, last)
+			node = yamlSetInMapping(node, keys, i, value, last)
 		case yaml.SequenceNode:
 			next, err := yamlSetInSequence(node, keys, i, value, last)
 			if err != nil {
@@ -580,17 +731,38 @@ func yamlPathSet(root *yaml.Node, keys []string, value string) error {
 		}
 	}
 
+	// config.Load reads the file through untyped YAML, which turns an
+	// unquoted 00112233 into the number 38043 and 1e5 into 100000. Tagging
+	// a string setting as a string makes the encoder quote such a value.
+	// Other settings stay unquoted, so compression_level 9 is a number.
+	// The encoder refuses to write a value that is not valid UTF-8 as a
+	// string. Left untagged, such a value is written as base64 !!binary and
+	// loads back unchanged.
+	if configKeyIsString(keys) && utf8.ValidString(value) {
+		node.Tag = yamlStringTag
+	}
+
 	return nil
 }
 
-// yamlSetInMapping resolves (creating if needed) the value node for key
-// within a mapping node, setting it to value when it is the final path
-// element, and returns the node to descend into.
-func yamlSetInMapping(node *yaml.Node, key, value string, last bool) *yaml.Node {
+// configKeyIsString reports whether the dotted key names a string in
+// config.Config, as s3.access_key_id and snapshots.home.exclude.0 do.
+func configKeyIsString(keys []string) bool {
+	typ, err := config.SettingType(keys)
+
+	return err == nil && typ.Kind() == reflect.String
+}
+
+// yamlSetInMapping resolves (creating if needed) the value node for the
+// path element keys[i] within a mapping node, setting it to value when it
+// is the final path element, and returns the node to descend into.
+func yamlSetInMapping(
+	node *yaml.Node, keys []string, i int, value string, last bool,
+) *yaml.Node {
 	var valueNode *yaml.Node
 
 	for j := 0; j+1 < len(node.Content); j += 2 {
-		if node.Content[j].Value == key {
+		if node.Content[j].Value == keys[i] {
 			valueNode = node.Content[j+1]
 
 			break
@@ -598,13 +770,8 @@ func yamlSetInMapping(node *yaml.Node, key, value string, last bool) *yaml.Node
 	}
 
 	if valueNode == nil {
-		keyNode := &yaml.Node{Kind: yaml.ScalarNode, Value: key}
-
-		valueNode = &yaml.Node{Kind: yaml.MappingNode}
-		if last {
-			valueNode = &yaml.Node{Kind: yaml.ScalarNode, Value: value}
-		}
-
+		keyNode := &yaml.Node{Kind: yaml.ScalarNode, Value: keys[i]}
+		valueNode = newPathNode(keys, i, value)
 		node.Content = append(node.Content, keyNode, valueNode)
 	} else if last {
 		setScalar(valueNode, value)
@@ -632,12 +799,7 @@ func yamlSetInSequence(
 	}
 
 	if idx == len(node.Content) {
-		newNode := &yaml.Node{Kind: yaml.MappingNode}
-		if last {
-			newNode = &yaml.Node{Kind: yaml.ScalarNode, Value: value}
-		}
-
-		node.Content = append(node.Content, newNode)
+		node.Content = append(node.Content, newPathNode(keys, i, value))
 	} else if last {
 		setScalar(node.Content[idx], value)
 	}
@@ -645,6 +807,22 @@ func yamlSetInSequence(
 	return node.Content[idx], nil
 }
 
+// newPathNode returns the node to create for the missing path element
+// keys[i]: the value when it is the final path element, otherwise an empty
+// list when the next path element is a number and an empty map when not.
+func newPathNode(keys []string, i int, value string) *yaml.Node {
+	if i == len(keys)-1 {
+		return &yaml.Node{Kind: yaml.ScalarNode, Value: value}
+	}
+
+	_, err := strconv.Atoi(keys[i+1])
+	if err == nil {
+		return &yaml.Node{Kind: yaml.SequenceNode}
+	}
+
+	return &yaml.Node{Kind: yaml.MappingNode}
+}
+
 // setScalar overwrites a node in place with a plain scalar value.
 func setScalar(n *yaml.Node, value string) {
 	n.Kind = yaml.ScalarNode

internal/cli/config_test.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"os"
 	"path/filepath"
+	"strconv"
 	"strings"
 	"testing"
 
@@ -12,20 +13,23 @@ import (
 	"sneak.berlin/go/vaultik/internal/ui"
 )
 
-// TestDefaultConfigTemplateParses ensures the init template is valid YAML
-// that unmarshals into the Config struct with the expected snapshots.
-func TestDefaultConfigTemplateParses(t *testing.T) {
+// TestDefaultConfigParsesOnMacOS ensures the config `config init` writes on
+// macOS is valid YAML that unmarshals into the Config struct with the
+// expected snapshots.
+func TestDefaultConfigParsesOnMacOS(t *testing.T) {
 	t.Parallel()
 
 	var cfg config.Config
 
-	err := yaml.Unmarshal([]byte(defaultConfigTemplate), &cfg)
+	err := yaml.Unmarshal([]byte(defaultConfigText("darwin")), &cfg)
 	if err != nil {
 		t.Fatalf("default config template is not valid YAML: %v", err)
 	}
 
-	if len(cfg.AgeRecipients) != 1 {
-		t.Errorf("expected 1 placeholder age recipient, got %d", len(cfg.AgeRecipients))
+	// A placeholder recipient would fail config.Load, so the template
+	// leaves the list empty.
+	if len(cfg.AgeRecipients) != 0 {
+		t.Errorf("expected no age recipients, got %d", len(cfg.AgeRecipients))
 	}
 
 	home, ok := cfg.Snapshots["home"]
@@ -55,6 +59,427 @@ func TestDefaultConfigTemplateParses(t *testing.T) {
 	}
 }
 
+// TestDefaultConfigOnLinux checks that the config `config init` writes on
+// Linux loads with its home snapshot and without the apps snapshot, whose
+// /Applications exists only on macOS.
+func TestDefaultConfigOnLinux(t *testing.T) {
+	t.Parallel()
+
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte(defaultConfigText("linux")), configFileMode)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+
+	err = writeConfigSet(ui.NewWithColor(&bytes.Buffer{}, false),
+		path, "storage_url", "file:///mnt/backups")
+	if err != nil {
+		t.Fatalf("config set storage_url: %v", err)
+	}
+
+	cfg, err := config.Load(path)
+	if err != nil {
+		t.Fatalf("config.Load: %v", err)
+	}
+
+	if _, ok := cfg.Snapshots["home"]; !ok {
+		t.Error("expected 'home' snapshot in the Linux default config")
+	}
+
+	if _, ok := cfg.Snapshots["apps"]; ok {
+		t.Error("the Linux default config should have no 'apps' snapshot")
+	}
+}
+
+// testRecipient is a valid age public key.
+const testRecipient = "age1278m9q7dp3chsh2dcy82qk27v047zywyvtxwnj4cvt0z65jw6a7q5dqhfj"
+
+// TestConfigSetRecipientOnFreshConfig follows the README quickstart: on the
+// file `config init` writes, `config set age_recipients.0` and
+// `config set storage_url` give a config that loads with that recipient.
+// Neither is refused, although the config does not load until both are set;
+// once it loads, a recipient that is not a valid key is.
+func TestConfigSetRecipientOnFreshConfig(t *testing.T) {
+	t.Parallel()
+
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte(defaultConfigText(macOS)), configFileMode)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+
+	out := ui.NewWithColor(&bytes.Buffer{}, false)
+
+	err = writeConfigSet(out, path, "age_recipients.0", testRecipient)
+	if err != nil {
+		t.Fatalf("config set age_recipients.0: %v", err)
+	}
+
+	err = writeConfigSet(out, path, "storage_url", "file:///mnt/backups")
+	if err != nil {
+		t.Fatalf("config set storage_url: %v", err)
+	}
+
+	cfg, err := config.Load(path)
+	if err != nil {
+		t.Fatalf("config.Load: %v", err)
+	}
+
+	if len(cfg.AgeRecipients) != 1 || cfg.AgeRecipients[0] != testRecipient {
+		t.Errorf("age_recipients = %v, want [%s]", cfg.AgeRecipients, testRecipient)
+	}
+
+	err = writeConfigSet(out, path, "age_recipients.0", "notakey")
+	if err == nil || !strings.Contains(err.Error(), "not a valid recipient") {
+		t.Errorf("config set age_recipients.0 notakey gives error %v, "+
+			"want one saying it is not a valid recipient", err)
+	}
+
+	_, err = config.Load(path)
+	if err != nil {
+		t.Errorf("config.Load after the refused change: %v", err)
+	}
+}
+
+// configWithoutRecipients loads, but has no age_recipients key and no
+// snapshot named work.
+const configWithoutRecipients = `storage_url: file:///mnt/backups
+snapshots:
+  home:
+    paths:
+      - "~"
+`
+
+// configWithUnsetReference loads only where VAULTIK_TEST_UNSET_SECRET is
+// set, as a backup job's environment might set it. No test sets it.
+const configWithUnsetReference = configWithoutRecipients + `s3:
+  secret_access_key: ${ENV:VAULTIK_TEST_UNSET_SECRET}
+`
+
+// configWithLevelReference returns configWithoutRecipients with a ${FILE:...}
+// reference on compression_level to a file holding 9, so that the config
+// loads once the reference is filled in.
+func configWithLevelReference(t *testing.T) string {
+	t.Helper()
+
+	levelPath := filepath.Join(t.TempDir(), "level")
+
+	err := os.WriteFile(levelPath, []byte("9"), configFileMode)
+	if err != nil {
+		t.Fatalf("write level: %v", err)
+	}
+
+	return configWithoutRecipients +
+		"compression_level: ${FILE:" + levelPath + "}\n"
+}
+
+// TestConfigSetIndexUnderMissingKey checks that a list index under a key
+// the file does not have creates a list, and that the result loads.
+func TestConfigSetIndexUnderMissingKey(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		key   string
+		value string
+		field func(cfg *config.Config) []string
+	}{
+		{"age_recipients.0", testRecipient,
+			func(cfg *config.Config) []string { return cfg.AgeRecipients }},
+		{"snapshots.work.paths.0", "/srv/work",
+			func(cfg *config.Config) []string { return cfg.Snapshots["work"].Paths }},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.key, func(t *testing.T) {
+			t.Parallel()
+
+			path := filepath.Join(t.TempDir(), "config.yml")
+
+			err := os.WriteFile(path, []byte(configWithoutRecipients), configFileMode)
+			if err != nil {
+				t.Fatalf("write config: %v", err)
+			}
+
+			err = writeConfigSet(ui.NewWithColor(&bytes.Buffer{}, false),
+				path, tt.key, tt.value)
+			if err != nil {
+				t.Fatalf("config set %s: %v", tt.key, err)
+			}
+
+			cfg, err := config.Load(path)
+			if err != nil {
+				t.Fatalf("config.Load: %v", err)
+			}
+
+			got := tt.field(cfg)
+			if len(got) != 1 || got[0] != tt.value {
+				t.Errorf("%s gives %v, want [%s]", tt.key, got, tt.value)
+			}
+		})
+	}
+}
+
+// TestConfigSetKeepsFileThatWouldNotLoad checks that config set refuses a
+// change after which the config would not load, with an error giving the
+// reason, and leaves the file as it was. On the file `config init` writes,
+// which does not load until storage_url is set, and on one that cannot be
+// read here, only an unknown key or a value of the wrong type is refused.
+func TestConfigSetKeepsFileThatWouldNotLoad(t *testing.T) {
+	t.Parallel()
+
+	levelReference := configWithLevelReference(t)
+
+	tests := []struct {
+		config string
+		key    string
+		value  string
+		reason string
+	}{
+		// Keys that no setting has, on a config that loads and on one
+		// that does not.
+		{configWithoutRecipients, "nosuchkey", "1",
+			`unknown config key "nosuchkey"`},
+		{configWithoutRecipients, "snapshots.home.exlude", "/tmp",
+			`unknown config key "snapshots.home.exlude"`},
+		{defaultConfigText(macOS), "compression_lvl", "9",
+			`unknown config key "compression_lvl"`},
+		// Index 1 of the empty list the missing key gives.
+		{configWithoutRecipients, "age_recipients.1", testRecipient,
+			"index out of range"},
+		// A list set to a single value.
+		{configWithoutRecipients, "snapshots.home.paths", "/srv",
+			"into []string"},
+		{configWithoutRecipients, "chunk_size", "fast",
+			`chunk_size: "fast" is not a size`},
+		// Values of the right type that fail validation.
+		{configWithoutRecipients, "compression_level", "99",
+			"compression_level must be between 1 and 19"},
+		{configWithoutRecipients, "chunk_size", "1KB",
+			"chunk_size must be at least 1MB"},
+		// With its reference filled in, the config loads.
+		{levelReference, "blob_size_limit", "1MB",
+			"blob_size_limit must be at least"},
+		// The type check on a config that does not load yet.
+		{defaultConfigText(macOS), "snapshots.apps.paths", "/srv",
+			"into []string"},
+		// The type check on a config that cannot be read here.
+		{configWithUnsetReference, "snapshots.home.paths", "/x",
+			"cannot unmarshal !!str `/x`"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.key+" "+tt.value, func(t *testing.T) {
+			t.Parallel()
+
+			path := filepath.Join(t.TempDir(), "config.yml")
+
+			err := os.WriteFile(path, []byte(tt.config), configFileMode)
+			if err != nil {
+				t.Fatalf("write config: %v", err)
+			}
+
+			err = writeConfigSet(ui.NewWithColor(&bytes.Buffer{}, false),
+				path, tt.key, tt.value)
+			if err == nil || !strings.Contains(err.Error(), tt.reason) {
+				t.Errorf("config set %s %s gives error %v, want one saying %q",
+					tt.key, tt.value, err, tt.reason)
+			}
+
+			data, err := os.ReadFile(path) //nolint:gosec // G304: test-controlled path
+			if err != nil {
+				t.Fatalf("read config: %v", err)
+			}
+
+			if string(data) != tt.config {
+				t.Errorf("config set %s changed the file to:\n%s",
+					tt.key, data)
+			}
+		})
+	}
+}
+
+// TestConfigSetRefusesOnlyItsOwnFailure checks that setting another key
+// works on a config with a ${...} reference on compression_level, which
+// loads because loading fills the reference in, on one whose
+// compression_level already fails to load, and on one with a reference
+// that cannot be filled in here.
+func TestConfigSetRefusesOnlyItsOwnFailure(t *testing.T) {
+	t.Parallel()
+
+	configs := map[string]string{
+		"reference":       configWithLevelReference(t),
+		"failing":         configWithoutRecipients + "compression_level: fast\n",
+		"unset reference": configWithUnsetReference,
+	}
+
+	for name, configYAML := range configs {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			path := filepath.Join(t.TempDir(), "config.yml")
+
+			err := os.WriteFile(path, []byte(configYAML), configFileMode)
+			if err != nil {
+				t.Fatalf("write config: %v", err)
+			}
+
+			err = writeConfigSet(ui.NewWithColor(&bytes.Buffer{}, false),
+				path, "hostname", "myhost")
+			if err != nil {
+				t.Fatalf("config set hostname: %v", err)
+			}
+
+			data, err := os.ReadFile(path) //nolint:gosec // G304: test-controlled path
+			if err != nil {
+				t.Fatalf("read config: %v", err)
+			}
+
+			if !strings.Contains(string(data), "hostname: myhost") {
+				t.Errorf("config set hostname wrote:\n%s", data)
+			}
+		})
+	}
+}
+
+// TestConfigSetWritesEnvSection checks that config set writes a key under
+// the top-level env section, whose keys are not settings, on a config that
+// loads.
+func TestConfigSetWritesEnvSection(t *testing.T) {
+	const name = "VAULTIK_TEST_CONFIG_SET_ENV"
+
+	// Loading the changed config sets the variable; t.Setenv restores it
+	// after the test.
+	t.Setenv(name, "")
+
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte(configWithoutRecipients), configFileMode)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+
+	err = writeConfigSet(ui.NewWithColor(&bytes.Buffer{}, false),
+		path, "env."+name, "value")
+	if err != nil {
+		t.Fatalf("config set env.%s: %v", name, err)
+	}
+
+	data, err := os.ReadFile(path) //nolint:gosec // G304: test-controlled path
+	if err != nil {
+		t.Fatalf("read config: %v", err)
+	}
+
+	if !strings.Contains(string(data), "env:\n  "+name+": value\n") {
+		t.Errorf("config set env.%s wrote:\n%s", name, data)
+	}
+}
+
+// TestConfigSetStringLooksLikeNumber sets string settings to values that
+// YAML reads as numbers or booleans when they are unquoted, and checks that
+// config.Load returns each one unchanged.
+func TestConfigSetStringLooksLikeNumber(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		key   string
+		value string
+		field func(cfg *config.Config) string
+	}{
+		{"s3.access_key_id", "00112233",
+			func(cfg *config.Config) string { return cfg.S3.AccessKeyID }},
+		{"s3.secret_access_key", "12345678901234567890123456789012",
+			func(cfg *config.Config) string { return cfg.S3.SecretAccessKey }},
+		{"hostname", "007",
+			func(cfg *config.Config) string { return cfg.Hostname }},
+		{"s3.prefix", "1e5",
+			func(cfg *config.Config) string { return cfg.S3.Prefix }},
+		{"s3.bucket", "true",
+			func(cfg *config.Config) string { return cfg.S3.Bucket }},
+		{"s3.region", "FALSE",
+			func(cfg *config.Config) string { return cfg.S3.Region }},
+		{"snapshots.home.exclude.0", "1.10",
+			func(cfg *config.Config) string { return cfg.Snapshots["home"].Exclude[0] }},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.key+"="+tt.value, func(t *testing.T) {
+			t.Parallel()
+
+			cfg := loadAfterConfigSet(t, tt.key, tt.value)
+
+			got := tt.field(cfg)
+			if got != tt.value {
+				t.Errorf("%s = %q after config set %q", tt.key, got, tt.value)
+			}
+		})
+	}
+}
+
+// TestConfigSetNonUTF8Path checks that config set still accepts a value that
+// is not valid UTF-8, such as a path with a Latin-1 file name, and that
+// config.Load returns it unchanged.
+func TestConfigSetNonUTF8Path(t *testing.T) {
+	t.Parallel()
+
+	const dir = "/srv/caf\xe9"
+
+	cfg := loadAfterConfigSet(t, "snapshots.home.paths.0", dir)
+
+	got := cfg.Snapshots["home"].Paths[0]
+	if got != dir {
+		t.Errorf("snapshots.home.paths.0 = %q, want %q", got, dir)
+	}
+}
+
+// TestConfigSetNumberStaysNumber checks that a number set for an integer
+// setting is still read as a number, not as a quoted string.
+func TestConfigSetNumberStaysNumber(t *testing.T) {
+	t.Parallel()
+
+	const level = 9
+
+	cfg := loadAfterConfigSet(t, "compression_level", strconv.Itoa(level))
+
+	if cfg.CompressionLevel != level {
+		t.Errorf("compression_level = %d, want %d", cfg.CompressionLevel, level)
+	}
+}
+
+// loadAfterConfigSet writes the file `config init` writes, sets storage_url
+// to a local directory so that the file passes validation, applies
+// `config set key value` and returns what config.Load reads back.
+func loadAfterConfigSet(t *testing.T, key, value string) *config.Config {
+	t.Helper()
+
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte(defaultConfigText(macOS)), configFileMode)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+
+	out := ui.NewWithColor(&bytes.Buffer{}, false)
+
+	err = writeConfigSet(out, path, "storage_url", "file:///mnt/backups")
+	if err != nil {
+		t.Fatalf("config set storage_url: %v", err)
+	}
+
+	err = writeConfigSet(out, path, key, value)
+	if err != nil {
+		t.Fatalf("config set %s: %v", key, err)
+	}
+
+	cfg, err := config.Load(path)
+	if err != nil {
+		t.Fatalf("config.Load: %v", err)
+	}
+
+	return cfg
+}
+
 const testYAML = `# top comment
 compression_level: 3
 age_recipients:
@@ -297,6 +722,30 @@ func TestWriteConfigSetTightensMode(t *testing.T) {
 	}
 }
 
+// TestConfigEditEditorWithArgument checks that an $EDITOR carrying an
+// argument works and that a config path with a space reaches it whole.
+// The "editor" is cp with the replacement file as its argument, so the
+// config ends up with the replacement's content only if both arrived.
+func TestConfigEditEditorWithArgument(t *testing.T) {
+	dir := t.TempDir()
+	path := seedFile(t, dir, "my config.yml", "compression_level: 3\n")
+	edited := seedFile(t, dir, "edited.yml", "compression_level: 9\n")
+
+	setRootFlags(t, RootFlags{ConfigPath: path})
+	t.Setenv("EDITOR", "cp '"+edited+"'")
+
+	mustExecute(t, newConfigEditCommand())
+
+	data, err := os.ReadFile(path) //nolint:gosec // G304: test-controlled path
+	if err != nil {
+		t.Fatalf("reading config back: %v", err)
+	}
+
+	if string(data) != "compression_level: 9\n" {
+		t.Errorf("config after edit = %q, want the edited content", data)
+	}
+}
+
 func splitPath(s string) []string {
 	return strings.Split(s, ".")
 }

internal/cli/database.go

@@ -6,7 +6,9 @@ import (
 
 	"github.com/spf13/cobra"
 	"sneak.berlin/go/vaultik/internal/config"
+	"sneak.berlin/go/vaultik/internal/database"
 	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/vaultik"
 )
 
 // NewDatabaseCommand creates the database command group
@@ -15,6 +17,12 @@ func NewDatabaseCommand() *cobra.Command {
 		Use:   "database",
 		Short: "Manage the local state database",
 		Long:  `Commands for managing the local SQLite state database.`,
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	cmd.AddCommand(
@@ -84,27 +92,25 @@ Use --force to skip the confirmation prompt.`,
 
 				var confirm string
 
-				_, err = fmt.Scanln(&confirm)
+				// A failure to read the answer means no confirmation.
+				_, err = fmt.Fscanln(cmd.InOrStdin(), &confirm)
 				if err != nil || confirm != "yes" {
 					_, _ = fmt.Fprintln(w, "Aborted.")
 
-					//nolint:nilerr // a failed/aborted confirmation is a clean abort
-					return nil
+					return vaultik.ErrNotConfirmed
 				}
 			}
 
-			// Delete the database file
-			err = os.Remove(dbPath)
-			if err != nil {
-				return fmt.Errorf("failed to delete database: %w", err)
+			// Delete the database file and its side files. A side file
+			// may not exist, so only failing to delete the database file
+			// is an error.
+			for _, indexFile := range database.IndexFiles(dbPath) {
+				err = os.Remove(indexFile)
+				if err != nil && indexFile == dbPath {
+					return fmt.Errorf("failed to delete database: %w", err)
+				}
 			}
 
-			// Also delete WAL and SHM files if they exist
-			walPath := dbPath + "-wal"
-			shmPath := dbPath + "-shm"
-			_ = os.Remove(walPath) // Ignore errors - files may not exist
-			_ = os.Remove(shmPath)
-
 			out.Infof("Local state database deleted: %s.", dbPath)
 			log.Info("Local state database deleted", "path", dbPath)
 

internal/cli/database_test.go

@@ -0,0 +1,75 @@
+package cli //nolint:testpackage // sets the unexported rootFlags directly
+
+import (
+	"bytes"
+	"errors"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// TestDatabaseDeleteDeclinedIsError checks that answering anything but
+// "yes" keeps the database and returns ErrNotConfirmed, so the command
+// exits non-zero.
+//
+//nolint:paralleltest // mutates the process-global rootFlags
+func TestDatabaseDeleteDeclinedIsError(t *testing.T) {
+	dir := t.TempDir()
+	dbPath, cfgPath := seedDatabaseDeleteConfig(t, dir)
+
+	setRootFlags(t, RootFlags{ConfigPath: cfgPath})
+
+	var out bytes.Buffer
+
+	cmd := newDatabaseDeleteCommand()
+	cmd.SetOut(&out)
+	cmd.SetErr(io.Discard)
+	cmd.SetIn(strings.NewReader("no\n"))
+	cmd.SetArgs(nil)
+
+	err := cmd.Execute()
+	if !errors.Is(err, vaultik.ErrNotConfirmed) {
+		t.Errorf("declined database delete returned %v, want %v",
+			err, vaultik.ErrNotConfirmed)
+	}
+
+	_, err = os.Stat(dbPath)
+	if err != nil {
+		t.Errorf("declined database delete removed the file: %v", err)
+	}
+
+	if !strings.Contains(out.String(), "Aborted.") {
+		t.Errorf("declined database delete printed %q, want Aborted.",
+			out.String())
+	}
+}
+
+// TestDatabaseDeleteRemovesSideFiles checks that the side files SQLite
+// keeps beside the database file are deleted with it.
+//
+//nolint:paralleltest // mutates the process-global rootFlags
+func TestDatabaseDeleteRemovesSideFiles(t *testing.T) {
+	dir := t.TempDir()
+	dbPath, cfgPath := seedDatabaseDeleteConfig(t, dir)
+
+	indexFiles := database.IndexFiles(dbPath)
+	for _, indexFile := range indexFiles {
+		seedFile(t, dir, filepath.Base(indexFile), "data")
+	}
+
+	setRootFlags(t, RootFlags{ConfigPath: cfgPath})
+
+	mustExecute(t, newDatabaseDeleteCommand(), "--force")
+
+	for _, indexFile := range indexFiles {
+		_, err := os.Stat(indexFile)
+		if !os.IsNotExist(err) {
+			t.Errorf("database delete left %s: stat err = %v", indexFile, err)
+		}
+	}
+}

internal/cli/entry.go

@@ -15,17 +15,28 @@ import (
 // the startup banner.
 const shortCommitLen = 12
 
+// exitCodeInterrupted is the exit status of a command that SIGINT or
+// SIGTERM stopped. It is 128 plus SIGINT's number, 2, which is what a
+// shell reports for a command stopped by Ctrl-C.
+const exitCodeInterrupted = 130
+
 // Entry is the main entry point for the CLI application.
-// It prints the startup banner to stdout (unless a banner-suppressing
+// It first calls globals.FillUnsetFromBuildInfo, which fills in the
+// version, commit and commit date that the linker flags did not set.
+// It prints the startup banner to stderr (unless a banner-suppressing
 // flag is present in os.Args — see bannerSuppressedInArgs), executes the
 // root cobra command, and routes any returned error through the
 // ui.Writer so the user sees a properly formatted "🛑 ERROR:" line.
+// The banner goes to stderr because stdout carries only the output the
+// user asked for, such as a completion script or a `config get` value.
 //
-// It returns the process exit code (0 on success, 1 on error) rather
-// than calling os.Exit, so that main's deferred profile writers run
-// before the process ends. See run in cmd/vaultik/main.go.
+// It returns the process exit code (0 on success, 130 when interrupted,
+// 1 on any other error) rather than calling os.Exit, so that main's
+// deferred profile writers run before the process ends. See run in
+// cmd/vaultik/main.go.
 func Entry() int {
-	emitStartupBanner(os.Args[1:], os.Stdout)
+	globals.FillUnsetFromBuildInfo()
+	emitStartupBanner(os.Args[1:], os.Stderr)
 
 	rootCmd := NewRootCommand()
 	rootCmd.SilenceErrors = true
@@ -33,14 +44,20 @@ func Entry() int {
 	err := rootCmd.Execute()
 	if err != nil {
 		// An operation that ran inside the fx app has already reported
-		// its own failure (and suppressed it under --json); errReported
-		// says so. Printing it again here would double the error line.
+		// its own failure (`snapshot verify --json` puts it in the
+		// document instead); errReported says so. Printing it again
+		// here would double the error line.
 		// Every other error — bad arguments, a config that would not
-		// load — reaches Entry unreported, so it is shown here.
+		// load, an interrupt — reaches Entry unreported, so it is shown
+		// here.
 		if !errors.Is(err, errReported) {
 			ReportErrorf("%s", err.Error())
 		}
 
+		if errors.Is(err, errInterrupted) {
+			return exitCodeInterrupted
+		}
+
 		return 1
 	}
 
@@ -49,9 +66,8 @@ func Entry() int {
 
 // emitStartupBanner writes the startup banner to w unless args (the
 // argument vector with the program name already stripped) contains a
-// flag that suppresses it. Split out of Entry so that the decision — the
-// only thing standing between a --json invocation and a parseable
-// stdout — is reachable from a test without running the whole CLI.
+// flag that suppresses it. Split out of Entry so that the decision is
+// reachable from a test without running the whole CLI.
 func emitStartupBanner(args []string, w io.Writer) {
 	if bannerSuppressedInArgs(args) {
 		return
@@ -86,8 +102,7 @@ func ReportErrorf(format string, args ...any) {
 // --json is a subcommand flag rather than a persistent one, but so is
 // --cron (it exists only on `snapshot create`), so this adds no new
 // class of imprecision. The only cost of a false positive is a missing
-// decorative banner; the cost of a false negative is a corrupt document
-// on stdout, so the scan errs deliberately in that direction.
+// decorative banner.
 func bannerSuppressedInArgs(args []string) bool {
 	for _, a := range args {
 		if a == "--" {

internal/cli/entry_banner_test.go

@@ -13,6 +13,7 @@ import (
 	"github.com/adrg/xdg"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/globals"
 )
 
 // Command words and flags used to build argument vectors below. They are
@@ -36,23 +37,14 @@ const (
 	// strips it before scanning, so it has to be present.
 	programName = "vaultik"
 
-	// someSnapshotID is any snapshot identifier: these tests never run
-	// the command, so it only has to occupy the positional argument.
+	// someSnapshotID only fills the positional argument; no test needs
+	// the snapshot to exist.
 	someSnapshotID = "host_2026-01-01T00:00:00Z"
 )
 
-// placeholderJSONDocument stands in for whatever document a --json
-// command writes to stdout. `snapshot list --json` with no snapshots
-// prints exactly this; the other --json commands print an object rather
-// than an array, but this test is not about their shape. It is about
-// what is on stdout *before* them, which is the same for all of them
-// because Entry prints the banner before cobra has parsed anything and
-// therefore before it can know which command is running.
-const placeholderJSONDocument = "[]\n"
-
 // jsonArgumentVectors are the argument vectors of every --json
 // invocation the CLI accepts, with the program name stripped exactly as
-// Entry strips it. Each one must leave stdout untouched by the banner.
+// Entry strips it. Each one must suppress the banner.
 //
 //nolint:gochecknoglobals // read-only test fixture shared by two tests
 var jsonArgumentVectors = map[string][]string{
@@ -74,39 +66,23 @@ var jsonArgumentVectors = map[string][]string{
 	},
 }
 
-// TestJSONInvocationStdoutIsExactlyOneDocument is the CLI-layer
-// regression guard for issue #106: `vaultik snapshot list --json | jq`
-// must work with no other flags.
-//
-// internal/vaultik's TestListSnapshots_JSONStdoutIsOnlyTheDocument
-// guards the same contract one layer down, but it calls the library
-// function directly and so cannot see Entry, which is where the
-// contamination was: the startup banner is written to stdout before
-// cobra parses anything, and the suppression scan did not know about
-// --json. The two banner lines and the blank line landed ahead of the
-// document and `jq` refused the result.
-//
-// The document is a constant here because this test is about the
-// argument vectors, one per --json command; the one that runs a real
-// command end to end is TestEntryJSONStdoutIsExactlyOneDocument below.
-func TestJSONInvocationStdoutIsExactlyOneDocument(t *testing.T) {
+// TestJSONInvocationSuppressesBanner checks that every --json
+// invocation suppresses the startup banner, as the README says --json
+// does along with --quiet and --cron. The scan is over the raw argument
+// vector, so each position and spelling of --json is listed.
+func TestJSONInvocationSuppressesBanner(t *testing.T) {
 	t.Parallel()
 
 	for name, argv := range jsonArgumentVectors {
 		t.Run(name, func(t *testing.T) {
 			t.Parallel()
 
-			var stdout bytes.Buffer
+			var banner bytes.Buffer
 
-			emitStartupBanner(argv, &stdout)
+			emitStartupBanner(argv, &banner)
 
-			require.Empty(t, stdout.String(),
-				"nothing may reach stdout ahead of a --json document")
-
-			_, err := stdout.WriteString(placeholderJSONDocument)
-			require.NoError(t, err)
-
-			requireExactlyOneJSONDocument(t, stdout.String())
+			assert.Empty(t, banner.String(),
+				"--json suppresses the banner")
 		})
 	}
 }
@@ -127,16 +103,45 @@ func TestBannerStillPrintedWithoutSuppressingFlag(t *testing.T) {
 		t.Run(name, func(t *testing.T) {
 			t.Parallel()
 
-			var stdout bytes.Buffer
+			var banner bytes.Buffer
 
-			emitStartupBanner(argv, &stdout)
+			emitStartupBanner(argv, &banner)
 
-			assert.Contains(t, stdout.String(), "starting up at",
+			assert.Contains(t, banner.String(), "starting up at",
 				"the banner belongs on invocations that did not opt out")
 		})
 	}
 }
 
+// TestBannerLabelsCommitDate checks that the banner prints the commit
+// and the commit date, each after its own label. Commit and CommitDate
+// get different values, so printing the wrong one fails.
+//
+//nolint:paralleltest // sets globals.Commit and globals.CommitDate
+func TestBannerLabelsCommitDate(t *testing.T) {
+	const (
+		commit     = "abc1234"
+		commitDate = "2026-11-09"
+	)
+
+	previousCommit, previousCommitDate := globals.Commit, globals.CommitDate
+
+	t.Cleanup(func() {
+		globals.Commit = previousCommit
+		globals.CommitDate = previousCommitDate
+	})
+
+	globals.Commit = commit
+	globals.CommitDate = commitDate
+
+	var banner bytes.Buffer
+
+	emitStartupBanner(nil, &banner)
+
+	assert.Contains(t, banner.String(),
+		"(commit "+commit+", commit date "+commitDate+")")
+}
+
 // TestBannerSuppressedInArgs covers the suppression scan directly,
 // including the flags that suppressed the banner before --json joined
 // them, so that adding --json cannot regress them.
@@ -172,9 +177,9 @@ func TestBannerSuppressedInArgs(t *testing.T) {
 
 // hermeticConfig is a complete, valid config that needs no network and
 // no credentials: file:// storage is exempt from the S3 credential
-// checks, and FileStorer over a directory that does not exist lists
-// zero objects without erroring. Chunk, blob and compression settings
-// are filled in by config.Load.
+// checks. A test that lists the destination must create its directory
+// first, because listing a directory that does not exist is an error.
+// Chunk, blob and compression settings are filled in by config.Load.
 const hermeticConfig = `age_recipients:
   - age1278m9q7dp3chsh2dcy82qk27v047zywyvtxwnj4cvt0z65jw6a7q5dqhfj
 snapshots:
@@ -196,22 +201,25 @@ hostname: test-host
 // `snapshot list` is the command chosen because it is the only --json
 // command that reaches its document without a populated destination
 // store: it reads the local index, streams `metadata/` (empty here),
-// and treats a barren destination as an empty list rather than a
-// failure.
+// and treats an empty destination directory as an empty list rather
+// than a failure.
 //
 // Not parallel: it replaces os.Args, os.Stdout and the xdg globals.
 func TestEntryJSONStdoutIsExactlyOneDocument(t *testing.T) {
 	dir := t.TempDir()
 	configPath := filepath.Join(dir, "config.yml")
+	storeDir := filepath.Join(dir, "store")
 
 	contents := fmt.Sprintf(hermeticConfig,
 		filepath.Join(dir, "source"),
-		filepath.Join(dir, "store"),
+		storeDir,
 		filepath.Join(dir, "index.sqlite"))
 
 	require.NoError(t,
 		os.WriteFile(configPath, []byte(contents), configFileMode))
 
+	require.NoError(t, os.Mkdir(storeDir, 0o750))
+
 	// The PID lock lives under xdg.DataHome, which xdg resolves at
 	// package init; point it at the temp dir so the test neither
 	// touches nor collides with the real one.
@@ -244,9 +252,7 @@ func TestEntryJSONStdoutIsExactlyOneDocument(t *testing.T) {
 // captureProcessStdout redirects the process's own stdout to a pipe for
 // the duration of fn and returns what was written to it. The redirection
 // has to be at the file-descriptor level rather than through an injected
-// writer, because the banner and the JSON encoder reach os.Stdout
-// independently and the point of the test is that both land in the same
-// place.
+// writer, because the commands Entry runs reach os.Stdout directly.
 //
 // Not parallel-safe: os.Stdout is process-global.
 func captureProcessStdout(t *testing.T, fn func()) string {

internal/cli/entry_interrupt_test.go

@@ -0,0 +1,203 @@
+package cli //nolint:testpackage // shares runEntry and the argument constants
+
+import (
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"os/signal"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/adrg/xdg"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// stalledStoreConfig is hermeticConfig with an s3:// destination store
+// in place of the file:// one. The server behind it accepts any
+// credentials.
+const stalledStoreConfig = `age_recipients:
+  - age1278m9q7dp3chsh2dcy82qk27v047zywyvtxwnj4cvt0z65jw6a7q5dqhfj
+snapshots:
+  test:
+    paths:
+      - %s
+storage_url: s3://bucket?endpoint=%s&ssl=false
+s3:
+  access_key_id: key
+  secret_access_key: secret
+index_path: %s
+hostname: test-host
+`
+
+// interruptRepeat is how often interruptOnFirstRequest sends SIGINT.
+const interruptRepeat = 50 * time.Millisecond
+
+// TestEntryInterruptedRun sends SIGINT to the test process while a
+// command waits on the destination store, and checks that Entry returns
+// 130 and prints one line on stderr saying the run was interrupted. The
+// store is a local HTTP server that holds every request open, so the
+// command is always mid-operation when the signal arrives. The two
+// cases cover --cron and --json, which silence other output.
+//
+// Not parallel: it signals the process and replaces os.Args, os.Stdout,
+// os.Stderr and the xdg globals.
+//
+//nolint:paralleltest // signals the process and replaces process globals
+func TestEntryInterruptedRun(t *testing.T) {
+	for _, testCase := range []struct {
+		name string
+		args []string
+	}{
+		{
+			name: "snapshot create --cron",
+			args: []string{cmdSnapshot, cmdCreate, "--cron"},
+		},
+		{
+			name: "snapshot verify --json",
+			args: []string{cmdSnapshot, cmdVerify, someSnapshotID, flagJSON},
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			endpoint, requestArrived := startStalledStore(t)
+			configPath := writeStalledStoreConfig(t, endpoint)
+			interruptOnFirstRequest(t, requestArrived)
+
+			code, _, stderr := runEntry(t,
+				append([]string{flagConfig, configPath}, testCase.args...)...)
+
+			assert.Equal(t, 130, code)
+			assert.Equal(t, 1,
+				strings.Count(stderr, errInterrupted.Error()), stderr)
+		})
+	}
+}
+
+// interruptOnFirstRequest sends SIGINT to the test process every
+// interruptRepeat, from the first request to the destination store until
+// the test ends. One signal is not enough: the command can reach the
+// store before fx has started catching signals. The test catches SIGINT
+// too, so that a signal fx is not catching does not kill the test
+// binary.
+func interruptOnFirstRequest(t *testing.T, requestArrived <-chan struct{}) {
+	t.Helper()
+
+	self, err := os.FindProcess(os.Getpid())
+	require.NoError(t, err)
+
+	caught := make(chan os.Signal, 1)
+	signal.Notify(caught, os.Interrupt)
+
+	testEnded := make(chan struct{})
+	senderDone := make(chan struct{})
+
+	// Stop catching SIGINT only after the sender has returned. The sender
+	// waits for each SIGINT it sends to arrive on caught; one still on
+	// its way after signal.Stop would kill the test binary.
+	t.Cleanup(func() {
+		close(testEnded)
+		<-senderDone
+		signal.Stop(caught)
+	})
+
+	go func() {
+		defer close(senderDone)
+
+		select {
+		case <-requestArrived:
+		case <-testEnded:
+			return
+		}
+
+		ticker := time.NewTicker(interruptRepeat)
+		defer ticker.Stop()
+
+		for {
+			// Empty caught, so that the receive below waits for this
+			// SIGINT rather than an earlier one.
+			select {
+			case <-caught:
+			default:
+			}
+
+			sendErr := self.Signal(os.Interrupt)
+			if sendErr != nil {
+				t.Errorf("sending SIGINT: %v", sendErr)
+
+				return
+			}
+
+			<-caught
+
+			select {
+			case <-testEnded:
+				return
+			case <-ticker.C:
+			}
+		}
+	}()
+}
+
+// startStalledStore starts an HTTP server that never answers: each
+// request is held until the client gives up on it or the test ends.
+// It returns the server's host:port and a channel that receives a value
+// when the first request arrives.
+func startStalledStore(t *testing.T) (string, <-chan struct{}) {
+	t.Helper()
+
+	requestArrived := make(chan struct{}, 1)
+	release := make(chan struct{})
+
+	server := httptest.NewServer(http.HandlerFunc(
+		func(_ http.ResponseWriter, r *http.Request) {
+			select {
+			case requestArrived <- struct{}{}:
+			default:
+			}
+
+			select {
+			case <-r.Context().Done():
+			case <-release:
+			}
+		}))
+
+	// Cleanups run last-registered first, so release lets any held
+	// request return before Close waits for it.
+	t.Cleanup(server.Close)
+	t.Cleanup(func() { close(release) })
+
+	return server.Listener.Addr().String(), requestArrived
+}
+
+// writeStalledStoreConfig writes a config whose destination store is the
+// server at endpoint and whose snapshot source holds one small file, so
+// that `snapshot create` has a blob to upload. Returns the config path.
+func writeStalledStoreConfig(t *testing.T, endpoint string) string {
+	t.Helper()
+
+	dir := t.TempDir()
+	configPath := filepath.Join(dir, "config.yml")
+	sourceDir := filepath.Join(dir, "source")
+
+	require.NoError(t, os.Mkdir(sourceDir, 0o750))
+	require.NoError(t, os.WriteFile(filepath.Join(sourceDir, "file.txt"),
+		[]byte("contents"), 0o600))
+
+	contents := fmt.Sprintf(stalledStoreConfig,
+		sourceDir, endpoint, filepath.Join(dir, "index.sqlite"))
+
+	require.NoError(t,
+		os.WriteFile(configPath, []byte(contents), configFileMode))
+
+	// The PID lock lives under xdg.DataHome, which xdg resolves at
+	// package init; point it at the temp dir so the test neither
+	// touches nor collides with the real one.
+	t.Setenv("XDG_DATA_HOME", filepath.Join(dir, "data"))
+	xdg.Reload()
+	t.Cleanup(xdg.Reload)
+
+	return configPath
+}

internal/cli/entry_prune_json_test.go

@@ -98,25 +98,31 @@ func TestEntryPruneJSONStdoutIsExactlyOneDocument(t *testing.T) {
 	}
 }
 
-// writeHermeticPruneConfig builds a config over a temp directory and, if
-// seedStale is set, creates the index database up front with one
-// snapshot record that has no counterpart on the destination store.
-// Returns the config path.
+// writeHermeticPruneConfig builds a config over a temp directory with an
+// empty destination directory and, if seedStale is set, creates the
+// index database up front with one snapshot record that has no
+// counterpart on the destination store, and gives the destination the
+// marker a backup writes. Returns the config path.
 func writeHermeticPruneConfig(t *testing.T, seedStale bool) string {
 	t.Helper()
 
 	dir := t.TempDir()
 	configPath := filepath.Join(dir, "config.yml")
 	indexPath := filepath.Join(dir, "index.sqlite")
+	storeDir := filepath.Join(dir, "store")
 
 	contents := fmt.Sprintf(hermeticConfig,
 		filepath.Join(dir, "source"),
-		filepath.Join(dir, "store"),
+		storeDir,
 		indexPath)
 
 	require.NoError(t,
 		os.WriteFile(configPath, []byte(contents), configFileMode))
 
+	// prune fails on a destination directory that does not exist, so the
+	// empty store is created here rather than left to a first backup.
+	require.NoError(t, os.Mkdir(storeDir, 0o750))
+
 	// The PID lock lives under xdg.DataHome, which xdg resolves at
 	// package init; point it at the temp dir so the test neither
 	// touches nor collides with the real one.
@@ -126,6 +132,11 @@ func writeHermeticPruneConfig(t *testing.T, seedStale bool) string {
 
 	if seedStale {
 		seedStaleSnapshotRecord(t, indexPath)
+
+		// While the local index holds a snapshot, prune fails on a
+		// destination that answers that it has no marker.
+		require.NoError(t, os.WriteFile(
+			filepath.Join(storeDir, "vaultik-destination.txt"), nil, 0o600))
 	}
 
 	return configPath

internal/cli/entry_prune_quiet_test.go

@@ -0,0 +1,127 @@
+package cli //nolint:testpackage // shares runEntry and the hermetic config helpers
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// flagCron is used to build the argument vectors below.
+const flagCron = "--cron"
+
+// plantUnreferencedBlob writes a blob that no snapshot references to the
+// destination store at storeDir, dated long before any prune grace period,
+// so that a prune deletes it. Returns its path.
+func plantUnreferencedBlob(t *testing.T, storeDir string) string {
+	t.Helper()
+
+	hash := strings.Repeat("b", 64)
+	path := filepath.Join(storeDir, "blobs", hash[:2], hash[2:4], hash)
+
+	require.NoError(t, os.MkdirAll(filepath.Dir(path), 0o750))
+	require.NoError(t, os.WriteFile(path, []byte("unreferenced"), 0o600))
+
+	longAgo := time.Unix(0, 0)
+	require.NoError(t, os.Chtimes(path, longAgo, longAgo))
+
+	return path
+}
+
+// TestEntrySnapshotCreatePruneIsSilent checks that a successful
+// `snapshot create --prune` under --cron or --quiet prints nothing on
+// stdout or stderr, whether or not the prune finds a blob to delete.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotCreatePruneIsSilent(t *testing.T) {
+	for _, testCase := range []struct {
+		name      string
+		args      []string
+		plantBlob bool
+	}{
+		{
+			name: "cron, no blob to delete",
+			args: []string{cmdSnapshot, cmdCreate, flagCron, flagPrune},
+		},
+		{
+			name:      "cron, blob to delete",
+			args:      []string{cmdSnapshot, cmdCreate, flagCron, flagPrune},
+			plantBlob: true,
+		},
+		{
+			name: "cron with keep-newer-than",
+			args: []string{
+				cmdSnapshot, cmdCreate, flagCron, flagPrune,
+				flagKeepNewerThan, "4w",
+			},
+			plantBlob: true,
+		},
+		{
+			name:      "quiet",
+			args:      []string{flagQuiet, cmdSnapshot, cmdCreate, flagPrune},
+			plantBlob: true,
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			configPath, storeDir := writeBackupConfig(t)
+
+			var blob string
+			if testCase.plantBlob {
+				blob = plantUnreferencedBlob(t, storeDir)
+			}
+
+			code, stdout, stderr := runEntry(t,
+				append([]string{flagConfig, configPath}, testCase.args...)...)
+
+			assert.Equal(t, 0, code)
+			assert.Empty(t, stdout)
+			assert.Empty(t, stderr)
+
+			if testCase.plantBlob {
+				assert.NoFileExists(t, blob, "the prune must delete the blob")
+			}
+		})
+	}
+}
+
+// TestEntryPruneQuietIsSilent checks that a successful `prune --force`
+// under --quiet prints nothing on stdout or stderr, both when there is
+// nothing to remove and when it removes a stale local snapshot record
+// and an unreferenced blob.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntryPruneQuietIsSilent(t *testing.T) {
+	for _, testCase := range []struct {
+		name               string
+		staleRecordAndBlob bool
+	}{
+		{name: "nothing to remove"},
+		{name: "stale record and blob to remove", staleRecordAndBlob: true},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			configPath := writeHermeticPruneConfig(t, testCase.staleRecordAndBlob)
+
+			var blob string
+			if testCase.staleRecordAndBlob {
+				// writeHermeticPruneConfig puts the store beside the config.
+				blob = plantUnreferencedBlob(t,
+					filepath.Join(filepath.Dir(configPath), "store"))
+			}
+
+			code, stdout, stderr := runEntry(t,
+				flagConfig, configPath, flagQuiet, cmdPrune, flagForce)
+
+			assert.Equal(t, 0, code)
+			assert.Empty(t, stdout)
+			assert.Empty(t, stderr)
+
+			if testCase.staleRecordAndBlob {
+				assert.NoFileExists(t, blob, "the prune must delete the blob")
+			}
+		})
+	}
+}

internal/cli/entry_snapshot_create_index_and_lock_files_test.go

@@ -0,0 +1,69 @@
+package cli //nolint:testpackage // shares runEntry and the hermetic config helpers
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/adrg/xdg"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// TestEntrySnapshotCreateSkipsIndexAndLockFiles backs up a directory
+// that holds vaultik's index and PID file beside one other file, with no
+// exclude pattern set. The summary counts only the other file. See
+// https://git.eeqj.de/sneak/vaultik/issues/421.
+//
+// Not parallel: it replaces os.Args, os.Stdout, os.Stderr and the xdg
+// globals.
+func TestEntrySnapshotCreateSkipsIndexAndLockFiles(t *testing.T) {
+	for _, testCase := range []struct {
+		name string
+		// linked sets index_path to a symlink outside the backed-up
+		// directory that points to the index inside it.
+		linked bool
+	}{
+		{name: "index in the backed-up directory"},
+		{name: "index_path links into the backed-up directory", linked: true},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			dir := t.TempDir()
+			configPath := filepath.Join(dir, "config.yml")
+			sourceDir := filepath.Join(dir, "source")
+			storeDir := filepath.Join(dir, "store")
+
+			indexPath := filepath.Join(sourceDir, "index.sqlite")
+
+			if testCase.linked {
+				link := filepath.Join(dir, "index.sqlite")
+				require.NoError(t, os.Symlink(indexPath, link))
+
+				indexPath = link
+			}
+
+			contents := fmt.Sprintf(hermeticConfig,
+				sourceDir, storeDir, indexPath)
+			require.NoError(t,
+				os.WriteFile(configPath, []byte(contents), configFileMode))
+			require.NoError(t, os.Mkdir(storeDir, 0o750))
+			require.NoError(t, os.Mkdir(sourceDir, 0o750))
+			require.NoError(t, os.WriteFile(
+				filepath.Join(sourceDir, "file.txt"), []byte("hello"), 0o600))
+
+			// The PID file goes in a directory under XDG_DATA_HOME, which
+			// xdg reads at package init.
+			t.Setenv("XDG_DATA_HOME", filepath.Join(sourceDir, "data"))
+			xdg.Reload()
+			t.Cleanup(xdg.Reload)
+
+			code, stdout, _ := runEntry(t,
+				flagConfig, configPath, cmdSnapshot, cmdCreate)
+
+			require.Equal(t, 0, code)
+			assert.Contains(t, stdout,
+				"Files: 1 examined, 1 backed up, 0 unchanged.")
+		})
+	}
+}

internal/cli/entry_snapshot_create_prune_test.go

@@ -0,0 +1,185 @@
+package cli //nolint:testpackage // shares runEntry and the hermetic config helpers
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/database"
+)
+
+// Flags of `snapshot create` used to build argument vectors below.
+const (
+	flagPrune         = "--prune"
+	flagKeepNewerThan = "--keep-newer-than"
+)
+
+// TestEntrySnapshotCreateRejectsKeepNewerThanBeforeBackup checks that a
+// --keep-newer-than the post-backup prune could not use stops `snapshot
+// create` before the backup: the command fails and the destination
+// store stays empty. It used to be read only by the prune, after the
+// snapshot was uploaded, and without --prune it was ignored.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotCreateRejectsKeepNewerThanBeforeBackup(t *testing.T) {
+	for _, testCase := range []struct {
+		name string
+		args []string
+	}{
+		{
+			name: "decimal in a day count",
+			args: []string{flagPrune, flagKeepNewerThan, "1.5d"},
+		},
+		{
+			name: "unknown unit",
+			args: []string{flagPrune, flagKeepNewerThan, "4x"},
+		},
+		{
+			name: "without --prune",
+			args: []string{flagKeepNewerThan, "4w"},
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			configPath, storeDir := writeBackupConfig(t)
+
+			code, _, _ := runEntry(t, append(
+				[]string{flagConfig, configPath, cmdSnapshot, cmdCreate},
+				testCase.args...)...)
+
+			assert.Equal(t, 1, code)
+
+			stored, err := os.ReadDir(storeDir)
+			require.NoError(t, err)
+			assert.Empty(t, stored, "nothing may be uploaded")
+		})
+	}
+}
+
+// TestEntrySnapshotCreateReportsPostBackupPruneFailure makes the
+// post-backup prune fail after a good backup, by planting a manifest
+// that cannot be decoded, and checks that the failure is reported as the
+// prune's and not as a failed snapshot creation.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotCreateReportsPostBackupPruneFailure(t *testing.T) {
+	configPath, storeDir := writeBackupConfig(t)
+
+	// A remote key is 64 lowercase hex characters. The prune reads every
+	// manifest on the store and fails on one it cannot decode.
+	plantedDir := filepath.Join(storeDir, "metadata", strings.Repeat("a", 64))
+	require.NoError(t, os.MkdirAll(plantedDir, 0o750))
+	require.NoError(t, os.WriteFile(
+		filepath.Join(plantedDir, "manifest.json.zst"),
+		[]byte("not a valid manifest"), 0o600))
+
+	code, _, stderr := runEntry(t,
+		flagConfig, configPath, cmdSnapshot, cmdCreate, flagPrune)
+
+	assert.Equal(t, 1, code)
+	assert.Contains(t, stderr, "Backup completed, but post-backup prune failed")
+	assert.NotContains(t, stderr, "Snapshot creation failed")
+
+	snapshots, err := os.ReadDir(filepath.Join(storeDir, "metadata"))
+	require.NoError(t, err)
+	assert.Len(t, snapshots, 2,
+		"the new snapshot is on the store beside the planted one")
+}
+
+// TestEntrySnapshotCreatePruneExitsNonZeroWhenSourceIsEmpty backs up,
+// empties the source directory, as a volume that is not mounted leaves
+// its mount point, and runs `snapshot create --prune`. The command exits
+// non-zero and reports the backup as completed, not as a failed snapshot
+// creation. See https://git.eeqj.de/sneak/vaultik/issues/388.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotCreatePruneExitsNonZeroWhenSourceIsEmpty(t *testing.T) {
+	configPath, storeDir := writeBackupConfig(t)
+
+	code, _, _ := runEntry(t, flagConfig, configPath, cmdSnapshot, cmdCreate)
+	require.Equal(t, 0, code)
+
+	// writeBackupConfig puts the source directory beside the config file.
+	require.NoError(t, os.Remove(
+		filepath.Join(filepath.Dir(configPath), "source", "file.txt")))
+
+	code, _, stderr := runEntry(t,
+		flagConfig, configPath, cmdSnapshot, cmdCreate, flagPrune)
+
+	assert.Equal(t, 1, code)
+	assert.Contains(t, stderr,
+		"Backup completed, but 1 source path(s) were empty")
+	assert.NotContains(t, stderr, "Snapshot creation failed")
+
+	snapshots, err := os.ReadDir(filepath.Join(storeDir, "metadata"))
+	require.NoError(t, err)
+	assert.Len(t, snapshots, 2, "the older snapshot is kept")
+}
+
+// TestEntrySnapshotCreatePruneExitsNonZeroWhenMountPointIsMissing backs
+// up, records in the local index that this backup found the source
+// directory as a mount point, and runs `snapshot create --prune`, which
+// does not find it as one. The command exits non-zero, reports the
+// backup as completed, not as a failed snapshot creation, and keeps the
+// older snapshot. See https://git.eeqj.de/sneak/vaultik/issues/395.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotCreatePruneExitsNonZeroWhenMountPointIsMissing(
+	t *testing.T,
+) {
+	configPath, storeDir := writeBackupConfig(t)
+
+	code, _, _ := runEntry(t, flagConfig, configPath, cmdSnapshot, cmdCreate)
+	require.Equal(t, 0, code)
+
+	// writeBackupConfig puts the source directory and the local index
+	// beside the config file.
+	dir := filepath.Dir(configPath)
+	ctx := context.Background()
+
+	db, err := database.New(ctx, filepath.Join(dir, "index.sqlite"))
+	require.NoError(t, err)
+
+	repos := database.NewRepositories(db)
+
+	backedUp, err := repos.Snapshots.GetOnlySnapshot(ctx)
+	require.NoError(t, err)
+	require.NoError(t, repos.Snapshots.AddMountPoint(ctx, nil,
+		backedUp.ID.String(), filepath.Join(dir, "source")))
+	require.NoError(t, db.Close())
+
+	code, _, stderr := runEntry(t,
+		flagConfig, configPath, cmdSnapshot, cmdCreate, flagPrune)
+
+	assert.Equal(t, 1, code)
+	assert.Contains(t, stderr, "Backup completed, but 1 mount point(s) "+
+		"an earlier snapshot recorded were not found")
+	assert.NotContains(t, stderr, "Snapshot creation failed")
+
+	snapshots, err := os.ReadDir(filepath.Join(storeDir, "metadata"))
+	require.NoError(t, err)
+	assert.Len(t, snapshots, 2, "the older snapshot is kept")
+}
+
+// writeBackupConfig builds a hermetic config with one small file in its
+// source directory and an empty destination store directory. Returns the
+// config path and the store directory.
+func writeBackupConfig(t *testing.T) (string, string) {
+	t.Helper()
+
+	configPath := writeHermeticPruneConfig(t, false)
+
+	// writeHermeticPruneConfig puts the source and store directories
+	// beside the config file.
+	dir := filepath.Dir(configPath)
+	sourceDir := filepath.Join(dir, "source")
+
+	require.NoError(t, os.Mkdir(sourceDir, 0o750))
+	require.NoError(t, os.WriteFile(
+		filepath.Join(sourceDir, "file.txt"), []byte("hello"), 0o600))
+
+	return configPath, filepath.Join(dir, "store")
+}

internal/cli/entry_snapshot_create_skip_errors_test.go

@@ -0,0 +1,30 @@
+package cli //nolint:testpackage // shares runEntry and the hermetic config helpers
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// TestEntrySnapshotCreateSkipErrorsExitsNonZeroWhenItSkips backs up,
+// under --skip-errors, a source directory that does not exist, which the
+// walk skips. The command exits non-zero and reports the backup as
+// completed, not as a failed snapshot creation. See
+// https://git.eeqj.de/sneak/vaultik/issues/310.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotCreateSkipErrorsExitsNonZeroWhenItSkips(t *testing.T) {
+	// writeBackupConfig would create the source directory.
+	configPath := writeHermeticPruneConfig(t, false)
+
+	code, stdout, stderr := runEntry(t,
+		flagConfig, configPath, "--skip-errors", cmdSnapshot, cmdCreate)
+
+	assert.Equal(t, 1, code)
+	assert.Contains(t, stdout,
+		"Files: 1 examined, 0 backed up, 0 unchanged, 1 failed.")
+	assert.NotContains(t, stdout, "Finished successfully.")
+	assert.Contains(t, stderr,
+		"Backup completed, but 1 file(s) could not be read")
+	assert.NotContains(t, stderr, "Snapshot creation failed")
+}

internal/cli/entry_snapshot_purge_test.go

@@ -0,0 +1,75 @@
+package cli //nolint:testpackage // shares runEntry and the hermetic config helpers
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/snapshot"
+)
+
+// Words and flags of `snapshot purge` used to build argument vectors
+// below.
+const (
+	cmdPurge      = "purge"
+	flagOlderThan = "--older-than"
+	flagForce     = "--force"
+)
+
+// TestEntrySnapshotPurgeKeepsMetadataWhenLocalIndexDeleteFails makes
+// the local index refuse to delete an old snapshot and checks that
+// `snapshot purge` exits non-zero and leaves that snapshot's metadata on
+// the destination store. It used to log the error, delete the metadata
+// and exit 0.
+//
+// `snapshot create --prune` runs the same purge, but cannot be driven
+// this way: the refusal also breaks the backup's metadata export, which
+// deletes the other snapshots from its copy of the local index.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotPurgeKeepsMetadataWhenLocalIndexDeleteFails(t *testing.T) {
+	configPath := writeHermeticPruneConfig(t, false)
+	dir := filepath.Dir(configPath)
+	indexPath := filepath.Join(dir, "index.sqlite")
+
+	// seedStaleSnapshotRecord writes an old snapshot to the local index
+	// only; the manifest planted here puts it on the destination store
+	// too, so the purge selects it rather than dropping it as stale.
+	seedStaleSnapshotRecord(t, indexPath)
+
+	// While the local index holds a snapshot, purge fails on a
+	// destination that answers that it has no marker.
+	require.NoError(t, os.WriteFile(
+		filepath.Join(dir, "store", "vaultik-destination.txt"), nil, 0o600))
+
+	manifestDir := filepath.Join(dir, "store", "metadata",
+		snapshot.RemoteSnapshotKey(stalePruneSnapshotID))
+	require.NoError(t, os.MkdirAll(manifestDir, 0o750))
+
+	manifestPath := filepath.Join(manifestDir, "manifest.json.zst")
+	require.NoError(t, os.WriteFile(manifestPath, []byte("stub"), 0o600))
+
+	ctx := context.Background()
+
+	db, err := database.New(ctx, indexPath)
+	require.NoError(t, err)
+
+	_, err = db.Conn().ExecContext(ctx, `
+		CREATE TRIGGER refuse_snapshot_delete BEFORE DELETE ON snapshots
+		BEGIN SELECT RAISE(ABORT, 'simulated index error'); END`)
+	require.NoError(t, err)
+	require.NoError(t, db.Close())
+
+	code, _, stderr := runEntry(t, flagConfig, configPath,
+		cmdSnapshot, cmdPurge, flagOlderThan, "30d", flagForce)
+
+	assert.Equal(t, 1, code)
+	assert.Contains(t, stderr, "simulated index error",
+		"the purge must fail at the local-index delete")
+	assert.FileExists(t, manifestPath,
+		"the snapshot is still in the local index, so its metadata must stay")
+}

internal/cli/entry_status_test.go

@@ -15,8 +15,8 @@ import (
 // run, so a failing command must come back with a non-zero code rather
 // than ending the process here.
 //
-// Stdout is captured only to keep the banner and command output off the
-// test log; the assertion is on the returned code.
+// Stdout and stderr are captured only to keep the banner and command
+// output off the test log; the assertion is on the returned code.
 //
 //nolint:paralleltest // replaces os.Args and rootFlags
 func TestEntryReturnsStatusCode(t *testing.T) {
@@ -50,7 +50,7 @@ func TestEntryReturnsStatusCode(t *testing.T) {
 
 			var code int
 
-			_ = captureProcessStdout(t, func() { code = Entry() })
+			_, _ = captureProcessStdoutAndStderr(t, func() { code = Entry() })
 
 			assert.Equal(t, testCase.want, code)
 		})

internal/cli/entry_stdout_stderr_test.go

@@ -0,0 +1,200 @@
+package cli //nolint:testpackage // shares hermeticConfig and the capture helpers
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"log/slog"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/adrg/xdg"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/database"
+)
+
+// TestEntryCompletionStdoutIsTheScript runs `vaultik completion bash`,
+// whose stdout the README tells the user to source. The script has to
+// start on the first line.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout and os.Stderr
+func TestEntryCompletionStdoutIsTheScript(t *testing.T) {
+	code, stdout, _ := runEntry(t, "completion", "bash")
+
+	require.Equal(t, 0, code)
+
+	firstLine, _, _ := strings.Cut(stdout, "\n")
+	assert.True(t, strings.HasPrefix(firstLine, "# bash completion"),
+		"the first line of stdout must be the script's, got %q", firstLine)
+}
+
+// TestEntryConfigGetStdoutIsTheValue runs `vaultik config get`, whose
+// stdout a script reads as the value and nothing else.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout and os.Stderr
+func TestEntryConfigGetStdoutIsTheValue(t *testing.T) {
+	configPath := filepath.Join(t.TempDir(), "config.yml")
+	require.NoError(t, os.WriteFile(configPath,
+		[]byte("hostname: test-host\n"), configFileMode))
+
+	code, stdout, _ := runEntry(t,
+		flagConfig, configPath, "config", "get", "hostname")
+
+	require.Equal(t, 0, code)
+	assert.Equal(t, "test-host\n", stdout)
+}
+
+// TestEntryJSONFailureIsReportedOnStderr runs each --json command that
+// writes no document when it fails, against a destination it cannot
+// use. The error must reach stderr, and stdout must stay empty.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntryJSONFailureIsReportedOnStderr(t *testing.T) {
+	for _, testCase := range []struct {
+		name         string
+		args         []string
+		wantOnStderr string
+	}{
+		{
+			name:         "remote info",
+			args:         []string{cmdRemote, cmdInfo, flagJSON},
+			wantOnStderr: "Failed to get remote info",
+		},
+		{
+			name:         "prune",
+			args:         []string{cmdPrune, flagJSON},
+			wantOnStderr: "Prune failed",
+		},
+		{
+			name:         "snapshot remove",
+			args:         []string{cmdSnapshot, cmdRemove, someSnapshotID, flagJSON},
+			wantOnStderr: "Failed to remove snapshot",
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			configPath := writeUnusableDestinationConfig(t)
+
+			code, stdout, stderr := runEntry(t,
+				append([]string{flagConfig, configPath}, testCase.args...)...)
+
+			assert.Equal(t, 1, code)
+			assert.Empty(t, stdout,
+				"a failed --json command has no document to write")
+			assert.Contains(t, stderr, testCase.wantOnStderr,
+				"the failure must be reported on stderr")
+		})
+	}
+}
+
+// TestEntrySnapshotRemoveJSONWarningIsOnStderr runs `snapshot remove
+// --json` on a snapshot in the local index, against a destination
+// directory that does not exist. The command removes the snapshot from
+// the local index and still exits 0. Its stdout must hold the document
+// alone, with the warning about the destination store on stderr: the
+// command to run again once it is reachable, and the snapshot's ID in
+// the record's snapshot_id field.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotRemoveJSONWarningIsOnStderr(t *testing.T) {
+	configPath, indexPath := writeMissingDestinationConfig(t)
+	seedStaleSnapshotRecord(t, indexPath)
+
+	code, stdout, stderr := runEntry(t, flagConfig, configPath,
+		cmdSnapshot, cmdRemove, stalePruneSnapshotID, flagJSON)
+
+	require.Equal(t, 0, code)
+	requireExactlyOneJSONDocument(t, stdout)
+
+	// stderr is a pipe here, so the logger writes one JSON record a line.
+	var warning map[string]any
+
+	for line := range strings.Lines(stderr) {
+		if strings.Contains(line,
+			"Could not remove snapshot metadata from remote storage") {
+			require.NoError(t, json.Unmarshal([]byte(line), &warning))
+		}
+	}
+
+	require.NotNil(t, warning, "the warning must reach stderr")
+	assert.Contains(t, warning[slog.MessageKey],
+		"run 'vaultik snapshot remove' with the snapshot's ID again")
+	assert.Equal(t, stalePruneSnapshotID, warning["snapshot_id"])
+}
+
+// writeMissingDestinationConfig builds a config whose destination
+// directory does not exist. Returns the config path and the path of
+// its local index, which is not created here.
+func writeMissingDestinationConfig(t *testing.T) (string, string) {
+	t.Helper()
+
+	dir := t.TempDir()
+	configPath := filepath.Join(dir, "config.yml")
+	indexPath := filepath.Join(dir, "index.sqlite")
+
+	contents := fmt.Sprintf(hermeticConfig,
+		filepath.Join(dir, "source"),
+		filepath.Join(dir, "missing-store"),
+		indexPath)
+
+	require.NoError(t,
+		os.WriteFile(configPath, []byte(contents), configFileMode))
+
+	// The PID lock lives under xdg.DataHome, which xdg resolves at
+	// package init; point it at the temp dir so the test neither
+	// touches nor collides with the real one.
+	t.Setenv("XDG_DATA_HOME", filepath.Join(dir, "data"))
+	xdg.Reload()
+	t.Cleanup(xdg.Reload)
+
+	return configPath, indexPath
+}
+
+// writeUnusableDestinationConfig builds a config whose destination
+// directory does not exist, which fails `remote info`, and whose local
+// index is bound to another destination, which fails `prune` and
+// `snapshot remove` (a missing destination alone only makes `snapshot
+// remove` warn). Returns the config path.
+func writeUnusableDestinationConfig(t *testing.T) string {
+	t.Helper()
+
+	configPath, indexPath := writeMissingDestinationConfig(t)
+
+	ctx := context.Background()
+
+	db, err := database.New(ctx, indexPath)
+	require.NoError(t, err)
+
+	defer func() { require.NoError(t, db.Close()) }()
+
+	require.NoError(t, database.NewRepositories(db).LocalMeta.Set(ctx,
+		database.LocalMetaKeyStorageURL, "file://"+t.TempDir()))
+
+	return configPath
+}
+
+// runEntry runs Entry with args after the program name and returns its
+// exit code and what it wrote to stdout and stderr.
+//
+// Not parallel-safe: it replaces os.Args, os.Stdout and os.Stderr.
+func runEntry(t *testing.T, args ...string) (int, string, string) {
+	t.Helper()
+
+	previousArgs := os.Args
+
+	t.Cleanup(func() {
+		os.Args = previousArgs
+		rootFlags = RootFlags{}
+	})
+
+	os.Args = append([]string{programName}, args...)
+
+	var code int
+
+	stdout, stderr := captureProcessStdoutAndStderr(t,
+		func() { code = Entry() })
+
+	return code, stdout, stderr
+}

internal/cli/entry_unknown_subcommand_test.go

@@ -0,0 +1,43 @@
+package cli //nolint:testpackage // shares runEntry and the command-word constants
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// TestEntryGroupRejectsUnknownSubcommand runs each group command with a
+// mistyped subcommand, as a crontab line might carry one. It must fail
+// with exit 1 and name the word, as `vaultik bogus` does at the top
+// level, instead of printing the group's help and exiting 0 having done
+// nothing. The group name alone still prints its help.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout and os.Stderr
+func TestEntryGroupRejectsUnknownSubcommand(t *testing.T) {
+	for _, testCase := range []struct {
+		group    string
+		mistyped string
+	}{
+		{group: cmdSnapshot, mistyped: "craete"},
+		{group: "config", mistyped: "inti"},
+		{group: cmdRemote, mistyped: "ifno"},
+		{group: "database", mistyped: "delte"},
+		{group: "completion", mistyped: "nosuchshell"},
+	} {
+		t.Run(testCase.group, func(t *testing.T) {
+			code, stdout, stderr := runEntry(t,
+				testCase.group, testCase.mistyped)
+
+			assert.Equal(t, 1, code)
+			assert.Empty(t, stdout)
+			assert.Contains(t, stderr,
+				`unknown command "`+testCase.mistyped+`"`)
+
+			code, stdout, _ = runEntry(t, testCase.group)
+
+			assert.Equal(t, 0, code)
+			assert.Contains(t, stdout,
+				programName+" "+testCase.group+" [command]")
+		})
+	}
+}

internal/cli/json_help_test.go

@@ -0,0 +1,24 @@
+package cli //nolint:testpackage // exercises the unexported command constructor
+
+import (
+	"testing"
+
+	"github.com/spf13/cobra"
+	"github.com/stretchr/testify/assert"
+)
+
+// TestJSONHelpSaysConfirmationPromptIsSkipped checks that the --json help
+// of `snapshot remove` and `prune` says the flag skips the confirmation
+// prompt. Both delete without asking under --json, since a prompt on
+// stdout would break the JSON document.
+func TestJSONHelpSaysConfirmationPromptIsSkipped(t *testing.T) {
+	t.Parallel()
+
+	for _, cmd := range []*cobra.Command{
+		newSnapshotRemoveCommand(),
+		NewPruneCommand(),
+	} {
+		assert.Contains(t, cmd.Flags().Lookup("json").Usage,
+			"skips the confirmation prompt", cmd.Name())
+	}
+}

internal/cli/prune.go

@@ -13,18 +13,30 @@ func NewPruneCommand() *cobra.Command {
 	cmd := &cobra.Command{
 		Use:   "prune",
 		Short: "Tidy local database and remote storage",
-		Long: `Removes orphaned data from both the local index database and
-unreferenced blobs from the backup destination store.
+		Long: `Removes orphaned data from the local index database, and
+unreferenced blobs and directories under metadata/ with no manifest
+from the backup destination store.
 
 Local cleanup drops incomplete snapshots and any files, chunks, or
 blobs no longer referenced by a completed snapshot. Remote cleanup
 scans every snapshot manifest in the destination store, builds the
 set of still-referenced blob hashes, and deletes any blob not in that
-set.
+set that is older than prune_grace_period (default 24h). A newer one
+is kept: it may belong to a backup still running, on this host or
+another sharing the destination store, that has not yet uploaded its
+manifest. The grace period does not protect a blob such a backup
+reuses from an earlier run that stopped before its manifest: that
+blob keeps the time of its first upload, so it can be deleted.
+Remote cleanup also deletes each directory under metadata/ with no
+manifest, as a backup that stopped before its manifest leaves, once
+all its objects are older than prune_grace_period and its snapshot is
+not in the local index.
 
-Snapshot create --prune and snapshot remove run the same cleanup
-automatically; this command is the manual entry point for the same
-work (e.g. after a crashed backup or to reclaim storage).`,
+Snapshot create --prune runs the same cleanup automatically; this
+command is the manual entry point for the same work (e.g. after a
+crashed backup or to reclaim storage). Snapshot remove leaves blobs in
+place; run this command afterwards to delete the ones no longer
+referenced.`,
 		Args: cobra.NoArgs,
 		RunE: func(cmd *cobra.Command, _ []string) error {
 			// Use unified config resolution
@@ -48,10 +60,6 @@ work (e.g. after a crashed backup or to reclaim storage).`,
 			}, func(v *vaultik.Vaultik) error {
 				return v.Prune(opts)
 			}, func(err error) {
-				if opts.JSON {
-					return
-				}
-
 				log.Error("Prune operation failed", "error", err)
 				ReportErrorf("Prune failed: %v", err)
 			})
@@ -59,7 +67,8 @@ work (e.g. after a crashed backup or to reclaim storage).`,
 	}
 
 	cmd.Flags().BoolVar(&opts.Force, "force", false, "Skip confirmation prompt")
-	cmd.Flags().BoolVar(&opts.JSON, "json", false, "Output pruning stats as JSON")
+	cmd.Flags().BoolVar(&opts.JSON, "json", false,
+		"Output pruning stats as JSON; skips the confirmation prompt, as --force does")
 
 	return cmd
 }

internal/cli/remote.go

@@ -18,6 +18,12 @@ func NewRemoteCommand() *cobra.Command {
 		Use:   "remote",
 		Short: "Remote storage management commands",
 		Long:  "Commands for inspecting and managing remote storage",
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	// Add subcommands
@@ -45,7 +51,7 @@ This is destructive and irreversible. Requires --force.`,
 				return errNukeNeedsForce
 			}
 
-			return runVaultikApp(cmd, mutating, false, false, "Remote nuke failed",
+			return runVaultikApp(cmd, mutating, false, "Remote nuke failed",
 				func(v *vaultik.Vaultik) error {
 					return v.NukeRemote(true)
 				})
@@ -92,10 +98,6 @@ func newRemoteInfoCommand() *cobra.Command {
 			}, func(v *vaultik.Vaultik) error {
 				return v.RemoteInfo(jsonOutput)
 			}, func(err error) {
-				if jsonOutput {
-					return
-				}
-
 				log.Error("Failed to get remote info", "error", err)
 				ReportErrorf("Failed to get remote info: %v", err)
 			})

internal/cli/root.go

@@ -60,7 +60,8 @@ on the source system.`,
 	cmd.PersistentFlags().BoolVar(&rootFlags.SkipErrors, "skip-errors", false,
 		"Skip files that cannot be read when creating a snapshot, or "+
 			"that cannot be restored when restoring, instead of aborting "+
-			"(packing and storage errors still abort)")
+			"(packing and storage errors while creating a snapshot still "+
+			"abort)")
 
 	// Add subcommands
 	cmd.AddCommand(
@@ -73,6 +74,21 @@ on the source system.`,
 		NewDatabaseCommand(),
 	)
 
+	// cobra adds the completion command only when Execute runs, and with
+	// no Run, so it skipped its own cobra.NoArgs and a mistyped shell name
+	// printed help and exited 0. Adding it here lets it have a Run. Its
+	// scripts go to os.Stdout as it is now: a later SetOut does not
+	// redirect them.
+	cmd.InitDefaultCompletionCmd()
+
+	for _, sub := range cmd.Commands() {
+		if sub.Name() == "completion" {
+			sub.Run = func(completion *cobra.Command, _ []string) {
+				_ = completion.Help()
+			}
+		}
+	}
+
 	return cmd
 }
 

internal/cli/snapshot.go

@@ -41,6 +41,12 @@ func NewSnapshotCommand() *cobra.Command {
 		Use:   "snapshot",
 		Short: "Snapshot management commands",
 		Long:  "Commands for creating, listing, and managing snapshots",
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	// Add subcommands
@@ -66,10 +72,16 @@ func newSnapshotCreateCommand() *cobra.Command {
 If snapshot names are provided, only those snapshots are created.
 If no names are provided, all configured snapshots are created.
 
-Config is located at /etc/vaultik/config.yml by default, but can be overridden by
-specifying a path using --config or by setting VAULTIK_CONFIG to a path.`,
+The config is read from the path given by --config or VAULTIK_CONFIG;
+otherwise from the platform config directory (~/.config/vaultik/config.yml
+on Linux), then /etc/vaultik/config.yml.`,
 		Args: cobra.ArbitraryArgs,
 		RunE: func(cmd *cobra.Command, args []string) error {
+			err := opts.Validate()
+			if err != nil {
+				return err
+			}
+
 			// Pass snapshot names from args
 			opts.Snapshots = args
 			// --skip-errors is a global flag on the root command.
@@ -96,6 +108,22 @@ specifying a path using --config or by setting VAULTIK_CONFIG to a path.`,
 			}, func(v *vaultik.Vaultik) error {
 				return v.CreateSnapshot(opts)
 			}, func(err error) {
+				if errors.Is(err, vaultik.ErrPostBackupPruneFailed) {
+					log.Error("Post-backup prune failed", "error", err)
+					ReportErrorf("Backup completed, but %v", err)
+
+					return
+				}
+
+				if errors.Is(err, vaultik.ErrFilesUnreadable) ||
+					errors.Is(err, vaultik.ErrSourcePathEmpty) ||
+					errors.Is(err, vaultik.ErrVolumeNotMounted) {
+					log.Error("Backup completed with an error", "error", err)
+					ReportErrorf("Backup completed, but %v", err)
+
+					return
+				}
+
 				log.Error("Snapshot creation failed", "error", err)
 				ReportErrorf("Snapshot creation failed: %v", err)
 			})
@@ -105,8 +133,13 @@ specifying a path using --config or by setting VAULTIK_CONFIG to a path.`,
 	cmd.Flags().BoolVar(&opts.Cron, "cron", false,
 		"Run in cron mode (silent unless warning or error)")
 	cmd.Flags().BoolVar(&opts.Prune, "prune", false,
-		"After backup, drop older snapshots of the same name and remove "+
-			"orphaned blobs")
+		"After backup, drop older snapshots of the same name, remove "+
+			"orphaned blobs older than prune_grace_period, and delete each "+
+			"directory under metadata/ with no manifest once all its "+
+			"objects are older than prune_grace_period and its snapshot is "+
+			"not in the local index (not after a backup that skipped "+
+			"unreadable files, found a configured path empty, or did not "+
+			"find a mount point an earlier snapshot found)")
 	cmd.Flags().StringVar(&opts.KeepNewerThan, "keep-newer-than", "",
 		"With --prune: keep snapshots newer than this duration "+
 			"(e.g. 30d, 4w, 6mo, 1y; m is minutes, mo is months) "+
@@ -126,7 +159,7 @@ func newSnapshotListCommand() *cobra.Command {
 		Long:    "Lists all snapshots with their ID, timestamp, and compressed size",
 		Args:    cobra.NoArgs,
 		RunE: func(cmd *cobra.Command, _ []string) error {
-			return runVaultikApp(cmd, readOnly, false, false,
+			return runVaultikApp(cmd, readOnly, false,
 				"Failed to list snapshots",
 				func(v *vaultik.Vaultik) error {
 					return v.ListSnapshots(jsonOutput)
@@ -162,7 +195,7 @@ restrict the operation to specific snapshot names.`,
 				return errPurgeCriteriaBoth
 			}
 
-			return runVaultikApp(cmd, mutating, false, false,
+			return runVaultikApp(cmd, mutating, false,
 				"Failed to purge snapshots",
 				func(v *vaultik.Vaultik) error {
 					return v.PurgeSnapshotsWithOptions(opts)
@@ -251,21 +284,29 @@ By default, this removes the snapshot from the local index database and
 strips the snapshot's metadata from the backup destination store. Blobs
 are NOT touched: deleting them requires reading every remaining remote
 manifest (the destination store may hold snapshots this host doesn't
-know about), which is what 'vaultik prune' does. On success the command
-prints the exact 'vaultik prune' invocation to run as a follow-up.
+know about), which is what 'vaultik prune' does. The command prints the
+'vaultik prune' invocation to run as a follow-up only when it removed the
+snapshot's metadata from the destination store, and never under --json.
+
+An ID that names no snapshot in the local index or on the destination
+store is an error.
 
 Use --local-only to skip the remote half (e.g. when you want to forget a
-snapshot locally without touching the destination store).
+snapshot locally without touching the destination store); the snapshot
+must then be in the local index.
 
-If the remote is unreachable, the local-database removal still completes
-and a warning is emitted; rerun 'vaultik prune' once the destination store
-is reachable to finish remote cleanup.
+If removing the snapshot's metadata from the destination store fails (for
+example because the remote is unreachable), the local-database removal
+still completes and a warning is emitted; run 'vaultik snapshot remove
+<snapshot-id>' again once the destination store is reachable to remove
+the snapshot's metadata from it ('vaultik prune' does not). For a
+snapshot that is only on the destination store, that failure is an error.
 
 To wipe the entire destination store and start over, use 'vaultik remote
 nuke --force' — it is the single supported entry point for that.`,
 		Args: requireSnapshotIDArg,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			return runVaultikApp(cmd, mutating, opts.JSON, opts.JSON,
+			return runVaultikApp(cmd, mutating, opts.JSON,
 				"Failed to remove snapshot",
 				func(v *vaultik.Vaultik) error {
 					_, err := v.RemoveSnapshot(args[0], opts)
@@ -278,7 +319,8 @@ nuke --force' — it is the single supported entry point for that.`,
 	cmd.Flags().BoolVarP(&opts.Force, "force", "f", false, "Skip confirmation prompt")
 	cmd.Flags().BoolVar(&opts.DryRun, "dry-run", false,
 		"Show what would be removed without removing")
-	cmd.Flags().BoolVar(&opts.JSON, "json", false, "Output result as JSON")
+	cmd.Flags().BoolVar(&opts.JSON, "json", false,
+		"Output result as JSON; skips the confirmation prompt, as --force does")
 	cmd.Flags().BoolVar(&opts.LocalOnly, "local-only", false,
 		"Skip remote cleanup; only touch the local index")
 

internal/cli/snapshot_restore.go

@@ -30,6 +30,8 @@ This command will restore files from the specified snapshot to the
 target directory.
 If no paths are specified, all files are restored.
 If paths are specified, only matching files/directories are restored.
+A path that matches nothing in the snapshot is named in a warning, and
+the command exits non-zero after restoring the paths that match.
 
 The snapshot may be named by its ID or, when restoring on a host with no
 local index, by the remote key that 'snapshot list' prints for a

internal/cli/version.go

@@ -37,13 +37,13 @@ func writeVersion(out *ui.Writer) {
 	w := out.Out()
 
 	_, _ = fmt.Fprintf(w, "vaultik %s\n", globals.Version)
-	_, _ = fmt.Fprintf(w, "  commit:     %s\n", globals.Commit)
-	_, _ = fmt.Fprintf(w, "  build date: %s\n", globals.CommitDate)
-	_, _ = fmt.Fprintf(w, "  go:         %s\n", runtime.Version())
-	_, _ = fmt.Fprintf(w, "  os/arch:    %s/%s\n", runtime.GOOS, runtime.GOARCH)
-	_, _ = fmt.Fprintf(w, "  author:     %s\n", globals.Author)
-	_, _ = fmt.Fprintf(w, "  homepage:   %s\n", globals.Homepage)
-	_, _ = fmt.Fprintf(w, "  license:    %s\n", globals.License)
+	_, _ = fmt.Fprintf(w, "  commit:      %s\n", globals.Commit)
+	_, _ = fmt.Fprintf(w, "  commit date: %s\n", globals.CommitDate)
+	_, _ = fmt.Fprintf(w, "  go:          %s\n", runtime.Version())
+	_, _ = fmt.Fprintf(w, "  os/arch:     %s/%s\n", runtime.GOOS, runtime.GOARCH)
+	_, _ = fmt.Fprintf(w, "  author:      %s\n", globals.Author)
+	_, _ = fmt.Fprintf(w, "  homepage:    %s\n", globals.Homepage)
+	_, _ = fmt.Fprintf(w, "  license:     %s\n", globals.License)
 
 	if globals.IsDevVersion(globals.Version) {
 		_, _ = fmt.Fprintln(w)

internal/cli/version_test.go

@@ -2,6 +2,7 @@ package cli_test
 
 import (
 	"bytes"
+	"runtime"
 	"strings"
 	"testing"
 
@@ -49,6 +50,42 @@ func TestVersionCommandReportsBuildVersion(t *testing.T) {
 	}
 }
 
+// TestVersionCommandReportLines checks the label, padding and value of
+// every line of the report. Commit and CommitDate get different values,
+// so a line that prints the wrong one fails.
+//
+//nolint:paralleltest // sets globals.Commit and globals.CommitDate
+func TestVersionCommandReportLines(t *testing.T) {
+	const (
+		commit     = "abc1234"
+		commitDate = "2026-11-09"
+	)
+
+	previousCommit, previousCommitDate := globals.Commit, globals.CommitDate
+
+	t.Cleanup(func() {
+		globals.Commit = previousCommit
+		globals.CommitDate = previousCommitDate
+	})
+
+	globals.Commit = commit
+	globals.CommitDate = commitDate
+
+	want := "vaultik " + globals.Version + "\n" +
+		"  commit:      " + commit + "\n" +
+		"  commit date: " + commitDate + "\n" +
+		"  go:          " + runtime.Version() + "\n" +
+		"  os/arch:     " + runtime.GOOS + "/" + runtime.GOARCH + "\n" +
+		"  author:      " + globals.Author + "\n" +
+		"  homepage:    " + globals.Homepage + "\n" +
+		"  license:     " + globals.License + "\n"
+
+	out := runVersionCommand(t)
+	if !strings.HasPrefix(out, want) {
+		t.Errorf("report:\n%s\nwant it to start with:\n%s", out, want)
+	}
+}
+
 // TestVersionCommandFlagsDevelopmentBuild is the regression test for
 // the thing this command exists to prevent: a build that is not a
 // release must say so. The notice used to be gated on the version

internal/config/config.go

@@ -4,12 +4,15 @@
 package config
 
 import (
+	"bytes"
 	"errors"
 	"fmt"
 	"os"
 	"path/filepath"
+	"reflect"
 	"sort"
 	"strings"
+	"time"
 
 	"filippo.io/age"
 	"git.eeqj.de/sneak/smartconfig"
@@ -29,36 +32,49 @@ const appName = "vaultik"
 //nolint:gosec // G101: marker for detecting a pasted secret key, not a credential
 const secretKeyPrefix = "AGE-SECRET-KEY-"
 
-// Defaults and validation bounds for tunable settings.
+// Defaults and validation bounds for tunable settings. MB and GB are
+// 1000-based, as the size parser reads them; the S3 part sizes are S3's own
+// limits, which are 1024-based (MiB, GiB).
 const (
-	defaultBlobSizeLimit    = Size(10 * 1024 * 1024 * 1024) // 10GB
-	defaultChunkSize        = Size(10 * 1024 * 1024)        // 10MB
-	defaultS3PartSize       = Size(5 * 1024 * 1024)         // 5MB
+	defaultBlobSizeLimit    = Size(10 * 1000 * 1000 * 1000) // 10GB
+	defaultChunkSize        = Size(10 * 1000 * 1000)        // 10MB
+	defaultS3PartSize       = Size(5 * 1024 * 1024)         // 5MiB
 	defaultCompressionLevel = 3
-	minChunkSize            = 1024 * 1024 // 1MB
+	defaultPruneGracePeriod = 24 * time.Hour
+	minChunkSize            = 1000 * 1000 // 1MB
 	minCompressionLevel     = 1
 	maxCompressionLevel     = 19
+	// S3 accepts a multipart upload part from 5MiB to 5GiB.
+	minS3PartSize = 5 * 1024 * 1024
+	maxS3PartSize = 5 * 1024 * 1024 * 1024
 )
 
 // Sentinel validation errors.
 var (
-	errNoConfigPath    = errors.New("config path not provided")
-	errNoAgeRecipients = errors.New(
-		"at least one age_recipient is required (generate with: age-keygen)")
+	errNoConfigPath         = errors.New("config path not provided")
 	errRecipientIsSecretKey = errors.New(
 		"an age secret key was given where a public key (age1...) belongs")
 	errRecipientNotX25519 = errors.New(
 		"not a valid recipient; only X25519 age1... public keys are supported")
 	errNoSnapshots = errors.New(
-		"at least one snapshot must be configured (see config.example.yml)")
+		"at least one snapshot must be configured (see configs/config.example.yml)")
 	errSnapshotNoPaths   = errors.New("snapshot must have at least one path")
 	errChunkSizeTooSmall = errors.New("chunk_size must be at least 1MB")
 	errBlobSizeTooSmall  = errors.New(
 		"blob_size_limit must be at least the largest chunk the chunker can " +
 			"emit (chunk_size times the FastCDC size spread)")
 	errBadCompression   = errors.New("compression_level must be between 1 and 19")
+	errBadS3PartSize    = errors.New("s3.part_size must be between 5MiB and 5GiB")
 	errBadStorageScheme = errors.New(
 		"storage_url must start with s3://, file://, or rclone://")
+	errFileURLNotAbsolute = errors.New(
+		"a file:// storage_url needs an absolute path, " +
+			"as in file:///path/to/backup")
+	errIndexPathNotAbsolute = errors.New(
+		"index_path and VAULTIK_INDEX_PATH need an absolute path, " +
+			"as in ~/.local/share/vaultik/index.sqlite")
+	errNegativePruneGracePeriod = errors.New(
+		"prune_grace_period must not be negative")
 	errStorageNotConfigured = errors.New(
 		"storage not configured; set storage_url or provide s3.endpoint + " +
 			"s3.bucket + credentials")
@@ -179,6 +195,15 @@ type Config struct {
 	Snapshots        map[string]SnapshotConfig `yaml:"snapshots"`
 	CompressionLevel int                       `yaml:"compression_level"`
 
+	// PruneGracePeriod is how old a blob no manifest lists must be before
+	// a prune deletes it, and how old every object in a directory under
+	// metadata/ with no manifest must be before a prune deletes that
+	// directory. A backup writes its manifest only once it has uploaded
+	// all its blobs and its db.zst.age, so a younger blob or directory may
+	// belong to a backup still running, on this host or another sharing
+	// the destination store.
+	PruneGracePeriod time.Duration `yaml:"prune_grace_period"`
+
 	// StorageURL specifies the storage backend using a URL format.
 	// Takes precedence over S3Config if set.
 	// Supported formats:
@@ -228,24 +253,84 @@ func New(path Path) (*Config, error) {
 	return cfg, nil
 }
 
-// Load reads and parses the configuration file from the specified path.
-// It applies default values for optional fields, performs environment variable
-// substitution using smartconfig, and validates the configuration.
-// The configuration file should be in YAML format. Returns an error if the file
-// cannot be read, parsed, or if validation fails.
+// Load reads the configuration file at path, parses it with Parse, warns
+// when others can read the file, and validates the configuration.
+// Returns an error if the file cannot be read, parsed, or if validation
+// fails.
 func Load(path string) (*Config, error) {
-	// Load config using smartconfig for interpolation
-	sc, err := smartconfig.NewFromConfigPath(path)
+	data, err := os.ReadFile(path) //nolint:gosec // G304: config path is operator-supplied
 	if err != nil {
 		return nil, fmt.Errorf("failed to load config file: %w", err)
 	}
 
+	cfg, err := Parse(data)
+	if err != nil {
+		return nil, err
+	}
+
+	// Check config file permissions (warn if world or group readable)
+	//nolint:gosec // G703: config path is operator-supplied by design
+	info, statErr := os.Stat(path)
+	if statErr == nil {
+		mode := info.Mode().Perm()
+		if mode&0044 != 0 { // group or world readable
+			log.Warn(cfg.readableByOthersWarning(),
+				"path", path,
+				"mode", fmt.Sprintf("%04o", mode),
+				"recommendation", "chmod 600 "+path)
+		}
+	}
+
+	err = cfg.Validate()
+	if err != nil {
+		return nil, fmt.Errorf("invalid config: %w", err)
+	}
+
+	return cfg, nil
+}
+
+// Parse reads YAML config data as Load does, without validating it. It
+// refuses a key that no setting has and a size setting whose value is not a
+// size, naming the setting. It applies default values for optional
+// fields, fills in ${...} references using smartconfig, and expands ~ in
+// paths.
+func Parse(data []byte) (*Config, error) {
+	// smartconfig drops line numbers, so keys are checked in the data as
+	// written.
+	var root yaml.Node
+
+	err := yaml.Unmarshal(data, &root)
+	if err != nil {
+		return nil, fmt.Errorf("failed to parse config: %w", err)
+	}
+
+	err = checkKeys(&root, reflect.TypeFor[Config](), "",
+		map[*yaml.Node]bool{}, map[nodeAndType]bool{})
+	if err != nil {
+		return nil, err
+	}
+
+	// Load config using smartconfig for interpolation
+	sc, err := smartconfig.NewFromReader(bytes.NewReader(data))
+	if err != nil {
+		return nil, fmt.Errorf("failed to load config file: %w", err)
+	}
+
+	// A ${...} reference can bring in keys that are not in the file, and
+	// is the value of a size setting only once it is filled in.
+	err = checkFilledConfig(sc.Data())
+	if err != nil {
+		return nil, err
+	}
+
 	cfg := &Config{
 		// Set defaults
 		BlobSizeLimit:    defaultBlobSizeLimit,
 		ChunkSize:        defaultChunkSize,
 		IndexPath:        filepath.Join(xdg.DataHome, appName, "index.sqlite"),
 		CompressionLevel: defaultCompressionLevel,
+		PruneGracePeriod: defaultPruneGracePeriod,
+		S3:               S3Config{PartSize: defaultS3PartSize},
 	}
 
 	// Convert smartconfig data to YAML then unmarshal
@@ -296,50 +381,29 @@ func Load(path string) (*Config, error) {
 		cfg.S3.Region = "us-east-1"
 	}
 
-	if cfg.S3.PartSize == 0 {
-		cfg.S3.PartSize = defaultS3PartSize
-	}
-
-	// Check config file permissions (warn if world or group readable)
-	//nolint:gosec // G703: config path is operator-supplied by design
-	info, statErr := os.Stat(path)
-	if statErr == nil {
-		mode := info.Mode().Perm()
-		if mode&0044 != 0 { // group or world readable
-			log.Warn("Config file has insecure permissions (contains S3 credentials)",
-				"path", path,
-				"mode", fmt.Sprintf("%04o", mode),
-				"recommendation", "chmod 600 "+path)
-		}
-	}
-
-	err = cfg.Validate()
-	if err != nil {
-		return nil, fmt.Errorf("invalid config: %w", err)
-	}
-
 	return cfg, nil
 }
 
 // Validate checks if the configuration is valid and complete.
 // It ensures all required fields are present and have valid values:
-//   - At least one age recipient must be specified, and every recipient must
-//     parse as an X25519 age1... public key (so a bad entry fails at load, not
-//     mid-backup); errors name the position, never the value
+//   - Every age recipient must parse as an X25519 age1... public key (so a
+//     bad entry fails at load, not mid-backup); errors name the position,
+//     never the value. An empty list is accepted, because only snapshot
+//     create needs a recipient and it checks for one itself
 //   - At least one snapshot must be configured with at least one path
 //   - Storage must be configured (either storage_url or s3.* fields)
+//   - The index path must be absolute: a relative one would resolve against
+//     the working directory, so each directory would get its own index
 //   - Chunk size must be at least 1MB
 //   - Blob size limit must be at least the largest chunk the chunker can emit
 //     (chunk_size times chunker.ChunkSizeSpread), so a single-chunk blob never
 //     exceeds the configured limit
 //   - Compression level must be between 1 and 19
+//   - S3 part size must be between 5MiB and 5GiB, the part sizes S3 accepts
+//   - Prune grace period must not be negative
 //
 // Returns an error describing the first validation failure encountered.
 func (c *Config) Validate() error {
-	if len(c.AgeRecipients) == 0 {
-		return errNoAgeRecipients
-	}
-
 	for i, recipient := range c.AgeRecipients {
 		err := validateAgeRecipient(recipient)
 		if err != nil {
@@ -363,6 +427,10 @@ func (c *Config) Validate() error {
 		return err
 	}
 
+	if !filepath.IsAbs(c.IndexPath) {
+		return fmt.Errorf("%w: %q", errIndexPathNotAbsolute, c.IndexPath)
+	}
+
 	if c.ChunkSize.Int64() < minChunkSize {
 		return errChunkSizeTooSmall
 	}
@@ -381,6 +449,15 @@ func (c *Config) Validate() error {
 		return errBadCompression
 	}
 
+	if c.S3.PartSize.Int64() < minS3PartSize ||
+		c.S3.PartSize.Int64() > maxS3PartSize {
+		return errBadS3PartSize
+	}
+
+	if c.PruneGracePeriod < 0 {
+		return errNegativePruneGracePeriod
+	}
+
 	return nil
 }
 
@@ -416,6 +493,18 @@ func (c *Config) setAgeSecretKey() {
 	}
 }
 
+// readableByOthersWarning is the warning Load logs when others can read
+// the config file. It says "may contain" because the S3 credentials are
+// seen only after smartconfig has replaced any ${...} reference in the
+// file with its value, so a set credential need not be in the file.
+func (c *Config) readableByOthersWarning() string {
+	if c.S3.AccessKeyID != "" || c.S3.SecretAccessKey != "" {
+		return "Config file is readable by others and may contain S3 credentials"
+	}
+
+	return "Config file is readable by others"
+}
+
 // validateStorage validates storage configuration.
 // If StorageURL is set, it takes precedence. S3 URLs require credentials.
 // File URLs don't require any S3 configuration.
@@ -447,11 +536,18 @@ func (c *Config) validateStorage() error {
 
 // validateStorageURL validates URL-based storage configuration. File and
 // rclone URLs need no credentials; S3 URLs require the legacy s3.*
-// credential fields.
+// credential fields. A file URL must name an absolute path: a relative one
+// would resolve against the working directory, so the destination would
+// move with it.
 func (c *Config) validateStorageURL() error {
 	switch {
 	case strings.HasPrefix(c.StorageURL, "file://"):
-		// File storage doesn't need S3 credentials
+		// storage.ParseStorageURL applies the same rule, but config
+		// cannot import storage, which imports config.
+		if !strings.HasPrefix(c.StorageURL, "file:///") {
+			return errFileURLNotAbsolute
+		}
+
 		return nil
 	case strings.HasPrefix(c.StorageURL, "rclone://"):
 		// Rclone storage uses rclone's own config

internal/config/config_test.go

@@ -6,8 +6,10 @@ import (
 	"path/filepath"
 	"strings"
 	"testing"
+	"time"
 
 	"sneak.berlin/go/vaultik/internal/chunker"
+	"sneak.berlin/go/vaultik/internal/log"
 )
 
 const (
@@ -88,12 +90,12 @@ func TestConfigLoad(t *testing.T) {
 }
 
 // TestExampleConfigIsScrubbedAndLoads checks that the shipped
-// config.example.yml carries only neutral placeholders (no real credentials,
-// private addresses, or internal host names) and still parses.
+// configs/config.example.yml carries only neutral placeholders (no real
+// credentials, private addresses, or internal host names) and still parses.
 func TestExampleConfigIsScrubbedAndLoads(t *testing.T) {
 	t.Parallel()
 
-	examplePath := filepath.Join("..", "..", "config.example.yml")
+	examplePath := filepath.Join("..", "..", "configs", "config.example.yml")
 
 	cfg, err := Load(examplePath)
 	if err != nil {
@@ -163,9 +165,11 @@ func TestValidateBlobSizeLimit(t *testing.T) {
 			AgeRecipients:    []string{testSneakAgePublicKey},
 			Snapshots:        map[string]SnapshotConfig{"test": {Paths: []string{"/tmp/src"}}},
 			StorageURL:       "file:///tmp/vaultik-test-store",
+			IndexPath:        "/tmp/vaultik-test-index.sqlite",
 			ChunkSize:        chunkSize,
 			BlobSizeLimit:    blobLimit,
 			CompressionLevel: 3,
+			S3:               S3Config{PartSize: defaultS3PartSize},
 		}
 	}
 
@@ -221,9 +225,355 @@ func TestValidateBlobSizeLimit(t *testing.T) {
 	}
 }
 
+// TestValidateS3PartSize checks that s3.part_size is held to the part sizes
+// S3 accepts, 5MiB to 5GiB, by changing only the part size of the test
+// config. "5MB" in the config file is 5,000,000 bytes, below the minimum.
+func TestValidateS3PartSize(t *testing.T) {
+	t.Parallel()
+
+	base, err := Load(os.Getenv("VAULTIK_CONFIG"))
+	if err != nil {
+		t.Fatalf("Failed to load config: %v", err)
+	}
+
+	tests := []struct {
+		name     string
+		partSize Size
+		wantErr  bool
+	}{
+		{
+			name:     "5MB is rejected",
+			partSize: 5_000_000,
+			wantErr:  true,
+		},
+		{
+			name:     "one byte below 5MiB is rejected",
+			partSize: minS3PartSize - 1,
+			wantErr:  true,
+		},
+		{
+			name:     "5MiB is accepted",
+			partSize: minS3PartSize,
+			wantErr:  false,
+		},
+		{
+			name:     "5GiB is accepted",
+			partSize: maxS3PartSize,
+			wantErr:  false,
+		},
+		{
+			name:     "one byte above 5GiB is rejected",
+			partSize: maxS3PartSize + 1,
+			wantErr:  true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			cfg := *base
+			cfg.S3.PartSize = tt.partSize
+
+			err := cfg.Validate()
+			if tt.wantErr {
+				if !errors.Is(err, errBadS3PartSize) {
+					t.Fatalf("Validate() error = %v, want errBadS3PartSize", err)
+				}
+
+				return
+			}
+
+			if err != nil {
+				t.Fatalf("Validate() unexpected error: %v", err)
+			}
+		})
+	}
+}
+
+// TestLoadS3PartSize checks that a config file without s3.part_size loads
+// with the 5MiB default, and that an explicit 0 fails at load like any other
+// part size S3 refuses.
+func TestLoadS3PartSize(t *testing.T) {
+	t.Parallel()
+
+	const withoutPartSize = "snapshots:\n" +
+		"  test:\n" +
+		"    paths: [/tmp/vaultik-test-source]\n" +
+		"storage_url: file:///tmp/vaultik-test-storage\n"
+
+	writeConfig := func(t *testing.T, text string) string {
+		t.Helper()
+
+		path := filepath.Join(t.TempDir(), "config.yml")
+
+		err := os.WriteFile(path, []byte(text), 0o600)
+		if err != nil {
+			t.Fatalf("write config: %v", err)
+		}
+
+		return path
+	}
+
+	t.Run("absent loads as 5MiB", func(t *testing.T) {
+		t.Parallel()
+
+		cfg, err := Load(writeConfig(t, withoutPartSize))
+		if err != nil {
+			t.Fatalf("Load() unexpected error: %v", err)
+		}
+
+		if cfg.S3.PartSize != defaultS3PartSize {
+			t.Errorf("s3.part_size = %d, want %d",
+				cfg.S3.PartSize, defaultS3PartSize)
+		}
+	})
+
+	t.Run("0 is rejected", func(t *testing.T) {
+		t.Parallel()
+
+		_, err := Load(writeConfig(t, withoutPartSize+"s3:\n  part_size: 0\n"))
+		if !errors.Is(err, errBadS3PartSize) {
+			t.Fatalf("Load() error = %v, want errBadS3PartSize", err)
+		}
+	})
+}
+
+// TestDocumentedSizeDefaults checks that the defaults are the documented
+// chunk_size 10MB and blob_size_limit 10GB as the parser reads those values
+// (1000-based), so writing a default out in the config file changes nothing.
+func TestDocumentedSizeDefaults(t *testing.T) {
+	t.Parallel()
+
+	cfg, err := Parse([]byte("chunk_size: 10MB\nblob_size_limit: 10GB\n"))
+	if err != nil {
+		t.Fatalf("Parse() unexpected error: %v", err)
+	}
+
+	if cfg.ChunkSize != defaultChunkSize {
+		t.Errorf("chunk_size: 10MB is %d bytes, the default is %d",
+			cfg.ChunkSize, defaultChunkSize)
+	}
+
+	if cfg.BlobSizeLimit != defaultBlobSizeLimit {
+		t.Errorf("blob_size_limit: 10GB is %d bytes, the default is %d",
+			cfg.BlobSizeLimit, defaultBlobSizeLimit)
+	}
+}
+
+// TestDocumentedSizeExamples checks the 1MB minimum chunk_size and that
+// every size the README, configs/config.example.yml and the config init template
+// list as an example is accepted.
+func TestDocumentedSizeExamples(t *testing.T) {
+	t.Parallel()
+
+	const withoutSizes = "snapshots:\n" +
+		"  test:\n" +
+		"    paths: [/tmp/vaultik-test-source]\n" +
+		"storage_url: file:///tmp/vaultik-test-storage\n"
+
+	tests := []struct {
+		setting string
+		wantErr error
+	}{
+		{"chunk_size: 1MB", nil},
+		{"chunk_size: 999999", errChunkSizeTooSmall},
+		{"chunk_size: 5M", nil},
+		{"chunk_size: 10M", nil},
+		{"chunk_size: 10MB", nil},
+		{"chunk_size: 64MiB", nil},
+		{"chunk_size: 1GB", nil},
+		// Four times the default chunk_size, the documented minimum.
+		{"blob_size_limit: 40MB", nil},
+		{"blob_size_limit: 500MB", nil},
+		{"blob_size_limit: 1GB", nil},
+		{"blob_size_limit: 1GiB", nil},
+		{"blob_size_limit: 10G", nil},
+		{"blob_size_limit: 10GB", nil},
+		{"s3: {part_size: 5MiB}", nil},
+		{"s3: {part_size: 10MB}", nil},
+		{"s3: {part_size: 16MiB}", nil},
+		{"s3: {part_size: 100MiB}", nil},
+		{"s3: {part_size: 5GiB}", nil},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.setting, func(t *testing.T) {
+			t.Parallel()
+
+			cfg, err := Parse([]byte(withoutSizes + tt.setting + "\n"))
+			if err != nil {
+				t.Fatalf("Parse() unexpected error: %v", err)
+			}
+
+			err = cfg.Validate()
+			if !errors.Is(err, tt.wantErr) {
+				t.Fatalf("Validate() error = %v, want %v", err, tt.wantErr)
+			}
+		})
+	}
+}
+
+// TestValidatePruneGracePeriod checks that a negative prune_grace_period
+// is refused rather than acting like 0s, which deletes unreferenced blobs
+// and directories under metadata/ with no manifest at once, and that 0s
+// and the documented examples are accepted.
+func TestValidatePruneGracePeriod(t *testing.T) {
+	t.Parallel()
+
+	const withoutGracePeriod = "snapshots:\n" +
+		"  test:\n" +
+		"    paths: [/tmp/vaultik-test-source]\n" +
+		"storage_url: file:///tmp/vaultik-test-storage\n"
+
+	tests := []struct {
+		setting string
+		wantErr error
+	}{
+		{"prune_grace_period: 0s", nil},
+		{"prune_grace_period: 90m", nil},
+		{"prune_grace_period: 36h", nil},
+		{"prune_grace_period: -1h", errNegativePruneGracePeriod},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.setting, func(t *testing.T) {
+			t.Parallel()
+
+			cfg, err := Parse([]byte(withoutGracePeriod + tt.setting + "\n"))
+			if err != nil {
+				t.Fatalf("Parse() unexpected error: %v", err)
+			}
+
+			err = cfg.Validate()
+			if !errors.Is(err, tt.wantErr) {
+				t.Fatalf("Validate() error = %v, want %v", err, tt.wantErr)
+			}
+		})
+	}
+}
+
+// TestLoadPruneGracePeriod checks that a config file without
+// prune_grace_period loads with the documented 24h. The value is written out
+// here rather than taken from defaultPruneGracePeriod, so that a change of the
+// default fails this test.
+func TestLoadPruneGracePeriod(t *testing.T) {
+	t.Parallel()
+
+	const withoutGracePeriod = "snapshots:\n" +
+		"  test:\n" +
+		"    paths: [/tmp/vaultik-test-source]\n" +
+		"storage_url: file:///tmp/vaultik-test-storage\n"
+
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte(withoutGracePeriod), 0o600)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+
+	cfg, err := Load(path)
+	if err != nil {
+		t.Fatalf("Load() unexpected error: %v", err)
+	}
+
+	if cfg.PruneGracePeriod != 24*time.Hour {
+		t.Errorf("prune_grace_period = %s, want 24h", cfg.PruneGracePeriod)
+	}
+}
+
+// TestLoadRejectsRelativeFileURL checks that a file:// storage_url with a
+// relative path fails at load. Accepted, it would resolve against the
+// working directory, so a backup run from another directory would write
+// to a different destination and still succeed.
+func TestLoadRejectsRelativeFileURL(t *testing.T) {
+	t.Parallel()
+
+	const relativeConfig = "snapshots:\n" +
+		"  test:\n" +
+		"    paths: [/tmp/vaultik-test-source]\n" +
+		"storage_url: file://dest\n"
+
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte(relativeConfig), 0o600)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+
+	_, err = Load(path)
+	if !errors.Is(err, errFileURLNotAbsolute) {
+		t.Fatalf("Load() error = %v, want errFileURLNotAbsolute", err)
+	}
+
+	if !strings.Contains(err.Error(), "absolute path") {
+		t.Errorf("error %q does not say to use an absolute path", err.Error())
+	}
+}
+
+// TestLoadIndexPath checks that a relative index_path fails at load, naming
+// the value. Accepted, it would resolve against the working directory, so a
+// backup run from another directory would start from an empty index. A path
+// starting with ~ is expanded and loads.
+func TestLoadIndexPath(t *testing.T) {
+	t.Parallel()
+
+	home, err := os.UserHomeDir()
+	if err != nil {
+		t.Fatalf("home directory: %v", err)
+	}
+
+	load := func(t *testing.T, indexPath string) (*Config, error) {
+		t.Helper()
+
+		data := "snapshots:\n" +
+			"  test:\n" +
+			"    paths: [/tmp/vaultik-test-source]\n" +
+			"storage_url: file:///tmp/vaultik-test-storage\n" +
+			"index_path: " + indexPath + "\n"
+
+		path := filepath.Join(t.TempDir(), "config.yml")
+
+		err := os.WriteFile(path, []byte(data), 0o600)
+		if err != nil {
+			t.Fatalf("write config: %v", err)
+		}
+
+		return Load(path)
+	}
+
+	t.Run("relative path is rejected", func(t *testing.T) {
+		t.Parallel()
+
+		_, err := load(t, "relative/index.sqlite")
+		if !errors.Is(err, errIndexPathNotAbsolute) {
+			t.Fatalf("Load() error = %v, want errIndexPathNotAbsolute", err)
+		}
+
+		if !strings.Contains(err.Error(), `"relative/index.sqlite"`) {
+			t.Errorf("error %q does not name the value", err.Error())
+		}
+	})
+
+	t.Run("path starting with ~ is expanded", func(t *testing.T) {
+		t.Parallel()
+
+		cfg, err := load(t, "~/vaultik/index.sqlite")
+		if err != nil {
+			t.Fatalf("Load() unexpected error: %v", err)
+		}
+
+		want := filepath.Join(home, "vaultik", "index.sqlite")
+		if cfg.IndexPath != want {
+			t.Errorf("IndexPath = %q, want %q", cfg.IndexPath, want)
+		}
+	})
+}
+
 // TestValidateAgeRecipients checks that recipients are parsed at config load
 // (a bad entry fails immediately, not mid-backup) and that no invalid entry —
-// least of all a pasted secret key — is echoed in the error.
+// least of all a pasted secret key — is echoed in the error. An empty list
+// loads, because only snapshot create needs a recipient.
 func TestValidateAgeRecipients(t *testing.T) {
 	t.Parallel()
 
@@ -232,9 +582,11 @@ func TestValidateAgeRecipients(t *testing.T) {
 			AgeRecipients:    recipients,
 			Snapshots:        map[string]SnapshotConfig{"test": {Paths: []string{"/tmp/src"}}},
 			StorageURL:       "file:///tmp/vaultik-test-store",
+			IndexPath:        "/tmp/vaultik-test-index.sqlite",
 			ChunkSize:        Size(10 * 1024 * 1024),
 			BlobSizeLimit:    Size(10 * 1024 * 1024 * 1024),
 			CompressionLevel: 3,
+			S3:               S3Config{PartSize: defaultS3PartSize},
 		}
 	}
 
@@ -244,7 +596,12 @@ func TestValidateAgeRecipients(t *testing.T) {
 		wantErr    bool
 	}{
 		{
-			name:       "config init placeholder is rejected",
+			name:       "no recipients is accepted",
+			recipients: nil,
+			wantErr:    false,
+		},
+		{
+			name:       "placeholder recipient is rejected",
 			recipients: []string{"age1REPLACE_WITH_YOUR_PUBLIC_KEY"},
 			wantErr:    true,
 		},
@@ -337,3 +694,125 @@ func TestAgeSecretKeySourceName(t *testing.T) {
 		})
 	}
 }
+
+// loadReadableConfig writes configYAML to a file that others can read,
+// loads it, and returns what the logger wrote to stderr meanwhile. The
+// logger writes to the os.Stderr it finds when it is initialized, so
+// os.Stderr is pointed at a file first. Not parallel-safe: os.Stderr and
+// the logger are process-global.
+func loadReadableConfig(t *testing.T, configYAML string) string {
+	t.Helper()
+
+	dir := t.TempDir()
+	configPath := filepath.Join(dir, "config.yml")
+	stderrPath := filepath.Join(dir, "stderr")
+
+	err := os.WriteFile(configPath, []byte(configYAML), 0o600)
+	if err != nil {
+		t.Fatalf("writing config: %v", err)
+	}
+
+	//nolint:gosec // G302: the test needs a config file others can read
+	err = os.Chmod(configPath, 0o644)
+	if err != nil {
+		t.Fatalf("chmod config: %v", err)
+	}
+
+	stderrFile, err := os.Create(stderrPath) //nolint:gosec // G304: test temp path
+	if err != nil {
+		t.Fatalf("creating stderr file: %v", err)
+	}
+
+	previous := os.Stderr
+	os.Stderr = stderrFile
+
+	log.Initialize(log.Config{})
+
+	_, loadErr := Load(configPath)
+
+	os.Stderr = previous
+
+	log.Initialize(log.Config{})
+
+	_ = stderrFile.Close()
+
+	if loadErr != nil {
+		t.Fatalf("Load() error = %v", loadErr)
+	}
+
+	captured, err := os.ReadFile(stderrPath) //nolint:gosec // G304: test temp path
+	if err != nil {
+		t.Fatalf("reading stderr file: %v", err)
+	}
+
+	return string(captured)
+}
+
+// TestLoadWarnsReadableConfigWithoutS3Credentials checks that a config
+// file others can read, holding no S3 credentials, is warned about
+// without a claim that it holds them.
+//
+//nolint:paralleltest // loadReadableConfig replaces os.Stderr
+func TestLoadWarnsReadableConfigWithoutS3Credentials(t *testing.T) {
+	stderr := loadReadableConfig(t, `
+storage_url: file:///var/backups/vaultik
+snapshots:
+  home:
+    paths:
+      - /home
+`)
+
+	if !strings.Contains(stderr, "Config file is readable by others") {
+		t.Errorf("expected a warning that the file is readable by others, got %q",
+			stderr)
+	}
+
+	if strings.Contains(stderr, "S3 credentials") {
+		t.Errorf("warning names S3 credentials the file does not set: %q", stderr)
+	}
+}
+
+// TestLoadWarnsReadableConfigWithS3Credentials checks that a config file
+// others can read and that sets S3 credentials, as values or as ${ENV:...}
+// references, is warned about as one that may contain them.
+//
+//nolint:paralleltest // loadReadableConfig replaces os.Stderr
+func TestLoadWarnsReadableConfigWithS3Credentials(t *testing.T) {
+	t.Setenv("VAULTIK_TEST_ACCESS_KEY_ID", "test-access-key")
+	t.Setenv("VAULTIK_TEST_SECRET_ACCESS_KEY", "test-secret-key")
+
+	configs := map[string]string{
+		"values": `
+storage_url: s3://bucket/prefix?endpoint=s3.example.com
+s3:
+  access_key_id: test-access-key
+  secret_access_key: test-secret-key
+snapshots:
+  home:
+    paths:
+      - /home
+`,
+		"references": `
+storage_url: s3://bucket/prefix?endpoint=s3.example.com
+s3:
+  access_key_id: ${ENV:VAULTIK_TEST_ACCESS_KEY_ID}
+  secret_access_key: ${ENV:VAULTIK_TEST_SECRET_ACCESS_KEY}
+snapshots:
+  home:
+    paths:
+      - /home
+`,
+	}
+
+	for name, configYAML := range configs {
+		t.Run(name, func(t *testing.T) {
+			stderr := loadReadableConfig(t, configYAML)
+
+			if !strings.Contains(stderr,
+				"Config file is readable by others and may contain S3 credentials") {
+				t.Errorf("expected a warning naming the S3 credentials, got %q",
+					stderr)
+			}
+		})
+	}
+}

internal/config/keys.go

@@ -0,0 +1,220 @@
+package config
+
+import (
+	"errors"
+	"fmt"
+	"reflect"
+	"strconv"
+	"strings"
+
+	"gopkg.in/yaml.v3"
+)
+
+// yamlMergeTag is the tag YAML gives a merge key (<<), which adds the keys
+// of the map it names to the map it sits in.
+const yamlMergeTag = "!!merge"
+
+// EnvSection is the top-level key of a map of environment variables that
+// smartconfig sets while it loads the config. It is not a vaultik setting,
+// so its keys are not checked.
+const EnvSection = "env"
+
+var (
+	errUnknownKey           = errors.New("unknown config key")
+	errAnchorContainsItself = errors.New("anchor contains itself")
+)
+
+// nodeAndType is a node of a parsed config and the setting type it is
+// checked against.
+type nodeAndType struct {
+	node *yaml.Node
+	typ  reflect.Type
+}
+
+// SettingType returns the type of the setting that the dotted key names in
+// Config, following the fields' yaml tags, as for s3.bucket or
+// snapshots.home.exclude.0. A snapshot name or a list index may be any
+// word. It returns an unknown key error when no setting has that key.
+func SettingType(keys []string) (reflect.Type, error) {
+	typ := reflect.TypeFor[Config]()
+
+	for _, key := range keys {
+		switch {
+		case typ.Kind() == reflect.Map || typ.Kind() == reflect.Slice:
+			typ = typ.Elem()
+		case typ.Kind() == reflect.Struct:
+			field, ok := yamlField(typ, key)
+			if !ok {
+				return nil, fmt.Errorf("%w %q", errUnknownKey, strings.Join(keys, "."))
+			}
+
+			typ = field.Type
+		default:
+			return nil, fmt.Errorf("%w %q", errUnknownKey, strings.Join(keys, "."))
+		}
+	}
+
+	return typ, nil
+}
+
+// yamlField returns the field of struct type typ whose yaml tag names key.
+func yamlField(typ reflect.Type, key string) (reflect.StructField, bool) {
+	for field := range typ.Fields() {
+		name, _, _ := strings.Cut(field.Tag.Get("yaml"), ",")
+		if name == key && name != "-" {
+			return field, true
+		}
+	}
+
+	return reflect.StructField{}, false
+}
+
+// checkKeys returns an error naming the first key in node, a parsed config
+// or part of one at the dotted path, that no setting in typ has. The error
+// gives the key's line in the file, when it has one. A value of the wrong
+// type is left for the YAML decoder to report. following holds the aliases
+// the walk is inside. Meeting one of them again means its anchor contains
+// itself, which is an error. checked holds each anchored node the walk has
+// checked completely, with the type it was checked against. An alias to
+// that node is not followed again for that type.
+func checkKeys(node *yaml.Node, typ reflect.Type, path string,
+	following map[*yaml.Node]bool, checked map[nodeAndType]bool,
+) error {
+	switch node.Kind {
+	case yaml.DocumentNode:
+		for _, child := range node.Content {
+			err := checkKeys(child, typ, path, following, checked)
+			if err != nil {
+				return err
+			}
+		}
+	case yaml.AliasNode:
+		if following[node] {
+			return fmt.Errorf("%w: %q on line %d",
+				errAnchorContainsItself, node.Value, node.Line)
+		}
+
+		target := nodeAndType{node.Alias, typ}
+		if checked[target] {
+			return nil
+		}
+
+		following[node] = true
+		defer delete(following, node)
+
+		err := checkKeys(node.Alias, typ, path, following, checked)
+		if err != nil {
+			return err
+		}
+
+		checked[target] = true
+	case yaml.SequenceNode:
+		if typ.Kind() != reflect.Slice {
+			return nil
+		}
+
+		for i, child := range node.Content {
+			err := checkKeys(child, typ.Elem(), joinKey(path, strconv.Itoa(i)),
+				following, checked)
+			if err != nil {
+				return err
+			}
+		}
+	case yaml.MappingNode:
+		return checkMappingKeys(node, typ, path, following, checked)
+	case yaml.ScalarNode:
+		// A scalar has no keys.
+	}
+
+	return nil
+}
+
+// checkFilledConfig checks data, the config after smartconfig has filled in
+// its ${...} references: its keys with checkKeys, then its size settings
+// with checkSizes. A key that a reference brings in is not in the file, and
+// the copy that Encode makes has no line numbers.
+func checkFilledConfig(data map[string]any) error {
+	var filled yaml.Node
+
+	err := filled.Encode(data)
+	if err != nil {
+		return fmt.Errorf("failed to marshal config data: %w", err)
+	}
+
+	err = checkKeys(&filled, reflect.TypeFor[Config](), "",
+		map[*yaml.Node]bool{}, map[nodeAndType]bool{})
+	if err != nil {
+		return err
+	}
+
+	return checkSizes(&filled, nil)
+}
+
+// checkMappingKeys is checkKeys for a map.
+func checkMappingKeys(node *yaml.Node, typ reflect.Type, path string,
+	following map[*yaml.Node]bool, checked map[nodeAndType]bool,
+) error {
+	for i := 0; i+1 < len(node.Content); i += 2 {
+		key, value := node.Content[i], node.Content[i+1]
+
+		if key.Tag == yamlMergeTag {
+			// The value is one map or a list of maps.
+			merged := []*yaml.Node{value}
+			if value.Kind == yaml.SequenceNode {
+				merged = value.Content
+			}
+
+			for _, m := range merged {
+				err := checkKeys(m, typ, path, following, checked)
+				if err != nil {
+					return err
+				}
+			}
+
+			continue
+		}
+
+		keyPath := joinKey(path, key.Value)
+		if keyPath == EnvSection {
+			continue
+		}
+
+		var valueType reflect.Type
+
+		switch {
+		case typ.Kind() == reflect.Map:
+			valueType = typ.Elem()
+		case typ.Kind() == reflect.Struct:
+			field, ok := yamlField(typ, key.Value)
+			if !ok && key.Line == 0 {
+				// A key that a ${...} reference brought in has no line.
+				return fmt.Errorf("%w %q", errUnknownKey, keyPath)
+			}
+
+			if !ok {
+				return fmt.Errorf("%w %q on line %d",
+					errUnknownKey, keyPath, key.Line)
+			}
+
+			valueType = field.Type
+		default:
+			return nil
+		}
+
+		err := checkKeys(value, valueType, keyPath, following, checked)
+		if err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+// joinKey appends key to the dotted path.
+func joinKey(path, key string) string {
+	if path == "" {
+		return key
+	}
+
+	return path + "." + key
+}

internal/config/keys_test.go

@@ -0,0 +1,163 @@
+package config //nolint:testpackage // checks the unexported errUnknownKey
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"slices"
+	"strconv"
+	"strings"
+	"testing"
+)
+
+// keysTestConfig loads; each case below adds one line to it.
+const keysTestConfig = "storage_url: file:///tmp/vaultik-test-storage\n" +
+	"s3:\n" +
+	"  region: us-east-1\n" +
+	"snapshots:\n" +
+	"  test:\n" +
+	"    paths: [/tmp/vaultik-test-source]\n"
+
+// TestParseRefusesUnknownKey checks that a key no setting has, at any
+// depth, fails the parse with an error naming the key and its line. A key
+// that a ${...} reference brings in is not in the file, so its error gives
+// no line.
+func TestParseRefusesUnknownKey(t *testing.T) {
+	t.Parallel()
+
+	snapshotPath := filepath.Join(t.TempDir(), "snapshot.yml")
+
+	err := os.WriteFile(snapshotPath, []byte("other:\n"+
+		"  paths: [/tmp/vaultik-test-source]\n"+
+		"  exlude: [/tmp/cache]\n"), 0o600)
+	if err != nil {
+		t.Fatalf("write snapshot: %v", err)
+	}
+
+	tests := []struct {
+		name string
+		yaml string
+		want string
+	}{
+		{"top level", keysTestConfig + "compression_lvl: 9\n",
+			`unknown config key "compression_lvl" on line 7`},
+		{"in a snapshot", keysTestConfig + "    exlude: [/tmp/cache]\n",
+			`unknown config key "snapshots.test.exlude" on line 7`},
+		{"in s3", strings.Replace(keysTestConfig, "  region:",
+			"  bukcet: b\n  region:", 1),
+			`unknown config key "s3.bukcet" on line 3`},
+		{"from a reference",
+			keysTestConfig + "  other: ${YAML:" + snapshotPath + ":other}\n",
+			`unknown config key "snapshots.other.exlude"`},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			_, err := Parse([]byte(tt.yaml))
+			if !errors.Is(err, errUnknownKey) {
+				t.Fatalf("Parse() error = %v, want errUnknownKey", err)
+			}
+
+			if err.Error() != tt.want {
+				t.Errorf("Parse() error = %q, want %q", err.Error(), tt.want)
+			}
+		})
+	}
+}
+
+// TestParseRefusesAnchorThatContainsItself checks that a YAML merge key (<<)
+// naming the anchor of the map it sits in fails the parse with an error.
+func TestParseRefusesAnchorThatContainsItself(t *testing.T) {
+	t.Parallel()
+
+	data := strings.Replace(keysTestConfig, "  test:", "  test: &test", 1) +
+		"    <<: *test\n"
+
+	_, err := Parse([]byte(data))
+	if !errors.Is(err, errAnchorContainsItself) {
+		t.Fatalf("Parse() error = %v, want errAnchorContainsItself", err)
+	}
+
+	want := `anchor contains itself: "test" on line 7`
+	if err.Error() != want {
+		t.Errorf("Parse() error = %q, want %q", err.Error(), want)
+	}
+}
+
+// TestParseRefusesExcessiveAliasing checks that Parse returns an error for
+// a file whose merge keys (<<) fan out. Each of eight snapshots merges the
+// one before it ten times. A key check that followed every alias would walk
+// the first snapshot over 10^8 times and run past the package's test
+// timeout.
+func TestParseRefusesExcessiveAliasing(t *testing.T) {
+	t.Parallel()
+
+	const snapshots, merges = 8, 10
+
+	var data strings.Builder
+
+	data.WriteString(strings.Replace(keysTestConfig, "  test:", "  test: &s0", 1))
+
+	for i := 1; i <= snapshots; i++ {
+		aliases := slices.Repeat([]string{"*s" + strconv.Itoa(i-1)}, merges)
+		fmt.Fprintf(&data, "  s%d: &s%d\n    <<: [%s]\n",
+			i, i, strings.Join(aliases, ", "))
+	}
+
+	_, err := Parse([]byte(data.String()))
+	if err == nil {
+		t.Fatal("Parse() returned no error")
+	}
+}
+
+// TestParseAcceptsKnownKeys checks that a config with a ${...} reference
+// and a YAML merge key (<<) still parses.
+func TestParseAcceptsKnownKeys(t *testing.T) {
+	t.Parallel()
+
+	levelPath := filepath.Join(t.TempDir(), "level")
+
+	err := os.WriteFile(levelPath, []byte("9"), 0o600)
+	if err != nil {
+		t.Fatalf("write level: %v", err)
+	}
+
+	data := strings.Replace(keysTestConfig, "  test:", "  test: &test", 1) +
+		"  other:\n" +
+		"    <<: *test\n" +
+		"    exclude: [/tmp/vaultik-test-source/cache]\n" +
+		"compression_level: ${FILE:" + levelPath + "}\n"
+
+	cfg, err := Parse([]byte(data))
+	if err != nil {
+		t.Fatalf("Parse() unexpected error: %v", err)
+	}
+
+	if cfg.CompressionLevel != 9 {
+		t.Errorf("compression_level = %d, want 9", cfg.CompressionLevel)
+	}
+
+	other := cfg.Snapshots["other"]
+	if len(other.Paths) != 1 || len(other.Exclude) != 1 {
+		t.Errorf("snapshot other = %+v, want the paths of test and one exclude",
+			other)
+	}
+}
+
+// TestParseAcceptsEnvSection checks that a config with a top-level env
+// section still parses. smartconfig sets the environment variables it
+// lists, so its keys are not settings.
+func TestParseAcceptsEnvSection(t *testing.T) {
+	const name = "VAULTIK_TEST_ENV_SECTION"
+
+	// Parse sets the variable; t.Setenv restores it after the test.
+	t.Setenv(name, "")
+
+	_, err := Parse([]byte(keysTestConfig + "env:\n  " + name + ": set\n"))
+	if err != nil {
+		t.Fatalf("Parse() unexpected error: %v", err)
+	}
+}

internal/config/size.go

@@ -4,20 +4,25 @@ import (
 	"errors"
 	"fmt"
 	"math"
+	"reflect"
+	"slices"
+	"strconv"
+	"strings"
 
 	"github.com/dustin/go-humanize"
+	"gopkg.in/yaml.v3"
 )
 
 var (
 	errSizeType     = errors.New("size must be a number or string")
 	errSizeTooLarge = errors.New("size exceeds maximum supported value")
+	errNotASize     = errors.New(
+		"is not a size; use a number of bytes or a unit such as 10MB")
 )
 
 // Size represents a byte size that can be specified in configuration files.
 // It can unmarshal from both numeric values (interpreted as bytes) and
 // human-readable strings like "10MB", "2.5GB", or "1TB".
-//
-//nolint:recvcheck // UnmarshalYAML requires a pointer; String/Int64 are value reads
 type Size int64
 
 // UnmarshalYAML implements yaml.Unmarshaler for Size, allowing it to be
@@ -34,6 +39,15 @@ func (s *Size) UnmarshalYAML(unmarshal func(any) error) error {
 		return nil
 	}
 
+	// A number too large for an int64 decodes as a float64. Its text can be
+	// 1e+20, which is not a size string.
+	var floatVal float64
+
+	err = unmarshal(&floatVal)
+	if err == nil && floatVal > math.MaxInt64 {
+		return errSizeTooLarge
+	}
+
 	// Try to unmarshal as string
 	var strVal string
 
@@ -42,17 +56,66 @@ func (s *Size) UnmarshalYAML(unmarshal func(any) error) error {
 		return errSizeType
 	}
 
-	// Parse the string using go-humanize
-	bytes, err := humanize.ParseBytes(strVal)
+	size, err := ParseSize(strVal)
 	if err != nil {
-		return fmt.Errorf("invalid size format: %w", err)
+		return err
 	}
 
-	if bytes > math.MaxInt64 {
-		return fmt.Errorf("%w: %s", errSizeTooLarge, strVal)
+	*s = size
+
+	return nil
+}
+
+// checkSizes returns an error naming the first size setting in node whose
+// value is not a size. node is the config, or the part of it at the dotted
+// keys, after smartconfig has filled in its ${...} references.
+func checkSizes(node *yaml.Node, keys []string) error {
+	typ, err := SettingType(keys)
+	if err != nil {
+		// Unknown keys are refused before this check, so keys is in one of
+		// two places. The env section holds no settings. A value of the
+		// wrong shape, such as a map given for hostname or inside exclude,
+		// is reported later by the YAML decoder.
+		return nil //nolint:nilerr // not an error here
 	}
 
-	*s = Size(bytes)
+	if typ == reflect.TypeFor[Size]() {
+		var size Size
+
+		err = node.Decode(&size)
+		if err != nil {
+			return fmt.Errorf("%s: %w", strings.Join(keys, "."), err)
+		}
+
+		return nil
+	}
+
+	switch node.Kind {
+	case yaml.DocumentNode:
+		for _, child := range node.Content {
+			err = checkSizes(child, keys)
+			if err != nil {
+				return err
+			}
+		}
+	case yaml.MappingNode:
+		for i := 0; i+1 < len(node.Content); i += 2 {
+			err = checkSizes(node.Content[i+1],
+				slices.Concat(keys, []string{node.Content[i].Value}))
+			if err != nil {
+				return err
+			}
+		}
+	case yaml.SequenceNode:
+		for i, child := range node.Content {
+			err = checkSizes(child, slices.Concat(keys, []string{strconv.Itoa(i)}))
+			if err != nil {
+				return err
+			}
+		}
+	case yaml.ScalarNode, yaml.AliasNode:
+		// A scalar holds no settings, and the filled config has no aliases.
+	}
 
 	return nil
 }
@@ -72,16 +135,17 @@ func (s Size) String() string {
 	return humanize.Bytes(uint64(s))
 }
 
-// ParseSize parses a size string into a Size value
+// ParseSize parses a size string, such as "10MB", into a Size value. An
+// error quotes s.
 func ParseSize(s string) (Size, error) {
-	bytes, err := humanize.ParseBytes(s)
+	bytes, err := humanize.ParseBigBytes(s)
 	if err != nil {
-		return 0, fmt.Errorf("invalid size format: %w", err)
+		return 0, fmt.Errorf("%q %w", s, errNotASize)
 	}
 
-	if bytes > math.MaxInt64 {
+	if !bytes.IsInt64() {
 		return 0, fmt.Errorf("%w: %s", errSizeTooLarge, s)
 	}
 
-	return Size(bytes), nil
+	return Size(bytes.Int64()), nil
 }

internal/config/size_test.go

@@ -0,0 +1,120 @@
+package config //nolint:testpackage // checks errNotASize and errSizeTooLarge
+
+import (
+	"errors"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+// TestParseRefusesValueThatIsNotASize checks that a size setting whose
+// value is not a size fails the parse with an error naming the setting and
+// quoting the value, also when a ${...} reference fills the value in.
+func TestParseRefusesValueThatIsNotASize(t *testing.T) {
+	t.Parallel()
+
+	sizePath := filepath.Join(t.TempDir(), "size")
+
+	err := os.WriteFile(sizePath, []byte("lots"), 0o600)
+	if err != nil {
+		t.Fatalf("write size: %v", err)
+	}
+
+	tests := []struct {
+		name string
+		yaml string
+		want string
+	}{
+		{"chunk_size", keysTestConfig + "chunk_size: banana\n",
+			`chunk_size: "banana" is not a size; ` +
+				"use a number of bytes or a unit such as 10MB"},
+		{"blob_size_limit", keysTestConfig + "blob_size_limit: 10XB\n",
+			`blob_size_limit: "10XB" is not a size; ` +
+				"use a number of bytes or a unit such as 10MB"},
+		{"s3.part_size", strings.Replace(keysTestConfig, "  region:",
+			"  part_size: big\n  region:", 1),
+			`s3.part_size: "big" is not a size; ` +
+				"use a number of bytes or a unit such as 10MB"},
+		{"from a reference",
+			keysTestConfig + "chunk_size: ${FILE:" + sizePath + "}\n",
+			`chunk_size: "lots" is not a size; ` +
+				"use a number of bytes or a unit such as 10MB"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			_, err := Parse([]byte(tt.yaml))
+			if !errors.Is(err, errNotASize) {
+				t.Fatalf("Parse() error = %v, want errNotASize", err)
+			}
+
+			if err.Error() != tt.want {
+				t.Errorf("Parse() error = %q, want %q", err.Error(), tt.want)
+			}
+		})
+	}
+}
+
+// TestParseRefusesSizeTooLarge checks that a size setting whose value is
+// too large for a size fails the parse with an error naming the setting.
+// YAML reads the second value as a float.
+func TestParseRefusesSizeTooLarge(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		value string
+		want  string
+	}{
+		{"100EB", "chunk_size: size exceeds maximum supported value: 100EB"},
+		{"99999999999999999999",
+			"chunk_size: size exceeds maximum supported value"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.value, func(t *testing.T) {
+			t.Parallel()
+
+			_, err := Parse([]byte(keysTestConfig + "chunk_size: " + tt.value + "\n"))
+			if !errors.Is(err, errSizeTooLarge) {
+				t.Fatalf("Parse() error = %v, want errSizeTooLarge", err)
+			}
+
+			if err.Error() != tt.want {
+				t.Errorf("Parse() error = %q, want %q", err.Error(), tt.want)
+			}
+		})
+	}
+}
+
+// TestParseAcceptsSizes checks that size settings given as a number of
+// bytes, as a unit, and through a ${...} reference still parse.
+func TestParseAcceptsSizes(t *testing.T) {
+	t.Parallel()
+
+	sizePath := filepath.Join(t.TempDir(), "size")
+
+	err := os.WriteFile(sizePath, []byte("20MB"), 0o600)
+	if err != nil {
+		t.Fatalf("write size: %v", err)
+	}
+
+	data := strings.Replace(keysTestConfig, "  region:",
+		"  part_size: 6000000\n  region:", 1) +
+		"chunk_size: ${FILE:" + sizePath + "}\n" +
+		"blob_size_limit: 1GB\n"
+
+	cfg, err := Parse([]byte(data))
+	if err != nil {
+		t.Fatalf("Parse() unexpected error: %v", err)
+	}
+
+	if cfg.ChunkSize != 20_000_000 || cfg.BlobSizeLimit != 1_000_000_000 ||
+		cfg.S3.PartSize != 6_000_000 {
+		t.Errorf("chunk_size, blob_size_limit, s3.part_size = %d, %d, %d, "+
+			"want 20000000, 1000000000, 6000000",
+			cfg.ChunkSize, cfg.BlobSizeLimit, cfg.S3.PartSize)
+	}
+}

internal/database/blob_chunks.go

@@ -58,12 +58,7 @@ func (r *BlobChunkRepository) GetByBlobID(
 		return nil, fmt.Errorf("querying blob chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var blobChunks []*BlobChunk
 

internal/database/blobs.go

@@ -83,12 +83,7 @@ func (r *BlobRepository) GetAll(ctx context.Context) (map[string]*Blob, error) {
 		return nil, fmt.Errorf("querying blobs: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	out := make(map[string]*Blob)
 
@@ -159,13 +154,16 @@ func (r *BlobRepository) UpdateFinished(
 	return nil
 }
 
-// UpdateUploaded marks a blob as uploaded
+// UpdateUploaded marks a blob as uploaded and records the size of the
+// stored object as its compressed size. That size can differ from the
+// local copy's when an identical blob was already stored with another
+// compression level or other recipients.
 func (r *BlobRepository) UpdateUploaded(
-	ctx context.Context, tx *sql.Tx, id string,
+	ctx context.Context, tx *sql.Tx, id string, storedSize int64,
 ) error {
 	query := `
-		UPDATE blobs 
-		SET uploaded_ts = ?
+		UPDATE blobs
+		SET uploaded_ts = ?, compressed_size = ?
 		WHERE id = ?
 	`
 
@@ -173,9 +171,9 @@ func (r *BlobRepository) UpdateUploaded(
 
 	var err error
 	if tx != nil {
-		_, err = tx.ExecContext(ctx, query, now, id)
+		_, err = tx.ExecContext(ctx, query, now, storedSize, id)
 	} else {
-		_, err = r.db.ExecWithLog(ctx, query, now, id)
+		_, err = r.db.ExecWithLog(ctx, query, now, storedSize, id)
 	}
 
 	if err != nil {

internal/database/blobs_test.go

@@ -122,8 +122,8 @@ func TestBlobRepositoryUpdates(t *testing.T) {
 		t.Errorf("expected compressed size 500, got %d", updated.CompressedSize)
 	}
 
-	// Test UpdateUploaded
-	err = repo.UpdateUploaded(ctx, nil, blob.ID.String())
+	// Test UpdateUploaded, with a stored object larger than the local copy
+	err = repo.UpdateUploaded(ctx, nil, blob.ID.String(), 600)
 	if err != nil {
 		t.Fatalf("failed to update blob as uploaded: %v", err)
 	}
@@ -141,6 +141,10 @@ func TestBlobRepositoryUpdates(t *testing.T) {
 	if uploaded.UploadedTS.Before(now.Add(-1 * time.Second)) {
 		t.Error("uploaded timestamp should be around test time")
 	}
+
+	if uploaded.CompressedSize != 600 {
+		t.Errorf("expected the stored size 600, got %d", uploaded.CompressedSize)
+	}
 }
 
 func TestBlobRepositoryDuplicate(t *testing.T) {

internal/database/chunk_files.go

@@ -61,12 +61,7 @@ func (r *ChunkFileRepository) GetByChunkHash(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanChunkFiles(rows)
 }
@@ -87,12 +82,7 @@ func (r *ChunkFileRepository) GetByFilePath(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanChunkFiles(rows)
 }
@@ -112,12 +102,7 @@ func (r *ChunkFileRepository) GetByFileID(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanChunkFiles(rows)
 }
@@ -220,7 +205,7 @@ func (r *ChunkFileRepository) CreateBatch(
 				cf.ChunkHash.String(), cf.FileID.String(), cf.FileOffset, cf.Length)
 		}
 
-		query += querySb183.String() //nolint:gosec // G202: appends "?" placeholders only
+		query += querySb183.String()
 
 		query += " ON CONFLICT(chunk_hash, file_id) DO NOTHING"
 

internal/database/chunks.go

@@ -98,7 +98,7 @@ func (r *ChunkRepository) GetByHashes(
 		args[i] = hash
 	}
 
-	query += querySb75.String() //nolint:gosec // G202: appends "?" placeholders only
+	query += querySb75.String()
 
 	query += ") ORDER BY chunk_hash"
 
@@ -107,12 +107,7 @@ func (r *ChunkRepository) GetByHashes(
 		return nil, fmt.Errorf("querying chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var chunks []*Chunk
 
@@ -152,12 +147,7 @@ func (r *ChunkRepository) ListUnpacked(
 		return nil, fmt.Errorf("querying unpacked chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var chunks []*Chunk
 

internal/database/chunks_ext.go

@@ -38,12 +38,7 @@ func (r *ChunkRepository) list(ctx context.Context, query string) ([]*Chunk, err
 		return nil, fmt.Errorf("querying chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var chunks []*Chunk
 

internal/database/database.go

@@ -42,6 +42,10 @@ var schemaFS embed.FS
 // table itself. It is applied before the normal migration loop.
 const bootstrapVersion = 0
 
+// busyTimeoutMs is how long a connection to the index waits for another
+// connection's lock before failing with "database is locked".
+const busyTimeoutMs = 10000
+
 // DB represents the Vaultik local index database connection.
 // It uses SQLite to track file metadata, content-defined chunks, and blob associations.
 // The database enables incremental backups by detecting changed files and
@@ -94,10 +98,27 @@ func ParseMigrationVersion(filename string) (int, error) {
 	return version, nil
 }
 
+// indexDSN returns the driver DSN that opens the database at path. The
+// driver runs each _pragma parameter on every connection it opens and drops
+// any parameter it does not know without an error, so a setting written in
+// another form silently does nothing. In WAL mode one connection can write
+// while others read; the busy timeout makes a connection wait for a lock
+// instead of failing at once.
+func indexDSN(path string) string {
+	return fmt.Sprintf(
+		"%s?_pragma=busy_timeout(%d)&_pragma=journal_mode(WAL)"+
+			"&_pragma=synchronous(NORMAL)&_pragma=foreign_keys(1)",
+		path, busyTimeoutMs)
+}
+
 // New creates a new database connection at the specified path.
-// It creates the schema if needed and configures SQLite with WAL mode for
-// better concurrency. SQLite handles crash recovery automatically when
-// opening a database with journal/WAL files present.
+// It creates the schema if needed. Every connection runs in WAL mode with
+// a busy timeout and foreign keys on (see indexDSN), so a read-only command
+// can read the index while a backup writes to it. Committed rows can sit in
+// the -wal file beside the database until a checkpoint, so a copy of the
+// database file alone may miss them.
+// SQLite handles crash recovery automatically when opening a database with
+// journal/WAL files present.
 // The path parameter can be a file path for persistent storage or ":memory:"
 // for an in-memory database (useful for testing).
 func New(ctx context.Context, path string) (*DB, error) {
@@ -110,11 +131,7 @@ func New(ctx context.Context, path string) (*DB, error) {
 	// First attempt with standard WAL mode
 	log.Debug("Attempting to open database with WAL mode", "path", path)
 
-	conn, err := sql.Open(
-		"sqlite",
-		path+"?_journal_mode=WAL&_synchronous=NORMAL&_busy_timeout=10000"+
-			"&_locking_mode=NORMAL&_foreign_keys=ON",
-	)
+	conn, err := sql.Open("sqlite", indexDSN(path))
 	if err == nil {
 		configureConnPool(conn)
 
@@ -134,8 +151,8 @@ func New(ctx context.Context, path string) (*DB, error) {
 		_ = conn.Close()
 	}
 
-	// If first attempt failed, try with TRUNCATE mode to clear any locks
-	return openWithRecovery(ctx, path)
+	// If the first attempt failed, try once more
+	return retryOpen(ctx, path)
 }
 
 // configureConnPool serializes all database access through one connection.
@@ -147,18 +164,12 @@ func configureConnPool(conn *sql.DB) {
 	conn.SetMaxIdleConns(1)
 }
 
-// finishOpen enables foreign keys, wraps the connection, and applies any
-// pending migrations. On migration failure the connection is closed.
+// finishOpen wraps the connection and applies any pending migrations. On
+// migration failure the connection is closed.
 func finishOpen(ctx context.Context, conn *sql.DB, path string) (*DB, error) {
-	// Enable foreign keys explicitly
-	_, err := conn.ExecContext(ctx, "PRAGMA foreign_keys = ON")
-	if err != nil {
-		log.Warn("Failed to enable foreign keys", "path", path, "error", err)
-	}
-
 	db := &DB{conn: conn, path: path}
 
-	err = applyMigrations(ctx, conn)
+	err := applyMigrations(ctx, conn)
 	if err != nil {
 		_ = conn.Close()
 
@@ -168,21 +179,15 @@ func finishOpen(ctx context.Context, conn *sql.DB, path string) (*DB, error) {
 	return db, nil
 }
 
-// openWithRecovery retries opening the database in TRUNCATE journal mode to
-// clear stale locks, then switches back to WAL mode.
-func openWithRecovery(ctx context.Context, path string) (*DB, error) {
-	log.Info(
-		"Database appears locked, attempting recovery with TRUNCATE mode",
-		"path", path,
-	)
+// retryOpen makes a second attempt to open the database, with the same
+// settings, after the first attempt failed, for example because another
+// process held a lock for longer than the busy timeout.
+func retryOpen(ctx context.Context, path string) (*DB, error) {
+	log.Info("Database appears locked, retrying open", "path", path)
 
-	conn, err := sql.Open(
-		"sqlite",
-		path+"?_journal_mode=TRUNCATE&_synchronous=NORMAL&_busy_timeout=10000"+
-			"&_foreign_keys=ON",
-	)
+	conn, err := sql.Open("sqlite", indexDSN(path))
 	if err != nil {
-		return nil, fmt.Errorf("opening database in recovery mode: %w", err)
+		return nil, fmt.Errorf("opening database on retry: %w", err)
 	}
 
 	configureConnPool(conn)
@@ -190,28 +195,18 @@ func openWithRecovery(ctx context.Context, path string) (*DB, error) {
 	err = conn.PingContext(ctx)
 	if err != nil {
 		log.Debug(
-			"Failed to ping database in recovery mode, closing",
+			"Failed to ping database on retry, closing",
 			"path", path, "error", err,
 		)
 
 		_ = conn.Close()
 
 		return nil, fmt.Errorf(
-			"database still locked after recovery attempt: %w",
+			"database still locked on retry: %w",
 			err,
 		)
 	}
 
-	log.Debug("Database opened in TRUNCATE mode", "path", path)
-
-	// Switch back to WAL mode
-	log.Debug("Switching database back to WAL mode", "path", path)
-
-	_, err = conn.ExecContext(ctx, "PRAGMA journal_mode=WAL")
-	if err != nil {
-		log.Warn("Failed to switch back to WAL mode", "path", path, "error", err)
-	}
-
 	db, err := finishOpen(ctx, conn, path)
 	if err != nil {
 		return nil, err

internal/database/database_test.go

@@ -120,6 +120,89 @@ func TestDatabaseConcurrentAccess(t *testing.T) {
 	}
 }
 
+// TestNewSetsJournalModeAndBusyTimeout checks that the connection settings
+// New passes reach SQLite. The driver drops a setting it does not recognise
+// without an error, so only reading the value back shows it took effect.
+func TestNewSetsJournalModeAndBusyTimeout(t *testing.T) {
+	t.Parallel()
+
+	ctx := context.Background()
+
+	db, err := New(ctx, filepath.Join(t.TempDir(), "index.db"))
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+
+	defer func() { _ = db.Close() }()
+
+	var journalMode string
+
+	err = db.conn.QueryRowContext(ctx, "PRAGMA journal_mode").Scan(&journalMode)
+	if err != nil {
+		t.Fatalf("reading journal_mode: %v", err)
+	}
+
+	if journalMode != "wal" {
+		t.Errorf("journal_mode = %q, want %q", journalMode, "wal")
+	}
+
+	var busyTimeout int
+
+	err = db.conn.QueryRowContext(ctx, "PRAGMA busy_timeout").Scan(&busyTimeout)
+	if err != nil {
+		t.Fatalf("reading busy_timeout: %v", err)
+	}
+
+	if busyTimeout != busyTimeoutMs {
+		t.Errorf("busy_timeout = %d, want %d", busyTimeout, busyTimeoutMs)
+	}
+}
+
+// TestNewWriteSucceedsWhileAnotherHandleReads opens the same index twice,
+// as a read-only command does while a backup runs, and checks that a write
+// on one handle commits while the other is in the middle of a read.
+func TestNewWriteSucceedsWhileAnotherHandleReads(t *testing.T) {
+	t.Parallel()
+
+	ctx := context.Background()
+	dbPath := filepath.Join(t.TempDir(), "index.db")
+
+	reader, err := New(ctx, dbPath)
+	if err != nil {
+		t.Fatalf("failed to open reading handle: %v", err)
+	}
+
+	defer func() { _ = reader.Close() }()
+
+	writer, err := New(ctx, dbPath)
+	if err != nil {
+		t.Fatalf("failed to open writing handle: %v", err)
+	}
+
+	defer func() { _ = writer.Close() }()
+
+	// The read lock taken by the SELECT is held until the transaction ends.
+	readTx, err := reader.BeginTx(ctx, nil)
+	if err != nil {
+		t.Fatalf("beginning read transaction: %v", err)
+	}
+
+	defer func() { _ = readTx.Rollback() }()
+
+	var count int
+
+	err = readTx.QueryRowContext(ctx, "SELECT COUNT(*) FROM chunks").Scan(&count)
+	if err != nil {
+		t.Fatalf("reading chunks: %v", err)
+	}
+
+	_, err = writer.ExecWithLog(ctx,
+		"INSERT INTO chunks (chunk_hash, size) VALUES (?, ?)", "hash", 1024)
+	if err != nil {
+		t.Fatalf("write while another handle reads: %v", err)
+	}
+}
+
 func TestParseMigrationVersion(t *testing.T) {
 	t.Parallel()
 

internal/database/errors.go

@@ -1,12 +0,0 @@
-package database
-
-import (
-	"fmt"
-	"os"
-)
-
-// Fatalf prints an error message to stderr and exits with status 1
-func Fatalf(format string, args ...any) {
-	fmt.Fprintf(os.Stderr, "FATAL: "+format+"\n", args...)
-	os.Exit(1)
-}

internal/database/file_chunks.go

@@ -62,12 +62,7 @@ func (r *FileChunkRepository) GetByPath(
 		return nil, fmt.Errorf("querying file chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanFileChunks(rows)
 }
@@ -88,12 +83,7 @@ func (r *FileChunkRepository) GetByFileID(
 		return nil, fmt.Errorf("querying file chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanFileChunks(rows)
 }
@@ -117,12 +107,7 @@ func (r *FileChunkRepository) GetByPathTx(
 		return nil, fmt.Errorf("querying file chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	fileChunks, err := r.scanFileChunks(rows)
 	LogSQL("GetByPathTx", "Complete", path, "count", len(fileChunks))
@@ -253,7 +238,7 @@ func (r *FileChunkRepository) CreateBatch(
 			args = append(args, fc.FileID.String(), fc.Idx, fc.ChunkHash.String())
 		}
 
-		query += querySb211.String() //nolint:gosec // G202: appends "?" placeholders only
+		query += querySb211.String()
 
 		query += " ON CONFLICT(file_id, idx) DO NOTHING"
 

internal/database/files.go

@@ -33,11 +33,13 @@ func (r *FileRepository) Create(ctx context.Context, tx *sql.Tx, file *File) err
 	}
 
 	query := `
-		INSERT INTO files (id, path, source_path, mtime, size, mode, uid, gid, link_target)
-		VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
+		INSERT INTO files
+			(id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target)
+		VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
 		ON CONFLICT(path) DO UPDATE SET
 			source_path = excluded.source_path,
 			mtime = excluded.mtime,
+			mtime_nsec = excluded.mtime_nsec,
 			size = excluded.size,
 			mode = excluded.mode,
 			uid = excluded.uid,
@@ -54,16 +56,19 @@ func (r *FileRepository) Create(ctx context.Context, tx *sql.Tx, file *File) err
 	if tx != nil {
 		LogSQL("Execute", query,
 			file.ID.String(), file.Path.String(), file.SourcePath.String(),
-			file.MTime.Unix(), file.Size, file.Mode, file.UID, file.GID,
+			file.MTime.Unix(), file.MTime.Nanosecond(),
+			file.Size, file.Mode, file.UID, file.GID,
 			file.LinkTarget.String())
 		err = tx.QueryRowContext(ctx, query,
 			file.ID.String(), file.Path.String(), file.SourcePath.String(),
-			file.MTime.Unix(), file.Size, file.Mode, file.UID, file.GID,
+			file.MTime.Unix(), file.MTime.Nanosecond(),
+			file.Size, file.Mode, file.UID, file.GID,
 			file.LinkTarget.String()).Scan(&idStr)
 	} else {
 		err = r.db.QueryRowWithLog(ctx, query,
 			file.ID.String(), file.Path.String(), file.SourcePath.String(),
-			file.MTime.Unix(), file.Size, file.Mode, file.UID, file.GID,
+			file.MTime.Unix(), file.MTime.Nanosecond(),
+			file.Size, file.Mode, file.UID, file.GID,
 			file.LinkTarget.String()).Scan(&idStr)
 	}
 
@@ -84,7 +89,7 @@ func (r *FileRepository) Create(ctx context.Context, tx *sql.Tx, file *File) err
 // in the index.
 func (r *FileRepository) GetByPath(ctx context.Context, path string) (*File, error) {
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
 		WHERE path = ?
 	`
@@ -104,7 +109,7 @@ func (r *FileRepository) GetByPath(ctx context.Context, path string) (*File, err
 // GetByID retrieves a file by its UUID
 func (r *FileRepository) GetByID(ctx context.Context, id types.FileID) (*File, error) {
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
 		WHERE id = ?
 	`
@@ -127,7 +132,7 @@ func (r *FileRepository) GetByPathTx(
 	ctx context.Context, tx *sql.Tx, path string,
 ) (*File, error) {
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
 		WHERE path = ?
 	`
@@ -158,23 +163,19 @@ func (r *FileRepository) ListModifiedSince(
 	ctx context.Context, since time.Time,
 ) ([]*File, error) {
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
-		WHERE mtime >= ?
+		WHERE (mtime, mtime_nsec) >= (?, ?)
 		ORDER BY path
 	`
 
-	rows, err := r.db.conn.QueryContext(ctx, query, since.Unix())
+	rows, err := r.db.conn.QueryContext(ctx, query,
+		since.Unix(), since.Nanosecond())
 	if err != nil {
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var files []*File
 
@@ -228,29 +229,29 @@ func (r *FileRepository) DeleteByID(
 	return nil
 }
 
-// ListByPrefix returns all files whose path starts with prefix, ordered by
-// path.
-func (r *FileRepository) ListByPrefix(
-	ctx context.Context, prefix string,
+// ListUnderPath returns the file at path and every file beneath it,
+// ordered by path. Paths are compared case-sensitively, and a trailing
+// slash on path is ignored, so "/" lists every file.
+func (r *FileRepository) ListUnderPath(
+	ctx context.Context, path string,
 ) ([]*File, error) {
+	path = strings.TrimRight(path, "/")
+	dirPrefix := path + "/"
+
+	// LIKE would ignore ASCII case and treat _ and % in path as wildcards.
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
-		WHERE path LIKE ? || '%'
+		WHERE path = ? OR substr(path, 1, length(?)) = ?
 		ORDER BY path
 	`
 
-	rows, err := r.db.conn.QueryContext(ctx, query, prefix)
+	rows, err := r.db.conn.QueryContext(ctx, query, path, dirPrefix, dirPrefix)
 	if err != nil {
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var files []*File
 
@@ -266,10 +267,94 @@ func (r *FileRepository) ListByPrefix(
 	return files, rows.Err()
 }
 
+// ListUnderPathInSnapshot returns the files that ListUnderPath returns
+// for path and that the snapshot lists, in no particular order.
+func (r *FileRepository) ListUnderPathInSnapshot(
+	ctx context.Context, path, snapshotID string,
+) ([]*File, error) {
+	path = strings.TrimRight(path, "/")
+	dirPrefix := path + "/"
+
+	query := `
+		SELECT f.id, f.path, f.source_path, f.mtime, f.mtime_nsec, f.size,
+			f.mode, f.uid, f.gid, f.link_target
+		FROM files f
+		JOIN snapshot_files sf ON sf.file_id = f.id
+		WHERE sf.snapshot_id = ?
+		  AND (f.path = ? OR substr(f.path, 1, length(?)) = ?)
+	`
+
+	rows, err := r.db.conn.QueryContext(ctx, query,
+		snapshotID, path, dirPrefix, dirPrefix)
+	if err != nil {
+		return nil, fmt.Errorf("querying files: %w", err)
+	}
+
+	defer func() { _ = rows.Close() }()
+
+	var files []*File
+
+	for rows.Next() {
+		file, err := r.scanFileRows(rows)
+		if err != nil {
+			return nil, fmt.Errorf("scanning file: %w", err)
+		}
+
+		files = append(files, file)
+	}
+
+	return files, rows.Err()
+}
+
+// ListIDsWithChunksNotInUploadedBlobs returns the IDs of the files whose
+// path starts with prefix and that list at least one chunk held by no
+// blob whose upload has completed (uploaded_ts set). A new snapshot
+// cannot reference such a chunk, so a backup must not treat the file as
+// unchanged even when its metadata matches the file on disk.
+func (r *FileRepository) ListIDsWithChunksNotInUploadedBlobs(
+	ctx context.Context, prefix string,
+) ([]types.FileID, error) {
+	query := `
+		SELECT DISTINCT f.id
+		FROM files f
+		JOIN file_chunks fc ON fc.file_id = f.id
+		WHERE f.path LIKE ? || '%'
+		  AND NOT EXISTS (
+			SELECT 1
+			FROM blob_chunks bc
+			JOIN blobs b ON bc.blob_id = b.id
+			WHERE bc.chunk_hash = fc.chunk_hash
+			  AND b.uploaded_ts IS NOT NULL
+		  )
+	`
+
+	rows, err := r.db.conn.QueryContext(ctx, query, prefix)
+	if err != nil {
+		return nil, fmt.Errorf("querying files: %w", err)
+	}
+
+	defer func() { _ = rows.Close() }()
+
+	var ids []types.FileID
+
+	for rows.Next() {
+		var id types.FileID
+
+		err := rows.Scan(&id)
+		if err != nil {
+			return nil, fmt.Errorf("scanning file ID: %w", err)
+		}
+
+		ids = append(ids, id)
+	}
+
+	return ids, rows.Err()
+}
+
 // ListAll returns all files in the database
 func (r *FileRepository) ListAll(ctx context.Context) ([]*File, error) {
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
 		ORDER BY path
 	`
@@ -279,12 +364,7 @@ func (r *FileRepository) ListAll(ctx context.Context) ([]*File, error) {
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var files []*File
 
@@ -310,7 +390,7 @@ func (r *FileRepository) CreateBatch(
 	}
 
 	// Each files row binds this many SQL variables.
-	const fileCols = 9
+	const fileCols = 10
 
 	// Batch at 100 rows to be safe with SQLite's variable limit.
 	const batchSize = 100
@@ -321,7 +401,7 @@ func (r *FileRepository) CreateBatch(
 		batch := files[i:end]
 
 		query := `INSERT INTO files
-			(id, path, source_path, mtime, size, mode, uid, gid, link_target)
+			(id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target)
 			VALUES `
 
 		args := make([]any, 0, len(batch)*fileCols)
@@ -333,19 +413,21 @@ func (r *FileRepository) CreateBatch(
 				querySb325.WriteString(", ")
 			}
 
-			querySb325.WriteString("(?, ?, ?, ?, ?, ?, ?, ?, ?)")
+			querySb325.WriteString("(?, ?, ?, ?, ?, ?, ?, ?, ?, ?)")
 
 			args = append(args,
 				f.ID.String(), f.Path.String(), f.SourcePath.String(),
-				f.MTime.Unix(), f.Size, f.Mode, f.UID, f.GID,
+				f.MTime.Unix(), f.MTime.Nanosecond(),
+				f.Size, f.Mode, f.UID, f.GID,
 				f.LinkTarget.String())
 		}
 
-		query += querySb325.String() //nolint:gosec // G202: appends "?" placeholders only
+		query += querySb325.String()
 
 		query += ` ON CONFLICT(path) DO UPDATE SET
 			source_path = excluded.source_path,
 			mtime = excluded.mtime,
+			mtime_nsec = excluded.mtime_nsec,
 			size = excluded.size,
 			mode = excluded.mode,
 			uid = excluded.uid,
@@ -405,7 +487,7 @@ func (r *FileRepository) scanFileFrom(row fileRowScanner) (*File, error) {
 	var (
 		file                          File
 		idStr, pathStr, sourcePathStr string
-		mtimeUnix                     int64
+		mtimeUnix, mtimeNsec          int64
 		linkTarget                    sql.NullString
 	)
 
@@ -414,6 +496,7 @@ func (r *FileRepository) scanFileFrom(row fileRowScanner) (*File, error) {
 		&pathStr,
 		&sourcePathStr,
 		&mtimeUnix,
+		&mtimeNsec,
 		&file.Size,
 		&file.Mode,
 		&file.UID,
@@ -432,7 +515,7 @@ func (r *FileRepository) scanFileFrom(row fileRowScanner) (*File, error) {
 	file.Path = types.FilePath(pathStr)
 	file.SourcePath = types.SourcePath(sourcePathStr)
 
-	file.MTime = time.Unix(mtimeUnix, 0).UTC()
+	file.MTime = time.Unix(mtimeUnix, mtimeNsec).UTC()
 	if linkTarget.Valid {
 		file.LinkTarget = types.FilePath(linkTarget.String)
 	}

internal/database/files_test.go

@@ -5,10 +5,12 @@ import (
 	"database/sql"
 	"errors"
 	"os"
+	"slices"
 	"testing"
 	"time"
 
 	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/types"
 )
 
 // errTestRollback is the sentinel returned from transaction bodies to
@@ -134,6 +136,141 @@ func TestFileRepositoryListDelete(t *testing.T) {
 	}
 }
 
+func TestFileRepositoryListUnderPath(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repo := database.NewFileRepository(db)
+
+	const (
+		docDir  = "/home/u/doc"
+		docFile = "/home/u/doc/a.txt"
+	)
+
+	// In path order, so the root case can expect all of them as listed.
+	paths := []string{
+		"/home/u/50%/x.txt",
+		"/home/u/50percent/y.txt",
+		"/home/u/DOC/c.txt",
+		"/home/u/a_b/x.txt",
+		"/home/u/axb/y.txt",
+		docDir,
+		"/home/u/doc.txt.bak",
+		docFile,
+		"/home/u/doc/sub/b.txt",
+		"/home/u/doc2/b.txt",
+	}
+
+	for _, path := range paths {
+		err := repo.Create(ctx, nil, &database.File{
+			Path:  types.FilePath(path),
+			MTime: time.Now().Truncate(time.Second),
+			Mode:  0644,
+		})
+		if err != nil {
+			t.Fatalf("failed to create %s: %v", path, err)
+		}
+	}
+
+	docTree := []string{docDir, docFile, "/home/u/doc/sub/b.txt"}
+
+	tests := []struct {
+		name string
+		path string
+		want []string
+	}{
+		{"directory", docDir, docTree},
+		{"directory with trailing slash", docDir + "/", docTree},
+		{"directory differing only in case", "/home/u/DOC",
+			[]string{"/home/u/DOC/c.txt"}},
+		{"file", docFile, []string{docFile}},
+		{"underscore is literal", "/home/u/a_b",
+			[]string{"/home/u/a_b/x.txt"}},
+		{"percent is literal", "/home/u/50%",
+			[]string{"/home/u/50%/x.txt"}},
+		{"root", "/", paths},
+	}
+
+	for _, tt := range tests {
+		files, err := repo.ListUnderPath(ctx, tt.path)
+		if err != nil {
+			t.Fatalf("%s: failed to list files: %v", tt.name, err)
+		}
+
+		got := make([]string, 0, len(files))
+		for _, f := range files {
+			got = append(got, f.Path.String())
+		}
+
+		if !slices.Equal(got, tt.want) {
+			t.Errorf("%s: listing %q got %q, want %q",
+				tt.name, tt.path, got, tt.want)
+		}
+	}
+}
+
+// ListUnderPathInSnapshot leaves out files outside path and files the
+// snapshot does not list, though another snapshot lists them.
+func TestFileRepositoryListUnderPathInSnapshot(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	files := database.NewFileRepository(db)
+	snapshots := database.NewSnapshotRepository(db)
+
+	listed := map[string][]string{
+		"one": {"/src/a.txt", "/src2/c.txt"},
+		"two": {"/src/b.txt"},
+	}
+
+	for snapshotID, paths := range listed {
+		err := snapshots.Create(ctx, nil, &database.Snapshot{
+			ID:        types.SnapshotID(snapshotID),
+			Hostname:  "host",
+			StartedAt: time.Now(),
+		})
+		if err != nil {
+			t.Fatalf("failed to create snapshot %s: %v", snapshotID, err)
+		}
+
+		for _, path := range paths {
+			err := files.Create(ctx, nil, &database.File{
+				Path:  types.FilePath(path),
+				MTime: time.Now().Truncate(time.Second),
+				Mode:  0644,
+			})
+			if err != nil {
+				t.Fatalf("failed to create %s: %v", path, err)
+			}
+
+			err = snapshots.AddFile(ctx, nil, snapshotID, path)
+			if err != nil {
+				t.Fatalf("failed to add %s to %s: %v", path, snapshotID, err)
+			}
+		}
+	}
+
+	found, err := files.ListUnderPathInSnapshot(ctx, "/src", "one")
+	if err != nil {
+		t.Fatalf("failed to list files: %v", err)
+	}
+
+	got := make([]string, 0, len(found))
+	for _, f := range found {
+		got = append(got, f.Path.String())
+	}
+
+	if !slices.Equal(got, []string{"/src/a.txt"}) {
+		t.Errorf("got %q, want only /src/a.txt", got)
+	}
+}
+
 func TestFileRepositorySymlink(t *testing.T) {
 	t.Parallel()
 
@@ -174,6 +311,115 @@ func TestFileRepositorySymlink(t *testing.T) {
 	}
 }
 
+// An mtime after 2262 or before 1678 does not fit in int64 nanoseconds
+// since the epoch, and must still come back from the database unchanged.
+func TestFileRepositoryMTimeOutsideInt64NanosecondRange(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repo := database.NewFileRepository(db)
+
+	mtimes := []time.Time{
+		time.Date(2300, time.January, 1, 0, 0, 0, 123456789, time.UTC),
+		time.Date(1601, time.January, 1, 0, 0, 0, 987654321, time.UTC),
+	}
+
+	for _, mtime := range mtimes {
+		created := &database.File{
+			Path:  types.FilePath("/created-" + mtime.Format(time.RFC3339Nano)),
+			MTime: mtime,
+		}
+
+		err := repo.Create(ctx, nil, created)
+		if err != nil {
+			t.Fatalf("failed to create file: %v", err)
+		}
+
+		batched := &database.File{
+			ID:    types.NewFileID(),
+			Path:  types.FilePath("/batched-" + mtime.Format(time.RFC3339Nano)),
+			MTime: mtime,
+		}
+
+		err = repo.CreateBatch(ctx, nil, []*database.File{batched})
+		if err != nil {
+			t.Fatalf("failed to batch create file: %v", err)
+		}
+
+		for _, path := range []types.FilePath{created.Path, batched.Path} {
+			retrieved, err := repo.GetByPath(ctx, path.String())
+			if err != nil {
+				t.Fatalf("failed to get file: %v", err)
+			}
+
+			if !retrieved.MTime.Equal(mtime) {
+				t.Errorf("%s: mtime got %v, want %v",
+					path, retrieved.MTime, mtime)
+			}
+		}
+	}
+}
+
+// A file already in the index and rewritten within the same second must get
+// its new nanoseconds stored, through both Create and CreateBatch.
+func TestFileRepositoryUpsertMTimeInSameSecond(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repo := database.NewFileRepository(db)
+
+	indexed := time.Date(2026, time.October, 7, 12, 0, 0, 100000000, time.UTC)
+	rewritten := indexed.Add(800 * time.Millisecond)
+
+	tests := []struct {
+		name   string
+		upsert func(file *database.File) error
+	}{
+		{"Create", func(file *database.File) error {
+			return repo.Create(ctx, nil, file)
+		}},
+		{"CreateBatch", func(file *database.File) error {
+			return repo.CreateBatch(ctx, nil, []*database.File{file})
+		}},
+	}
+
+	for _, tt := range tests {
+		file := &database.File{
+			ID:    types.NewFileID(),
+			Path:  types.FilePath("/" + tt.name),
+			MTime: indexed,
+		}
+
+		err := tt.upsert(file)
+		if err != nil {
+			t.Fatalf("%s: failed to create file: %v", tt.name, err)
+		}
+
+		file.MTime = rewritten
+
+		err = tt.upsert(file)
+		if err != nil {
+			t.Fatalf("%s: failed to update file: %v", tt.name, err)
+		}
+
+		retrieved, err := repo.GetByPath(ctx, file.Path.String())
+		if err != nil {
+			t.Fatalf("%s: failed to get file: %v", tt.name, err)
+		}
+
+		if !retrieved.MTime.Equal(rewritten) {
+			t.Errorf("%s: mtime got %v, want %v",
+				tt.name, retrieved.MTime, rewritten)
+		}
+	}
+}
+
 func TestFileRepositoryTransaction(t *testing.T) {
 	t.Parallel()
 

internal/database/models.go

@@ -14,8 +14,7 @@ type File struct {
 	ID   types.FileID   // UUID primary key
 	Path types.FilePath // Absolute path of the file
 
-	// SourcePath is the source directory this file came from (used for
-	// restore path stripping).
+	// SourcePath is the source directory this file came from.
 	SourcePath types.SourcePath
 	MTime      time.Time
 	Size       int64
@@ -99,8 +98,8 @@ type Snapshot struct {
 	StartedAt          time.Time
 	CompletedAt        *time.Time // nil if still in progress
 	FileCount          int64
-	ChunkCount         int64
-	BlobCount          int64
+	ChunkCount         int64 // Chunks this snapshot stored that were not stored before
+	BlobCount          int64 // Blobs this snapshot created
 	TotalSize          int64 // Total size of all referenced files
 
 	// BlobSize is the total size of all referenced blobs (compressed and

internal/database/module.go

@@ -20,6 +20,17 @@ const indexDirPerm = 0o700
 // backed-up path and chunk hash and must stay private.
 const indexFilePerm = 0o600
 
+// IndexFiles returns the path of the index at indexPath and the paths of
+// the side files SQLite keeps beside it while it writes.
+func IndexFiles(indexPath string) []string {
+	return []string{
+		indexPath,
+		indexPath + "-wal",
+		indexPath + "-shm",
+		indexPath + "-journal",
+	}
+}
+
 // ensureIndexFileMode makes the index file owner-only before the SQLite
 // driver opens it: it creates the file 0600 if absent, or chmods an existing
 // one to 0600. Doing this first matters because SQLite creates its -wal and

internal/database/repositories.go

@@ -3,6 +3,7 @@ package database
 import (
 	"context"
 	"database/sql"
+	"errors"
 	"fmt"
 )
 
@@ -60,22 +61,24 @@ func (r *Repositories) WithTx(ctx context.Context, fn TxFunc) error {
 
 	defer func() {
 		if p := recover(); p != nil {
-			rollbackErr := tx.Rollback()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
+			// The panic is the failure to report, so a rollback error
+			// is dropped.
+			_ = tx.Rollback()
 
 			panic(p)
-		} else if err != nil {
-			rollbackErr := tx.Rollback()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
 		}
 	}()
 
 	err = fn(ctx, tx)
 	if err != nil {
+		// Once ctx is cancelled, database/sql rolls the transaction
+		// back itself and Rollback returns sql.ErrTxDone.
+		rollbackErr := tx.Rollback()
+		if rollbackErr != nil && !errors.Is(rollbackErr, sql.ErrTxDone) {
+			return fmt.Errorf("%w; rolling back transaction: %w",
+				err, rollbackErr)
+		}
+
 		return err
 	}
 
@@ -103,22 +106,24 @@ func (r *Repositories) WithReadTx(ctx context.Context, fn TxFunc) error {
 
 	defer func() {
 		if p := recover(); p != nil {
-			rollbackErr := tx.Rollback()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
+			// The panic is the failure to report, so a rollback error
+			// is dropped.
+			_ = tx.Rollback()
 
 			panic(p)
-		} else if err != nil {
-			rollbackErr := tx.Rollback()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
 		}
 	}()
 
 	err = fn(ctx, tx)
 	if err != nil {
+		// Once ctx is cancelled, database/sql rolls the transaction
+		// back itself and Rollback returns sql.ErrTxDone.
+		rollbackErr := tx.Rollback()
+		if rollbackErr != nil && !errors.Is(rollbackErr, sql.ErrTxDone) {
+			return fmt.Errorf("%w; rolling back transaction: %w",
+				err, rollbackErr)
+		}
+
 		return err
 	}
 

internal/database/repositories_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"database/sql"
 	"errors"
+	"strings"
 	"testing"
 	"time"
 
@@ -235,6 +236,125 @@ func TestRepositoriesTransactionRollback(t *testing.T) {
 	}
 }
 
+// When the context is cancelled during the body, database/sql rolls the
+// transaction back itself, so WithTx's own rollback finds it already
+// done. WithTx must return the body's error alone, without that rollback.
+func TestRepositoriesTransactionContextCancelled(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	repos := database.NewRepositories(db)
+
+	err := repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		file := &database.File{
+			Path:  "/test/cancelled_file.txt",
+			MTime: time.Now().Truncate(time.Second),
+			Size:  1024,
+			Mode:  0644,
+			UID:   1000,
+			GID:   1000,
+		}
+
+		err := repos.Files.Create(ctx, tx, file)
+		if err != nil {
+			return err
+		}
+
+		cancel()
+
+		// database/sql rolls back from a goroutine of its own; wait
+		// until it has, so the test does not depend on that race.
+		for {
+			_, err = tx.ExecContext(context.WithoutCancel(ctx), "SELECT 1")
+			if errors.Is(err, sql.ErrTxDone) {
+				return ctx.Err()
+			}
+
+			time.Sleep(time.Millisecond)
+		}
+	})
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("expected context.Canceled, got: %v", err)
+	}
+
+	if errors.Is(err, sql.ErrTxDone) {
+		t.Fatalf("the already-done rollback was reported: %v", err)
+	}
+
+	file, err := repos.Files.GetByPath(
+		context.Background(), "/test/cancelled_file.txt",
+	)
+	if err != nil {
+		t.Fatalf("error checking for file: %v", err)
+	}
+
+	if file != nil {
+		t.Error("file should not exist after the cancelled transaction")
+	}
+}
+
+// A body that ends the transaction itself with ROLLBACK makes WithTx's
+// own rollback fail. WithTx must return that failure together with the
+// body's error.
+func TestRepositoriesTransactionRollbackFails(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repos := database.NewRepositories(db)
+
+	err := repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		_, err := tx.ExecContext(ctx, "ROLLBACK")
+		if err != nil {
+			return err
+		}
+
+		return errIntentionalRollback
+	})
+	if !errors.Is(err, errIntentionalRollback) {
+		t.Fatalf("expected the body's error, got: %v", err)
+	}
+
+	if !strings.Contains(err.Error(), "rolling back transaction") {
+		t.Errorf("expected the failed rollback in the error, got: %v", err)
+	}
+}
+
+// When the body ends the transaction itself with ROLLBACK and then
+// panics, WithTx's own rollback fails; the panic must still reach the
+// caller.
+func TestRepositoriesTransactionPanicRollbackFails(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repos := database.NewRepositories(db)
+
+	defer func() {
+		if recover() == nil {
+			t.Error("expected the body's panic to reach the caller")
+		}
+	}()
+
+	_ = repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		_, err := tx.ExecContext(ctx, "ROLLBACK")
+		if err != nil {
+			return err
+		}
+
+		panic(errIntentionalRollback)
+	})
+}
+
 func TestRepositoriesReadTransaction(t *testing.T) {
 	t.Parallel()
 
@@ -291,3 +411,41 @@ func TestRepositoriesReadTransaction(t *testing.T) {
 		t.Error("expected to retrieve file in read transaction")
 	}
 }
+
+// When the context is cancelled during a read transaction's body,
+// database/sql rolls the transaction back itself, so WithReadTx's own
+// rollback finds it already done. WithReadTx must return the body's error
+// alone, without that rollback.
+func TestRepositoriesReadTransactionContextCancelled(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	repos := database.NewRepositories(db)
+
+	err := repos.WithReadTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		cancel()
+
+		// database/sql rolls back from a goroutine of its own; wait
+		// until it has, so the test does not depend on that race.
+		for {
+			_, err := tx.ExecContext(context.WithoutCancel(ctx), "SELECT 1")
+			if errors.Is(err, sql.ErrTxDone) {
+				return ctx.Err()
+			}
+
+			time.Sleep(time.Millisecond)
+		}
+	})
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("expected context.Canceled, got: %v", err)
+	}
+
+	if errors.Is(err, sql.ErrTxDone) {
+		t.Fatalf("the already-done rollback was reported: %v", err)
+	}
+}

internal/database/repository_comprehensive_test.go

@@ -824,7 +824,7 @@ func TestTransactionIsolation(t *testing.T) {
 	}
 
 	// Verify the file was not created (transaction rolled back)
-	files, err := repos.Files.ListByPrefix(ctx, "/tx-test")
+	files, err := repos.Files.ListUnderPath(ctx, "/tx-test.txt")
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -916,7 +916,7 @@ func TestConcurrentOrphanedCleanup(t *testing.T) {
 	}
 
 	// Verify correct files were deleted
-	files, err := repos.Files.ListByPrefix(ctx, "/concurrent-")
+	files, err := repos.Files.ListAll(ctx)
 	if err != nil {
 		t.Fatal(err)
 	}

internal/database/repository_debug_test.go

@@ -147,7 +147,7 @@ func TestOrphanedFileCleanupDebug(t *testing.T) {
 	t.Logf("Files count after cleanup: %d", count)
 
 	// List remaining files
-	files, err := repos.Files.ListByPrefix(ctx, "/")
+	files, err := repos.Files.ListUnderPath(ctx, "/")
 	if err != nil {
 		t.Fatal(err)
 	}

internal/database/repository_edge_cases_test.go

@@ -3,6 +3,7 @@ package database
 
 import (
 	"context"
+	"database/sql"
 	"fmt"
 	"strings"
 	"testing"
@@ -367,7 +368,7 @@ func verifyBlobNullUploadTS(
 }
 
 // createLargeDatasetFiles creates fileCount files and adds every other
-// one to the snapshot.
+// one to the snapshot, in one transaction as a backup writes them.
 func createLargeDatasetFiles(
 	t *testing.T,
 	repos *Repositories,
@@ -376,31 +377,38 @@ func createLargeDatasetFiles(
 ) {
 	t.Helper()
 
-	ctx := context.Background()
 	start := time.Now()
 
-	for i := range fileCount {
-		file := &File{
-			Path:  types.FilePath(fmt.Sprintf("/large/file%05d.txt", i)),
-			MTime: time.Now(),
-			Size:  int64(i * 1024),
-			Mode:  0644,
-			UID:   uint32(1000 + (i % 10)),
-			GID:   uint32(1000 + (i % 10)),
-		}
+	err := repos.WithTx(context.Background(),
+		func(ctx context.Context, tx *sql.Tx) error {
+			for i := range fileCount {
+				file := &File{
+					Path:  types.FilePath(fmt.Sprintf("/large/file%05d.txt", i)),
+					MTime: time.Now(),
+					Size:  int64(i * 1024),
+					Mode:  0644,
+					UID:   uint32(1000 + (i % 10)),
+					GID:   uint32(1000 + (i % 10)),
+				}
 
-		err := repos.Files.Create(ctx, nil, file)
-		if err != nil {
-			t.Fatalf("failed to create file %d: %v", i, err)
-		}
+				err := repos.Files.Create(ctx, tx, file)
+				if err != nil {
+					return fmt.Errorf("creating file %d: %w", i, err)
+				}
 
-		// Add half to snapshot
-		if i%2 == 0 {
-			err = repos.Snapshots.AddFileByID(ctx, nil, snapshotID, file.ID)
-			if err != nil {
-				t.Fatal(err)
+				// Add half to snapshot
+				if i%2 == 0 {
+					err = repos.Snapshots.AddFileByID(ctx, tx, snapshotID, file.ID)
+					if err != nil {
+						return err
+					}
+				}
 			}
-		}
+
+			return nil
+		})
+	if err != nil {
+		t.Fatal(err)
 	}
 
 	t.Logf("Created %d files in %v", fileCount, time.Since(start))
@@ -442,12 +450,12 @@ func TestLargeDatasets(t *testing.T) {
 		createLargeDatasetFiles(t, repos, snapshot.ID.String(), fileCount)
 	})
 
-	// Test ListByPrefix performance
+	// Test ListUnderPath performance
 	//nolint:paralleltest // phases share one database and are order-dependent
-	t.Run("list by prefix performance", func(t *testing.T) {
+	t.Run("list under path performance", func(t *testing.T) {
 		start := time.Now()
 
-		files, err := repos.Files.ListByPrefix(ctx, "/large/")
+		files, err := repos.Files.ListUnderPath(ctx, "/large/")
 		if err != nil {
 			t.Fatal(err)
 		}
@@ -472,7 +480,7 @@ func TestLargeDatasets(t *testing.T) {
 		t.Logf("Cleaned up orphaned files in %v", time.Since(start))
 
 		// Verify correct number remain
-		files, err := repos.Files.ListByPrefix(ctx, "/large/")
+		files, err := repos.Files.ListUnderPath(ctx, "/large/")
 		if err != nil {
 			t.Fatal(err)
 		}
@@ -606,8 +614,7 @@ func TestTimezoneHandling(t *testing.T) {
 		t.Skip("timezone not available")
 	}
 
-	// Use Truncate to remove sub-second precision since we store as Unix timestamps
-	nyTime := time.Now().In(loc).Truncate(time.Second)
+	nyTime := time.Now().In(loc)
 	file := &File{
 		Path:  "/timezone-test.txt",
 		MTime: nyTime,

internal/database/schema/001.sql

@@ -5,8 +5,9 @@
 CREATE TABLE IF NOT EXISTS files (
     id TEXT PRIMARY KEY,  -- UUID
     path TEXT NOT NULL UNIQUE,
-    source_path TEXT NOT NULL DEFAULT '',  -- The source directory this file came from (for restore path stripping)
-    mtime INTEGER NOT NULL,
+    source_path TEXT NOT NULL DEFAULT '',  -- The source directory this file came from
+    mtime INTEGER NOT NULL,  -- whole seconds since the Unix epoch
+    mtime_nsec INTEGER NOT NULL,  -- nanoseconds within that second, 0 to 999999999
     size INTEGER NOT NULL,
     mode INTEGER NOT NULL,
     uid INTEGER NOT NULL,
@@ -85,8 +86,10 @@ CREATE TABLE IF NOT EXISTS snapshots (
     completed_at INTEGER,
     file_count INTEGER NOT NULL DEFAULT 0,
     chunk_count INTEGER NOT NULL DEFAULT 0,
+    new_chunk_size INTEGER NOT NULL DEFAULT 0,
     blob_count INTEGER NOT NULL DEFAULT 0,
     total_size INTEGER NOT NULL DEFAULT 0,
+    uncompressed_chunk_size INTEGER NOT NULL DEFAULT 0,
     blob_size INTEGER NOT NULL DEFAULT 0,
     blob_uncompressed_size INTEGER NOT NULL DEFAULT 0,
     compression_ratio REAL NOT NULL DEFAULT 1.0,
@@ -120,6 +123,16 @@ CREATE TABLE IF NOT EXISTS snapshot_blobs (
 -- Index for efficient blob lookups (used in orphan detection)
 CREATE INDEX IF NOT EXISTS idx_snapshot_blobs_blob_id ON snapshot_blobs(blob_id);
 
+-- Snapshot mount points table: the directories at or below a configured
+-- path that the snapshot's backup found on a different filesystem from
+-- their parent directory, as a mount point is, and found an entry in
+CREATE TABLE IF NOT EXISTS snapshot_mount_points (
+    snapshot_id TEXT NOT NULL,
+    path TEXT NOT NULL,
+    PRIMARY KEY (snapshot_id, path),
+    FOREIGN KEY (snapshot_id) REFERENCES snapshots(id) ON DELETE CASCADE
+);
+
 -- Uploads table: tracks blob upload metrics
 CREATE TABLE IF NOT EXISTS uploads (
     blob_hash TEXT PRIMARY KEY,

internal/database/snapshots.go

@@ -24,7 +24,8 @@ var (
 )
 
 // SnapshotRepository provides access to the snapshots table and its
-// snapshot_files / snapshot_blobs association tables.
+// snapshot_files / snapshot_blobs association tables, and to the mount
+// points each snapshot's backup found (snapshot_mount_points).
 type SnapshotRepository struct {
 	db *DB
 }
@@ -120,14 +121,20 @@ func (r *SnapshotRepository) UpdateCounts(
 	return nil
 }
 
-// UpdateExtendedStats updates extended statistics for a snapshot
+// UpdateExtendedStats updates extended statistics for a snapshot. It also
+// records the total plaintext size of the distinct chunks the snapshot's
+// files reference. Only the snapshot's own backup can take that total: a
+// file's row is shared by every snapshot that lists it, and a later
+// backup of a changed file rewrites the row with the new chunks.
 func (r *SnapshotRepository) UpdateExtendedStats(
 	ctx context.Context,
 	tx *sql.Tx,
 	snapshotID string,
 	blobUncompressedSize int64,
 	compressionLevel int,
+	uploadBytes int64,
 	uploadDurationMs int64,
+	newChunkSize int64,
 ) error {
 	compressionRatio, err := r.extendedCompressionRatio(
 		ctx, tx, snapshotID, blobUncompressedSize,
@@ -141,19 +148,30 @@ func (r *SnapshotRepository) UpdateExtendedStats(
 		SET blob_uncompressed_size = ?,
 		    compression_ratio = ?,
 		    compression_level = ?,
-		    upload_bytes = blob_size,
-		    upload_duration_ms = ?
+		    upload_bytes = ?,
+		    upload_duration_ms = ?,
+		    new_chunk_size = ?,
+		    uncompressed_chunk_size = (
+		        SELECT COALESCE(SUM(c.size), 0)
+		        FROM (
+		            SELECT DISTINCT fc.chunk_hash
+		            FROM snapshot_files sf
+		            JOIN file_chunks fc ON sf.file_id = fc.file_id
+		            WHERE sf.snapshot_id = snapshots.id
+		        ) sc
+		        JOIN chunks c ON sc.chunk_hash = c.chunk_hash
+		    )
 		WHERE id = ?
 	`
 
 	if tx != nil {
 		_, err = tx.ExecContext(ctx, query,
 			blobUncompressedSize, compressionRatio, compressionLevel,
-			uploadDurationMs, snapshotID)
+			uploadBytes, uploadDurationMs, newChunkSize, snapshotID)
 	} else {
 		_, err = r.db.ExecWithLog(ctx, query,
 			blobUncompressedSize, compressionRatio, compressionLevel,
-			uploadDurationMs, snapshotID)
+			uploadBytes, uploadDurationMs, newChunkSize, snapshotID)
 	}
 
 	if err != nil {
@@ -238,12 +256,7 @@ func (r *SnapshotRepository) GetOnlySnapshot(ctx context.Context) (*Snapshot, er
 		return nil, fmt.Errorf("querying snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	snapshots, err := r.scanSnapshotRows(rows)
 	if err != nil {
@@ -278,12 +291,7 @@ func (r *SnapshotRepository) ListRecent(
 		return nil, fmt.Errorf("querying snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanSnapshotRows(rows)
 }
@@ -395,7 +403,7 @@ func (r *SnapshotRepository) AddFilesByIDBatch(
 			args = append(args, snapshotID, fileID.String())
 		}
 
-		query += querySb312.String() //nolint:gosec // G202: appends "?" placeholders only
+		query += querySb312.String()
 
 		var err error
 		if tx != nil {
@@ -487,38 +495,63 @@ func (r *SnapshotRepository) GetBlobHashes(
 	ctx context.Context, snapshotID string,
 ) ([]string, error) {
 	query := `
-		SELECT sb.blob_hash 
+		SELECT sb.blob_hash
 		FROM snapshot_blobs sb
 		WHERE sb.snapshot_id = ?
 		ORDER BY sb.blob_hash
 	`
 
-	rows, err := r.db.conn.QueryContext(ctx, query, snapshotID)
+	blobs, err := r.queryStrings(ctx, query, snapshotID)
 	if err != nil {
 		return nil, fmt.Errorf("querying blob hashes: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	return blobs, nil
+}
 
-	var blobs []string
+// AddMountPoint records dir as a directory that the backup of a snapshot
+// found on a different filesystem from its parent directory, as a mount
+// point is, using tx when non-nil.
+func (r *SnapshotRepository) AddMountPoint(
+	ctx context.Context, tx *sql.Tx, snapshotID string, dir string,
+) error {
+	query := `
+		INSERT OR IGNORE INTO snapshot_mount_points (snapshot_id, path)
+		VALUES (?, ?)
+	`
 
-	for rows.Next() {
-		var blobHash string
-
-		err := rows.Scan(&blobHash)
-		if err != nil {
-			return nil, fmt.Errorf("scanning blob hash: %w", err)
-		}
-
-		blobs = append(blobs, blobHash)
+	var err error
+	if tx != nil {
+		_, err = tx.ExecContext(ctx, query, snapshotID, dir)
+	} else {
+		_, err = r.db.ExecWithLog(ctx, query, snapshotID, dir)
 	}
 
-	return blobs, rows.Err()
+	if err != nil {
+		return fmt.Errorf("adding mount point to snapshot: %w", err)
+	}
+
+	return nil
+}
+
+// GetMountPoints returns the directories AddMountPoint recorded for a
+// snapshot, sorted.
+func (r *SnapshotRepository) GetMountPoints(
+	ctx context.Context, snapshotID string,
+) ([]string, error) {
+	query := `
+		SELECT path
+		FROM snapshot_mount_points
+		WHERE snapshot_id = ?
+		ORDER BY path
+	`
+
+	dirs, err := r.queryStrings(ctx, query, snapshotID)
+	if err != nil {
+		return nil, fmt.Errorf("querying mount points: %w", err)
+	}
+
+	return dirs, nil
 }
 
 // GetSnapshotTotalCompressedSize returns the total compressed size of all
@@ -543,71 +576,66 @@ func (r *SnapshotRepository) GetSnapshotTotalCompressedSize(
 	return totalSize, nil
 }
 
-// GetSnapshotUncompressedChunkSize returns the sum of plaintext sizes of all unique
-// chunks referenced by a snapshot (via snapshot_files → file_chunks → chunks).
+// GetSnapshotBlobSizes returns the total compressed and uncompressed sizes
+// of all blobs referenced by a snapshot.
+func (r *SnapshotRepository) GetSnapshotBlobSizes(
+	ctx context.Context, snapshotID string,
+) (int64, int64, error) {
+	query := `
+		SELECT COALESCE(SUM(b.compressed_size), 0),
+			COALESCE(SUM(b.uncompressed_size), 0)
+		FROM snapshot_blobs sb
+		JOIN blobs b ON sb.blob_hash = b.blob_hash
+		WHERE sb.snapshot_id = ?
+	`
+
+	var compressed, uncompressed int64
+
+	err := r.db.conn.QueryRowContext(ctx, query, snapshotID).Scan(
+		&compressed, &uncompressed)
+	if err != nil {
+		return 0, 0, fmt.Errorf("querying snapshot blob sizes: %w", err)
+	}
+
+	return compressed, uncompressed, nil
+}
+
+// GetSnapshotUncompressedChunkSize returns the total plaintext size of the
+// distinct chunks the snapshot's files reference, as the snapshot's backup
+// recorded it (see UpdateExtendedStats).
 func (r *SnapshotRepository) GetSnapshotUncompressedChunkSize(
 	ctx context.Context, snapshotID string,
 ) (int64, error) {
-	query := `
-		SELECT COALESCE(SUM(c.size), 0)
-		FROM (
-			SELECT DISTINCT fc.chunk_hash
-			FROM snapshot_files sf
-			JOIN file_chunks fc ON sf.file_id = fc.file_id
-			WHERE sf.snapshot_id = ?
-		) sc
-		JOIN chunks c ON sc.chunk_hash = c.chunk_hash
-	`
+	query := `SELECT uncompressed_chunk_size FROM snapshots WHERE id = ?`
 
-	var totalSize int64
+	var uncompressedChunkSize int64
 
-	err := r.db.conn.QueryRowContext(ctx, query, snapshotID).Scan(&totalSize)
+	err := r.db.conn.QueryRowContext(ctx, query, snapshotID).Scan(&uncompressedChunkSize)
 	if err != nil {
 		return 0, fmt.Errorf("querying uncompressed chunk size: %w", err)
 	}
 
-	return totalSize, nil
+	return uncompressedChunkSize, nil
 }
 
-// GetSnapshotNewChunkSize returns the sum of plaintext sizes of chunks that are
-// referenced by this snapshot but not by any earlier completed snapshot known to
-// the local database. The result is the marginal uncompressed data this snapshot
-// added to the dedup pool — i.e., the delta from prior snapshots.
+// GetSnapshotNewChunkSize returns the total plaintext size of the chunks the
+// snapshot's backup stored that were not stored before, as the backup
+// recorded it. It is not derived from the snapshot's files: a file's row is
+// shared by every snapshot that lists it and is rewritten when the file
+// changes, so it shows only the file's latest chunks.
 func (r *SnapshotRepository) GetSnapshotNewChunkSize(
 	ctx context.Context, snapshotID string,
 ) (int64, error) {
-	query := `
-		WITH this_snap_chunks AS (
-			SELECT DISTINCT fc.chunk_hash
-			FROM snapshot_files sf
-			JOIN file_chunks fc ON sf.file_id = fc.file_id
-			WHERE sf.snapshot_id = ?
-		),
-		prior_chunks AS (
-			SELECT DISTINCT fc.chunk_hash
-			FROM snapshots s
-			JOIN snapshot_files sf ON sf.snapshot_id = s.id
-			JOIN file_chunks fc ON fc.file_id = sf.file_id
-			WHERE s.completed_at IS NOT NULL
-			  AND s.id != ?
-			  AND s.started_at < (SELECT started_at FROM snapshots WHERE id = ?)
-		)
-		SELECT COALESCE(SUM(c.size), 0)
-		FROM chunks c
-		JOIN this_snap_chunks t ON c.chunk_hash = t.chunk_hash
-		WHERE c.chunk_hash NOT IN (SELECT chunk_hash FROM prior_chunks)
-	`
+	query := `SELECT new_chunk_size FROM snapshots WHERE id = ?`
 
-	var totalSize int64
+	var newChunkSize int64
 
-	err := r.db.conn.QueryRowContext(
-		ctx, query, snapshotID, snapshotID, snapshotID,
-	).Scan(&totalSize)
+	err := r.db.conn.QueryRowContext(ctx, query, snapshotID).Scan(&newChunkSize)
 	if err != nil {
 		return 0, fmt.Errorf("querying new chunk size: %w", err)
 	}
 
-	return totalSize, nil
+	return newChunkSize, nil
 }
 
 // GetIncompleteSnapshots returns all snapshots that haven't been completed
@@ -628,12 +656,7 @@ func (r *SnapshotRepository) GetIncompleteSnapshots(
 		return nil, fmt.Errorf("querying incomplete snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanSnapshotRows(rows)
 }
@@ -656,12 +679,7 @@ func (r *SnapshotRepository) GetIncompleteByHostname(
 		return nil, fmt.Errorf("querying incomplete snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	// Same column set as every other multi-row snapshot query, so the
 	// shared scanner applies — including its timestamp normalization.
@@ -722,6 +740,34 @@ func (r *SnapshotRepository) DeleteSnapshotUploads(
 	return nil
 }
 
+// queryStrings runs query, which selects one text column, and returns
+// its values in the order of the rows.
+func (r *SnapshotRepository) queryStrings(
+	ctx context.Context, query string, args ...any,
+) ([]string, error) {
+	rows, err := r.db.conn.QueryContext(ctx, query, args...)
+	if err != nil {
+		return nil, err
+	}
+
+	defer func() { _ = rows.Close() }()
+
+	var values []string
+
+	for rows.Next() {
+		var value string
+
+		err := rows.Scan(&value)
+		if err != nil {
+			return nil, err
+		}
+
+		values = append(values, value)
+	}
+
+	return values, rows.Err()
+}
+
 // extendedCompressionRatio computes the compression ratio for a snapshot
 // from its stored blob_size and the given uncompressed size. Returns 1.0
 // when the uncompressed size is zero.

internal/database/snapshots_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"fmt"
 	"math"
+	"slices"
 	"testing"
 	"time"
 
@@ -145,6 +146,65 @@ func TestSnapshotRepositoryUpdateCounts(t *testing.T) {
 	}
 }
 
+// GetSnapshotBlobSizes totals the blobs the snapshot references, and only
+// those.
+func TestSnapshotRepositoryGetSnapshotBlobSizes(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repos := database.NewRepositories(db)
+
+	snapshot := &database.Snapshot{
+		ID:             "2024-01-03T12:00:00Z",
+		Hostname:       testHostname,
+		VaultikVersion: testVersion,
+		StartedAt:      time.Now().Truncate(time.Second),
+	}
+
+	err := repos.Snapshots.Create(ctx, nil, snapshot)
+	if err != nil {
+		t.Fatalf("failed to create snapshot: %v", err)
+	}
+
+	blobs := []*database.Blob{
+		{Hash: "referenced-1", CompressedSize: 10, UncompressedSize: 100},
+		{Hash: "referenced-2", CompressedSize: 20, UncompressedSize: 200},
+		{Hash: "unreferenced", CompressedSize: 40, UncompressedSize: 400},
+	}
+
+	for _, blob := range blobs {
+		blob.ID = types.NewBlobID()
+		blob.CreatedTS = time.Now().Truncate(time.Second)
+
+		err = repos.Blobs.Create(ctx, nil, blob)
+		if err != nil {
+			t.Fatalf("failed to create blob %s: %v", blob.Hash, err)
+		}
+	}
+
+	for _, blob := range blobs[:2] {
+		err = repos.Snapshots.AddBlob(ctx, nil, snapshot.ID.String(),
+			blob.ID, blob.Hash)
+		if err != nil {
+			t.Fatalf("failed to add blob %s to snapshot: %v", blob.Hash, err)
+		}
+	}
+
+	compressed, uncompressed, err := repos.Snapshots.GetSnapshotBlobSizes(
+		ctx, snapshot.ID.String())
+	if err != nil {
+		t.Fatalf("failed to get snapshot blob sizes: %v", err)
+	}
+
+	if compressed != 30 || uncompressed != 300 {
+		t.Errorf("blob sizes: got %d and %d, want 30 and 300",
+			compressed, uncompressed)
+	}
+}
+
 func TestSnapshotRepositoryListRecent(t *testing.T) {
 	t.Parallel()
 
@@ -363,3 +423,52 @@ func TestSnapshotRepositoryDuplicate(t *testing.T) {
 		t.Error("expected error for duplicate snapshot")
 	}
 }
+
+// Deleting a snapshot from the local index deletes the mount points its
+// backup recorded, and no other snapshot's. See
+// https://git.eeqj.de/sneak/vaultik/issues/395.
+func TestSnapshotRepositoryDeleteRemovesMountPoints(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repo := database.NewSnapshotRepository(db)
+
+	for _, id := range []types.SnapshotID{"kept", "deleted"} {
+		err := repo.Create(ctx, nil, &database.Snapshot{
+			ID:             id,
+			Hostname:       testHostname,
+			VaultikVersion: testVersion,
+			StartedAt:      time.Now().Truncate(time.Second),
+		})
+		if err != nil {
+			t.Fatalf("failed to create snapshot: %v", err)
+		}
+
+		err = repo.AddMountPoint(ctx, nil, id.String(), "/data/nas")
+		if err != nil {
+			t.Fatalf("failed to add mount point: %v", err)
+		}
+	}
+
+	err := repo.Delete(ctx, "deleted")
+	if err != nil {
+		t.Fatalf("failed to delete snapshot: %v", err)
+	}
+
+	for id, want := range map[string][]string{
+		"kept":    {"/data/nas"},
+		"deleted": nil,
+	} {
+		got, err := repo.GetMountPoints(ctx, id)
+		if err != nil {
+			t.Fatalf("failed to read mount points: %v", err)
+		}
+
+		if !slices.Equal(got, want) {
+			t.Errorf("mount points of %s: got %v, want %v", id, got, want)
+		}
+	}
+}

internal/database/uploads.go

@@ -158,19 +158,3 @@ type UploadStats struct {
 	MinDurationMs int64
 	MaxDurationMs int64
 }
-
-// GetCountBySnapshot returns the count of uploads for a specific snapshot
-func (r *UploadRepository) GetCountBySnapshot(
-	ctx context.Context, snapshotID string,
-) (int64, error) {
-	query := `SELECT COUNT(*) FROM uploads WHERE snapshot_id = ?`
-
-	var count int64
-
-	err := r.conn.QueryRowContext(ctx, query, snapshotID).Scan(&count)
-	if err != nil {
-		return 0, err
-	}
-
-	return count, nil
-}

internal/globals/globals.go

@@ -1,31 +1,89 @@
 // Package globals holds application-wide metadata (name, version,
-// commit) that is populated at build time via linker flags.
+// commit) that is populated at build time via linker flags, or from the
+// build info Go records in the binary when the linker flags set none.
 package globals
 
 import (
+	"regexp"
+	"runtime/debug"
 	"strings"
 	"time"
+
+	"golang.org/x/mod/module"
 )
 
 // Appname is the application name, populated from main().
 var Appname = "vaultik" //nolint:gochecknoglobals // set via -ldflags at build time
 
-// DevVersion is the version a binary reports when it was not built
-// from a tagged commit. script/version emits either this exact string
-// (outside a git checkout) or this string followed by "-" and the
-// commit it was built from, and goreleaser's snapshot template matches
-// that shape. It is deliberately not a number: a build that is not a
-// release must not name itself like one.
+// DevVersion is the version a binary reports when it was built without
+// git metadata: script/version emits it outside a git checkout, and a
+// binary keeps it when neither its linker flags nor its Go build info
+// name a version. goreleaser's snapshot template stamps
+// it followed by "-" and the commit it was built from. It is
+// deliberately not a number.
 const DevVersion = "dev"
 
+// Unknown is what Commit and CommitDate hold when the build did not
+// stamp them, and the version script/docker and script/cibuild stamp
+// when the host has no git checkout.
+const Unknown = "unknown"
+
 // Version is the application version, populated from main().
 var Version = DevVersion //nolint:gochecknoglobals // set via -ldflags at build time
 
 // Commit is the git commit hash, populated from main().
-var Commit = "unknown" //nolint:gochecknoglobals // set via -ldflags at build time
+var Commit = Unknown //nolint:gochecknoglobals // set via -ldflags at build time
 
 // CommitDate is the ISO-8601 date of the commit, populated from main().
-var CommitDate = "unknown" //nolint:gochecknoglobals // set via -ldflags at build time
+var CommitDate = Unknown //nolint:gochecknoglobals // set via -ldflags at build time
+
+// develVersion is the main module version Go records when it cannot
+// tell which version it built.
+const develVersion = "(devel)"
+
+// FillUnsetFromBuildInfo applies FillUnset to Version, Commit and
+// CommitDate, with the build info of the running binary.
+func FillUnsetFromBuildInfo() {
+	info, ok := debug.ReadBuildInfo()
+	if !ok {
+		return
+	}
+
+	Version, Commit, CommitDate = FillUnset(
+		info, Version, Commit, CommitDate)
+}
+
+// FillUnset returns version, commit and commitDate. Each one that
+// still holds its default ("dev" or empty for the version, "unknown"
+// for the others) is replaced by the value Go recorded in info. A
+// `go install` of a tag sets no linker flags, but Go records the tag
+// as the main module's version. Go records the commit and its time
+// only when it builds from a git checkout. The time is cut to a date,
+// as the Makefile stamps it.
+func FillUnset(
+	info *debug.BuildInfo, version, commit, commitDate string,
+) (string, string, string) {
+	if (version == "" || version == DevVersion) &&
+		info.Main.Version != "" && info.Main.Version != develVersion {
+		version = info.Main.Version
+	}
+
+	for _, s := range info.Settings {
+		switch s.Key {
+		case "vcs.revision":
+			if commit == Unknown {
+				commit = s.Value
+			}
+		case "vcs.time":
+			t, err := time.Parse(time.RFC3339, s.Value)
+			if commitDate == Unknown && err == nil {
+				commitDate = t.UTC().Format(time.DateOnly)
+			}
+		}
+	}
+
+	return version, commit, commitDate
+}
 
 // Author identifies the upstream author of vaultik.
 const Author = "Jeffrey Paul <sneak@sneak.berlin>"
@@ -60,18 +118,31 @@ func New() (*Globals, error) {
 }
 
 // IsDevVersion reports whether v names a development build rather than
-// a release. Both "dev" and "dev-<sha>" (and its "-dirty" variant)
-// count: a caller that compares against "dev" exactly would treat every
-// commit-stamped development build as a release.
+// a release. "dev" and goreleaser's snapshot "dev-<sha>" count, and so
+// does what `git describe --tags --always --dirty` gives a make or
+// docker build of an untagged commit: the bare short commit, or
+// tag-N-gHASH on a commit after a tag. A Go pseudo-version counts: Go
+// records one for a build of an untagged commit without linker flags.
+// Any version ending in "-dirty" or "+dirty" counts, a modified
+// checkout of a tag ("v1.0.0-dirty", "v1.0.0+dirty") included.
+// A plain tag such as "v1.0.0" or "1.0.0" is a release.
 //
 // The empty string counts too. Nothing that knows its version reports
 // no version, so an empty Version means the stamping failed, and the
 // safe reading of "we could not establish that this is a release" is
 // that it is not one. The Makefile refuses to build at all in that
 // case; this is the second line of defence, for a binary linked by
-// something other than the Makefile.
+// something other than the Makefile. Unknown counts for the same
+// reason.
 func IsDevVersion(v string) bool {
-	return v == "" || v == DevVersion || strings.HasPrefix(v, DevVersion+"-")
+	if v == "" || v == Unknown || v == DevVersion ||
+		strings.HasPrefix(v, DevVersion+"-") ||
+		strings.HasSuffix(v, "-dirty") || strings.HasSuffix(v, "+dirty") ||
+		module.IsPseudoVersion(v) {
+		return true
+	}
+
+	return regexp.MustCompile(`^(.+-[0-9]+-g)?[0-9a-f]+$`).MatchString(v)
 }
 
 // shortCommitLen is the number of commit-hash characters ShortCommit keeps.

internal/globals/globals_test.go

@@ -1,6 +1,7 @@
 package globals_test
 
 import (
+	"runtime/debug"
 	"testing"
 
 	"sneak.berlin/go/vaultik/internal/globals"
@@ -34,10 +35,11 @@ func TestGlobalsNew(t *testing.T) {
 }
 
 // TestIsDevVersion covers the boundary that matters: everything
-// script/version and goreleaser's snapshot template can emit for an
-// untagged build must be recognised as a development build, and a real
-// tag must not be. A plain equality check against "dev" used to decide
-// this, which classified every commit-stamped dev build as a release.
+// script/version, a plain docker build and goreleaser's snapshot
+// template can emit for an untagged build must be recognised as a
+// development build, and a real tag must not be. A plain equality check
+// against "dev" used to decide this, which classified every
+// commit-stamped dev build as a release.
 func TestIsDevVersion(t *testing.T) {
 	t.Parallel()
 
@@ -49,12 +51,29 @@ func TestIsDevVersion(t *testing.T) {
 		{"dev", true},
 		{"dev-b6e4a218a39e", true},
 		{"dev-b6e4a218a39e-dirty", true},
-		// What a tagged build produces (script/version strips the
-		// leading "v", matching goreleaser's .Version).
+		// What `git describe --tags --always --dirty` produces with no
+		// tag reachable, and on a commit after a tag.
+		{"877eb2f", true},
+		{"877eb2f-dirty", true},
+		{"v1.0.0-3-g877eb2f", true},
+		{"v1.0.0-3-g877eb2f-dirty", true},
+		{"1.0.0-rc.1-12-g877eb2f", true},
+		// A tagged commit with uncommitted changes is not that tag.
+		{"v1.0.0-dirty", true},
+		// What Go records for a build without linker flags: a
+		// pseudo-version for an untagged commit, "+dirty" for a
+		// modified checkout.
+		{"v1.0.0-rc3.0.20261111001226-abcdefabcdef", true},
+		{"v0.0.0-20261111001226-abcdefabcdef", true},
+		{"v1.0.1-0.20261111001226-abcdefabcdef", true},
+		{"v1.0.0+dirty", true},
+		// What a tagged build produces (goreleaser's .Version strips
+		// the leading "v"; script/version keeps it).
 		{"1.0.0", false},
 		{"0.1.0", false},
 		{"1.0.0-rc.1", false},
 		{"v1.0.0", false},
+		{"v1.0.0-rc3", false},
 		// A release must not be mistaken for a dev build just because
 		// the string happens to contain "dev".
 		{"1.0.0-dev", false},
@@ -64,6 +83,9 @@ func TestIsDevVersion(t *testing.T) {
 		// as one. The Makefile refuses to build when script/version
 		// yields nothing; this covers a binary linked some other way.
 		{"", true},
+		// What script/docker and script/cibuild stamp when the host
+		// has no git checkout.
+		{"unknown", true},
 	}
 
 	for _, tc := range cases {
@@ -73,10 +95,9 @@ func TestIsDevVersion(t *testing.T) {
 	}
 }
 
-// TestDefaultVersionIsDev pins the linker-flag contract: an unstamped
-// binary (no -ldflags at all, which is what `go build ./...` and `go
-// install` produce) must report itself as a development build rather
-// than as some default release number.
+// TestDefaultVersionIsDev pins the default: a binary whose linker flags
+// and build info name no version must report itself as a development
+// build rather than as some default release number.
 func TestDefaultVersionIsDev(t *testing.T) {
 	t.Parallel()
 
@@ -85,3 +106,104 @@ func TestDefaultVersionIsDev(t *testing.T) {
 			globals.DevVersion)
 	}
 }
+
+// buildInfo returns build info with the given main module version and,
+// when revision is not empty, the commit settings Go records for a
+// build from a git checkout.
+func buildInfo(version, revision string) *debug.BuildInfo {
+	info := &debug.BuildInfo{
+		Main: debug.Module{Path: "sneak.berlin/go/vaultik", Version: version},
+	}
+
+	if revision != "" {
+		info.Settings = []debug.BuildSetting{
+			{Key: "vcs.revision", Value: revision},
+			{Key: "vcs.time", Value: "2026-11-10T23:12:26Z"},
+		}
+	}
+
+	return info
+}
+
+// TestFillUnset covers a binary from `go install` of a tag, which
+// carries no linker flags: it must report the tag. Values the linker
+// flags set are kept.
+func TestFillUnset(t *testing.T) {
+	t.Parallel()
+
+	const (
+		tag = "v1.0.0-rc3"
+		rev = "0123456789abcdef0123456789abcdef01234567"
+	)
+
+	cases := []struct {
+		name                    string
+		info                    *debug.BuildInfo
+		version, commit, date   string
+		wantVersion, wantCommit string
+		wantDate                string
+	}{
+		{
+			name:        "go install of a tag",
+			info:        buildInfo(tag, ""),
+			version:     globals.DevVersion,
+			commit:      globals.Unknown,
+			date:        globals.Unknown,
+			wantVersion: tag,
+			wantCommit:  globals.Unknown,
+			wantDate:    globals.Unknown,
+		},
+		{
+			name:        "empty version",
+			info:        buildInfo(tag, ""),
+			version:     "",
+			commit:      globals.Unknown,
+			date:        globals.Unknown,
+			wantVersion: tag,
+			wantCommit:  globals.Unknown,
+			wantDate:    globals.Unknown,
+		},
+		{
+			name:        "go build in a git checkout",
+			info:        buildInfo(tag, rev),
+			version:     globals.DevVersion,
+			commit:      globals.Unknown,
+			date:        globals.Unknown,
+			wantVersion: tag,
+			wantCommit:  rev,
+			wantDate:    "2026-11-10",
+		},
+		{
+			name:        "Go could not tell the version",
+			info:        buildInfo("(devel)", ""),
+			version:     globals.DevVersion,
+			commit:      globals.Unknown,
+			date:        globals.Unknown,
+			wantVersion: globals.DevVersion,
+			wantCommit:  globals.Unknown,
+			wantDate:    globals.Unknown,
+		},
+		{
+			name:        "linker flags set",
+			info:        buildInfo(tag, rev),
+			version:     "v1.0.0-rc2-4-gabcdef1",
+			commit:      "abcdef1",
+			date:        "2026-11-01",
+			wantVersion: "v1.0.0-rc2-4-gabcdef1",
+			wantCommit:  "abcdef1",
+			wantDate:    "2026-11-01",
+		},
+	}
+
+	for _, tc := range cases {
+		version, commit, date := globals.FillUnset(
+			tc.info, tc.version, tc.commit, tc.date)
+
+		if version != tc.wantVersion || commit != tc.wantCommit ||
+			date != tc.wantDate {
+			t.Errorf("%s: got %q, %q, %q; want %q, %q, %q", tc.name,
+				version, commit, date,
+				tc.wantVersion, tc.wantCommit, tc.wantDate)
+		}
+	}
+}

internal/log/tty_handler.go

@@ -10,6 +10,8 @@ import (
 	"sync"
 	"time"
 	"unicode"
+
+	"github.com/dustin/go-humanize"
 )
 
 // groupSeparator joins an open group path to an attribute key. This
@@ -304,18 +306,13 @@ func formatDuration(d time.Duration) string {
 	}
 }
 
-// formatBytes formats bytes in a human-readable way
+// formatBytes formats a byte count with humanize.Bytes, as every command
+// prints one. humanize.Bytes takes no negative count, so one is printed
+// as it is.
 func formatBytes(b int64) string {
-	const unit = 1024
-	if b < unit {
+	if b < 0 {
 		return fmt.Sprintf("%d B", b)
 	}
 
-	div, exp := int64(unit), 0
-	for n := b / unit; n >= unit; n /= unit {
-		div *= unit
-		exp++
-	}
-
-	return fmt.Sprintf("%.1f %cB", float64(b)/float64(div), "KMGTPE"[exp])
+	return humanize.Bytes(uint64(b))
 }

internal/log/tty_handler_test.go

@@ -200,7 +200,8 @@ func TestTTYHandlerWithGroupQualifiesKeys(t *testing.T) {
 func TestTTYHandlerByteFormattingSurvivesGrouping(t *testing.T) {
 	t.Parallel()
 
-	const oneAndAHalfKiB = 1536
+	// Printed as "1.0 MB", in powers of 1000 like every other byte count.
+	const oneMegabyte = 1_000_000
 
 	for name, testCase := range map[string]struct {
 		derive func(*slog.Logger) *slog.Logger
@@ -223,13 +224,13 @@ func TestTTYHandlerByteFormattingSurvivesGrouping(t *testing.T) {
 			var buf bytes.Buffer
 
 			logger := slog.New(log.NewTTYHandler(&buf, debugHandlerOptions()))
-			testCase.derive(logger).Info("uploaded", "bytes", oneAndAHalfKiB)
+			testCase.derive(logger).Info("uploaded", "bytes", oneMegabyte)
 
 			line := ansiEscape.ReplaceAllString(buf.String(), "")
 
-			assert.Contains(t, line, testCase.key+"=1.5 KB",
+			assert.Contains(t, line, testCase.key+"=1.0 MB",
 				"a byte count must be human-readable however it is qualified")
-			assert.NotContains(t, line, strconv.Itoa(oneAndAHalfKiB),
+			assert.NotContains(t, line, strconv.Itoa(oneMegabyte),
 				"the raw number must not survive the formatting")
 		})
 	}

internal/pidlock/pidlock.go

@@ -10,15 +10,19 @@ import (
 	"path/filepath"
 	"strconv"
 	"strings"
-	"syscall"
+
+	"github.com/adrg/xdg"
+	"golang.org/x/sys/unix"
 )
 
 // ErrAlreadyRunning indicates another vaultik instance is running.
 var ErrAlreadyRunning = errors.New("another vaultik instance is already running")
 
-// Lock represents an acquired PID lock.
+// Lock represents an acquired PID lock: an flock(2) on the PID file,
+// held while the file stays open. The kernel drops it when the process
+// exits, however it exits, so a crashed run never leaves the lock held.
 type Lock struct {
-	path string
+	file *os.File
 }
 
 const (
@@ -26,13 +30,24 @@ const (
 	lockDirPerm = 0o700
 	// pidFilePerm is the mode for the PID file (owner-only).
 	pidFilePerm = 0o600
+	// pidFileName is the name of the PID file in the lock directory.
+	pidFileName = "vaultik.pid"
 )
 
+// Dir returns the directory that holds vaultik's PID file.
+func Dir() string {
+	return filepath.Join(xdg.DataHome, "vaultik")
+}
+
+// Path returns the path of the PID file that Acquire uses in lockDir.
+func Path(lockDir string) string {
+	return filepath.Join(lockDir, pidFileName)
+}
+
 // Acquire attempts to acquire a PID lock in the specified directory.
-// If the lock file exists and the process is still running, it returns
-// ErrAlreadyRunning with details about the existing process.
-// On success, it writes the current PID to the lock file and returns
-// a Lock that must be released with Release().
+// If another process holds the lock, it returns ErrAlreadyRunning with
+// that process's PID. On success, it writes the current PID to the lock
+// file and returns a Lock that must be released with Release().
 func Acquire(lockDir string) (*Lock, error) {
 	// Ensure lock directory exists
 	err := os.MkdirAll(lockDir, lockDirPerm)
@@ -40,58 +55,84 @@ func Acquire(lockDir string) (*Lock, error) {
 		return nil, fmt.Errorf("creating lock directory: %w", err)
 	}
 
-	lockPath := filepath.Join(lockDir, "vaultik.pid")
+	lockPath := Path(lockDir)
 
-	// Check for existing lock
-	existingPID, err := readPIDFile(lockPath)
-	if err == nil {
-		// Lock file exists, check if process is running
-		if isProcessRunning(existingPID) {
-			return nil, fmt.Errorf("%w (PID %d)", ErrAlreadyRunning, existingPID)
-		}
-		// Process is not running, stale lock file - we can take over
-	}
-
-	// Write our PID
-	pid := os.Getpid()
-
-	err = os.WriteFile(lockPath, []byte(strconv.Itoa(pid)), pidFilePerm)
+	// No O_TRUNC: the file may hold the PID of the process that has the
+	// lock, which the error below reports.
+	file, err := os.OpenFile( //nolint:gosec // G304: path is our own lock file
+		lockPath, os.O_RDWR|os.O_CREATE, pidFilePerm)
 	if err != nil {
-		return nil, fmt.Errorf("writing PID file: %w", err)
+		return nil, fmt.Errorf("opening PID file: %w", err)
 	}
 
-	return &Lock{path: lockPath}, nil
+	err = unix.Flock(int(file.Fd()), unix.LOCK_EX|unix.LOCK_NB)
+	if err != nil {
+		_ = file.Close()
+
+		if errors.Is(err, unix.EWOULDBLOCK) {
+			return nil, alreadyRunningError(lockPath)
+		}
+
+		return nil, fmt.Errorf("locking PID file: %w", err)
+	}
+
+	err = writePID(file)
+	if err != nil {
+		_ = file.Close()
+
+		return nil, err
+	}
+
+	return &Lock{file: file}, nil
 }
 
-// Release removes the PID lock file.
+// Release empties the PID file and closes it, which drops the lock.
 // It is safe to call Release multiple times.
 func (l *Lock) Release() error {
-	if l == nil || l.path == "" {
+	if l == nil || l.file == nil {
 		return nil
 	}
 
-	// Verify we still own the lock (our PID is in the file)
-	existingPID, err := readPIDFile(l.path)
+	file := l.file
+	l.file = nil
+
+	// Do not remove the file here. A process that opened it a moment
+	// earlier could then lock the removed file while another creates and
+	// locks a new one, and both would run.
+	truncateErr := file.Truncate(0)
+	closeErr := file.Close()
+
+	return errors.Join(truncateErr, closeErr)
+}
+
+// writePID replaces the contents of the locked PID file with the current
+// PID.
+func writePID(file *os.File) error {
+	err := file.Truncate(0)
 	if err != nil {
-		// File already gone or unreadable - that's fine
-		return nil //nolint:nilerr // unreadable lock file means nothing to release
+		return fmt.Errorf("truncating PID file: %w", err)
 	}
 
-	if existingPID != os.Getpid() {
-		// Someone else wrote to our lock file - don't remove it
-		return nil
+	_, err = file.WriteAt([]byte(strconv.Itoa(os.Getpid())), 0)
+	if err != nil {
+		return fmt.Errorf("writing PID file: %w", err)
 	}
 
-	err = os.Remove(l.path)
-	if err != nil && !os.IsNotExist(err) {
-		return fmt.Errorf("removing PID file: %w", err)
-	}
-
-	l.path = "" // Prevent double-release
-
 	return nil
 }
 
+// alreadyRunningError reports that another process holds the lock,
+// naming its PID when the file holds one. The holder writes its PID just
+// after it locks, so the file can briefly be empty.
+func alreadyRunningError(lockPath string) error {
+	pid, err := readPIDFile(lockPath)
+	if err != nil {
+		return ErrAlreadyRunning
+	}
+
+	return fmt.Errorf("%w (PID %d)", ErrAlreadyRunning, pid)
+}
+
 // readPIDFile reads and parses the PID from a lock file.
 func readPIDFile(path string) (int, error) {
 	data, err := os.ReadFile(path) //nolint:gosec // G304: path is our own lock file
@@ -106,16 +147,3 @@ func readPIDFile(path string) (int, error) {
 
 	return pid, nil
 }
-
-// isProcessRunning checks if a process with the given PID is running.
-func isProcessRunning(pid int) bool {
-	process, err := os.FindProcess(pid)
-	if err != nil {
-		return false
-	}
-
-	// On Unix, FindProcess always succeeds. We need to send signal 0 to check.
-	err = process.Signal(syscall.Signal(0))
-
-	return err == nil
-}

internal/pidlock/pidlock_test.go

@@ -4,6 +4,7 @@ import (
 	"os"
 	"path/filepath"
 	"strconv"
+	"sync"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -33,9 +34,10 @@ func TestAcquireAndRelease(t *testing.T) {
 	err = lock.Release()
 	require.NoError(t, err)
 
-	// Verify PID file is gone
-	_, err = os.Stat(pidPath)
-	assert.True(t, os.IsNotExist(err))
+	// Verify PID file is empty
+	data, err = os.ReadFile(pidPath) //nolint:gosec // G304: test's own temp file
+	require.NoError(t, err)
+	assert.Empty(t, data)
 }
 
 func TestAcquireBlocksSecondInstance(t *testing.T) {
@@ -55,6 +57,64 @@ func TestAcquireBlocksSecondInstance(t *testing.T) {
 	lock2, err := pidlock.Acquire(tmpDir)
 	require.ErrorIs(t, err, pidlock.ErrAlreadyRunning)
 	assert.Nil(t, lock2)
+
+	// Once the first lock is released, the next Acquire succeeds
+	require.NoError(t, lock1.Release())
+
+	lock3, err := pidlock.Acquire(tmpDir)
+	require.NoError(t, err)
+	require.NoError(t, lock3.Release())
+}
+
+// TestConcurrentAcquireAdmitsOne starts many Acquire calls at the same
+// moment, as two cron entries firing together would, and checks that
+// exactly one of them gets the lock.
+func TestConcurrentAcquireAdmitsOne(t *testing.T) {
+	t.Parallel()
+
+	const callers = 50
+
+	tmpDir := t.TempDir()
+	start := make(chan struct{})
+
+	var (
+		mu       sync.Mutex
+		acquired []*pidlock.Lock
+		failures []error
+		wg       sync.WaitGroup
+	)
+
+	for range callers {
+		wg.Go(func() {
+			<-start
+
+			lock, err := pidlock.Acquire(tmpDir)
+
+			mu.Lock()
+			defer mu.Unlock()
+
+			if err != nil {
+				failures = append(failures, err)
+
+				return
+			}
+
+			acquired = append(acquired, lock)
+		})
+	}
+
+	close(start)
+	wg.Wait()
+
+	for _, lock := range acquired {
+		require.NoError(t, lock.Release())
+	}
+
+	assert.Len(t, acquired, 1, "exactly one caller should hold the lock")
+
+	for _, err := range failures {
+		require.ErrorIs(t, err, pidlock.ErrAlreadyRunning)
+	}
 }
 
 func TestAcquireWithStaleLock(t *testing.T) {

internal/s3/client.go

@@ -6,7 +6,9 @@ import (
 	"context"
 	"errors"
 	"io"
+	"strings"
 	"sync/atomic"
+	"time"
 
 	"github.com/aws/aws-sdk-go-v2/aws"
 	"github.com/aws/aws-sdk-go-v2/config"
@@ -17,6 +19,13 @@ import (
 	"github.com/aws/smithy-go/logging"
 )
 
+// ErrNoContinuationToken is returned by ListObjects and ListObjectsStream
+// when the server marks a page of a listing truncated but sends no
+// continuation token to fetch the rest with, as a server answering V2
+// listings V1-style does.
+var ErrNoContinuationToken = errors.New(
+	"server marked the listing truncated but sent no continuation token")
+
 // Client wraps the AWS S3 client for vaultik operations.
 // It provides a simplified interface for S3 operations with automatic
 // prefix handling and connection management. All operations are performed
@@ -26,10 +35,14 @@ type Client struct {
 	bucket   string
 	prefix   string
 	endpoint string
+	partSize int64
 }
 
 // Config contains S3 client configuration.
-// All fields are required except Prefix, which defaults to an empty string.
+// All fields are required except Prefix, which defaults to an empty string,
+// and PartSize, where zero means the SDK default of 5 MiB.
+// A non-empty Prefix is joined to every key with one "/", whether or not
+// it ends with one.
 // The Endpoint field should include the protocol (http:// or https://).
 type Config struct {
 	Endpoint        string
@@ -38,6 +51,9 @@ type Config struct {
 	AccessKeyID     string
 	SecretAccessKey string
 	Region          string
+	// PartSize is the size in bytes of each part of a multipart upload.
+	// An upload too large for S3's limit of 10,000 parts gets larger parts.
+	PartSize int64
 }
 
 // nopLogger is a logger that discards all output.
@@ -75,11 +91,19 @@ func NewClient(ctx context.Context, cfg Config) (*Client, error) {
 
 	s3Client := s3.NewFromConfig(awsCfg, s3Opts)
 
+	// Every method below builds a key as prefix + key, so the prefix
+	// must carry its own trailing "/".
+	prefix := strings.TrimRight(cfg.Prefix, "/")
+	if prefix != "" {
+		prefix += "/"
+	}
+
 	return &Client{
 		s3Client: s3Client,
 		bucket:   cfg.Bucket,
-		prefix:   cfg.Prefix,
+		prefix:   prefix,
 		endpoint: cfg.Endpoint,
+		partSize: cfg.PartSize,
 	}, nil
 }
 
@@ -113,12 +137,9 @@ func (c *Client) PutObjectWithProgress(
 ) error {
 	fullKey := c.prefix + key
 
-	// uploadPartSize is 10MB for better progress granularity.
-	const uploadPartSize = 10 * 1024 * 1024
-
 	// Create an uploader with the S3 client
 	uploader := manager.NewUploader(c.s3Client, func(u *manager.Uploader) {
-		u.PartSize = uploadPartSize
+		u.PartSize = uploadPartSize(c.partSize, size)
 	})
 
 	// Create a progress reader that tracks upload progress
@@ -139,6 +160,21 @@ func (c *Client) PutObjectWithProgress(
 	return err
 }
 
+// uploadPartSize returns the part size for an upload of size bytes: the
+// configured part size (the SDK default when zero), raised where needed so
+// the upload fits in S3's limit of 10,000 parts. The uploader cannot raise
+// it itself, because it cannot seek the progress reader to learn its size.
+func uploadPartSize(configured, size int64) int64 {
+	if configured == 0 {
+		configured = manager.DefaultUploadPartSize
+	}
+
+	maxParts := int64(manager.MaxUploadParts)
+	smallestThatFits := (size + maxParts - 1) / maxParts // rounded up
+
+	return max(configured, smallestThatFits)
+}
+
 // GetObject downloads an object from S3 with the specified key.
 // The key is automatically prefixed with the configured prefix.
 // Returns a ReadCloser containing the object data. The caller must
@@ -186,7 +222,7 @@ func (c *Client) ListObjects(ctx context.Context, prefix string) ([]string, erro
 	})
 
 	for paginator.HasMorePages() {
-		page, err := paginator.NextPage(ctx)
+		page, err := nextListPage(ctx, paginator)
 		if err != nil {
 			return nil, err
 		}
@@ -207,6 +243,26 @@ func (c *Client) ListObjects(ctx context.Context, prefix string) ([]string, erro
 	return keys, nil
 }
 
+// nextListPage fetches the next page of a listing. The SDK paginator takes
+// a page marked truncated without a continuation token for the last page
+// and stops without an error, so such a page is ErrNoContinuationToken
+// here instead.
+func nextListPage(
+	ctx context.Context, paginator *s3.ListObjectsV2Paginator,
+) (*s3.ListObjectsV2Output, error) {
+	page, err := paginator.NextPage(ctx)
+	if err != nil {
+		return nil, err
+	}
+
+	truncated := aws.ToBool(page.IsTruncated)
+	if truncated && aws.ToString(page.NextContinuationToken) == "" {
+		return nil, ErrNoContinuationToken
+	}
+
+	return page, nil
+}
+
 // HeadObject checks if an object exists in S3.
 // Returns true if the object exists, false otherwise.
 // The key is automatically prefixed with the configured prefix.
@@ -247,7 +303,10 @@ func IsNotFound(err error) bool {
 type ObjectInfo struct {
 	Key  string
 	Size int64
-	Err  error
+	// ModTime is the object's LastModified time. Only ListObjectsStream
+	// sets it.
+	ModTime time.Time
+	Err     error
 }
 
 // ListObjectsStream lists objects with the given prefix and returns a channel.
@@ -272,7 +331,7 @@ func (c *Client) ListObjectsStream(
 		})
 
 		for paginator.HasMorePages() {
-			page, err := paginator.NextPage(ctx)
+			page, err := nextListPage(ctx, paginator)
 			if err != nil {
 				ch <- ObjectInfo{Err: err}
 
@@ -288,8 +347,9 @@ func (c *Client) ListObjectsStream(
 					}
 
 					ch <- ObjectInfo{
-						Key:  key,
-						Size: *obj.Size,
+						Key:     key,
+						Size:    *obj.Size,
+						ModTime: aws.ToTime(obj.LastModified),
 					}
 				}
 			}

internal/s3/client_internal_test.go

@@ -0,0 +1,55 @@
+package s3
+
+import "testing"
+
+// TestUploadPartSize checks that an upload too large for 10,000 parts of the
+// configured size gets parts just large enough to fit in 10,000.
+func TestUploadPartSize(t *testing.T) {
+	t.Parallel()
+
+	const mib = 1024 * 1024
+
+	tests := []struct {
+		name       string
+		configured int64
+		size       int64
+		want       int64
+	}{
+		{
+			name:       "an upload that fits keeps the configured size",
+			configured: 5 * mib,
+			size:       10 * 1024 * mib,
+			want:       5 * mib,
+		},
+		{
+			name:       "exactly 10,000 parts keeps the configured size",
+			configured: 6 * mib,
+			size:       10_000 * 6 * mib,
+			want:       6 * mib,
+		},
+		{
+			name:       "one byte more than 10,000 parts adds a byte to each",
+			configured: 6 * mib,
+			size:       10_000*6*mib + 1,
+			want:       6*mib + 1,
+		},
+		{
+			name:       "zero means the SDK default of 5MiB",
+			configured: 0,
+			size:       1,
+			want:       5 * mib,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			got := uploadPartSize(tt.configured, tt.size)
+			if got != tt.want {
+				t.Errorf("uploadPartSize(%d, %d) = %d, want %d",
+					tt.configured, tt.size, got, tt.want)
+			}
+		})
+	}
+}

internal/s3/module.go

@@ -28,6 +28,7 @@ func provideClient(lc fx.Lifecycle, cfg *config.Config) (*Client, error) {
 		AccessKeyID:     cfg.S3.AccessKeyID,
 		SecretAccessKey: cfg.S3.SecretAccessKey,
 		Region:          cfg.S3.Region,
+		PartSize:        cfg.S3.PartSize.Int64(),
 	})
 	if err != nil {
 		return nil, err

internal/snapshot/backup_test.go

@@ -71,7 +71,7 @@ func verifyBackupFiles(
 ) {
 	t.Helper()
 
-	files, err := repos.Files.ListByPrefix(ctx, "")
+	files, err := repos.Files.ListAll(ctx)
 	if err != nil {
 		t.Fatalf("Failed to list files: %v", err)
 	}

internal/snapshot/blob_temp_file_test.go

@@ -0,0 +1,62 @@
+//nolint:testpackage // exercises the unexported cleanupBlobTempFile
+package snapshot
+
+import (
+	"errors"
+	"os"
+	"testing"
+
+	"github.com/spf13/afero"
+	"sneak.berlin/go/vaultik/internal/blob"
+)
+
+// A blob temp file that cannot be closed or removed must make the cleanup
+// return an error, not end the process.
+func TestCleanupBlobTempFileReturnsErrors(t *testing.T) {
+	t.Parallel()
+
+	fs := afero.NewOsFs()
+	s := &Scanner{fs: fs}
+
+	tests := []struct {
+		name string
+		// prepare runs before the cleanup and makes its close or remove
+		// fail.
+		prepare func(tempFile afero.File) error
+		want    error
+	}{
+		{
+			name:    "close fails",
+			prepare: func(tempFile afero.File) error { return tempFile.Close() },
+			want:    os.ErrClosed,
+		},
+		{
+			name: "remove fails",
+			prepare: func(tempFile afero.File) error {
+				return fs.Remove(tempFile.Name())
+			},
+			want: os.ErrNotExist,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			tempFile, err := afero.TempFile(fs, t.TempDir(), "blob-*.tmp")
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			err = tt.prepare(tempFile)
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			err = s.cleanupBlobTempFile(&blob.WithReader{TempFile: tempFile})
+			if !errors.Is(err, tt.want) {
+				t.Fatalf("expected an error wrapping %v, got: %v", tt.want, err)
+			}
+		})
+	}
+}

internal/snapshot/device_number_test.go

@@ -0,0 +1,40 @@
+//nolint:testpackage // exercises the unexported deviceNumberFromStat
+package snapshot
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// The device number a real backup reads is the same for a directory and
+// a file in it, and differs between the temp directory and /dev.
+// Without it no mount point is ever found. See
+// https://git.eeqj.de/sneak/vaultik/issues/395.
+func TestDeviceNumberFromStat(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	file := filepath.Join(dir, "file.txt")
+	require.NoError(t, os.WriteFile(file, nil, 0o600))
+
+	deviceOf := func(path string) uint64 {
+		t.Helper()
+
+		info, err := os.Stat(path)
+		require.NoError(t, err)
+
+		device, ok := deviceNumberFromStat(info)
+		require.True(t, ok, "no device number read for %s", path)
+
+		return device
+	}
+
+	assert.Equal(t, deviceOf(dir), deviceOf(file))
+
+	// /dev is a different filesystem from the temp dir on Linux, macOS and in Docker.
+	assert.NotEqual(t, deviceOf(dir), deviceOf("/dev"))
+}

internal/snapshot/exclude_test.go

@@ -531,3 +531,101 @@ func TestExcludePatterns_UnanchoredPatternFile(t *testing.T) {
 	require.Equal(t, 1, result.FilesScanned,
 		"Unanchored pattern for file should match anywhere")
 }
+
+func TestExcludePatterns_DoubleStarDirectoryAtEveryDepth(t *testing.T) {
+	t.Parallel()
+
+	// As in gitignore, a leading **/ also matches directly under the source dir
+	fs := setupAnchoredTestFS(t)
+
+	scanner, repos, cleanup := createTestScanner(t, fs, []string{"**/projectname"})
+	defer cleanup()
+
+	ctx := context.Background()
+	createSnapshotRecord(ctx, t, repos, "test-snapshot")
+
+	result, err := scanner.Scan(ctx, "/backup", "test-snapshot")
+	require.NoError(t, err)
+
+	// **/projectname should exclude BOTH:
+	// - /backup/projectname/file.txt
+	// - /backup/otherproject/projectname/file.txt
+	// Total files: 4, excluded: 2
+	require.Equal(t, 2, result.FilesScanned,
+		"**/ directory pattern should match at the top level and nested")
+}
+
+func TestExcludePatterns_DoubleStarFileAtEveryDepth(t *testing.T) {
+	t.Parallel()
+
+	// As in gitignore, a leading **/ also matches directly under the source dir
+	fs := setupAnchoredTestFS(t)
+
+	scanner, repos, cleanup := createTestScanner(t, fs, []string{"**/*.txt"})
+	defer cleanup()
+
+	ctx := context.Background()
+	createSnapshotRecord(ctx, t, repos, "test-snapshot")
+
+	result, err := scanner.Scan(ctx, "/backup", "test-snapshot")
+	require.NoError(t, err)
+
+	// **/*.txt should exclude ALL .txt files:
+	// - /backup/file.txt
+	// - /backup/projectname/file.txt
+	// - /backup/otherproject/projectname/file.txt
+	// Total files: 4, excluded: 3
+	require.Equal(t, 1, result.FilesScanned,
+		"**/ file pattern should match at the top level and nested")
+}
+
+func TestExcludePatterns_AnchoredDoubleStarAtEveryDepth(t *testing.T) {
+	t.Parallel()
+
+	// As in gitignore, /**/ also matches directly under the source dir
+	fs := setupAnchoredTestFS(t)
+
+	scanner, repos, cleanup := createTestScanner(t, fs, []string{"/**/projectname"})
+	defer cleanup()
+
+	ctx := context.Background()
+	createSnapshotRecord(ctx, t, repos, "test-snapshot")
+
+	result, err := scanner.Scan(ctx, "/backup", "test-snapshot")
+	require.NoError(t, err)
+
+	// /**/projectname should exclude BOTH:
+	// - /backup/projectname/file.txt
+	// - /backup/otherproject/projectname/file.txt
+	// Total files: 4, excluded: 2
+	require.Equal(t, 2, result.FilesScanned,
+		"/**/ pattern should match at the top level and nested")
+}
+
+func TestExcludePatterns_DoubleStarInMiddleMatchesZeroDirectories(t *testing.T) {
+	t.Parallel()
+
+	// As in gitignore, a/**/b matches a/b as well as a/c/b. The last part
+	// is a glob: with a plain name the glob library already matched a/b.
+	fs := &afero.MemMapFs{}
+	files := []string{"/backup/a/x.log", "/backup/a/c/y.log", "/backup/a/c/kept.txt"}
+
+	for _, path := range files {
+		require.NoError(t, fs.MkdirAll(filepath.Dir(path), 0755))
+		require.NoError(t, afero.WriteFile(fs, path, []byte(path), 0644))
+	}
+
+	scanner, repos, cleanup := createTestScanner(t, fs, []string{"a/**/*.log"})
+	defer cleanup()
+
+	ctx := context.Background()
+	createSnapshotRecord(ctx, t, repos, "test-snapshot")
+
+	result, err := scanner.Scan(ctx, "/backup", "test-snapshot")
+	require.NoError(t, err)
+
+	// a/**/*.log should exclude /backup/a/x.log and /backup/a/c/y.log
+	// Total files: 3, excluded: 2
+	require.Equal(t, 1, result.FilesScanned,
+		"a/**/*.log should match a/x.log as well as a/c/y.log")
+}

internal/snapshot/export_perm_test.go

@@ -1,40 +1,48 @@
-//nolint:testpackage // exercises the unexported copyFile helper
+//nolint:testpackage // exercises the unexported copyDatabase helper
 package snapshot
 
 import (
+	"context"
 	"os"
 	"path/filepath"
 	"syscall"
 	"testing"
 
 	"github.com/spf13/afero"
+	"sneak.berlin/go/vaultik/internal/database"
 )
 
-// TestCopyFileExportCopyMode verifies that the exported snapshot database
-// copy is created owner-only (0600), even under a lenient 022 umask that
-// would otherwise leave a fresh file world-readable.
+// TestCopyDatabaseExportCopyMode verifies that the exported snapshot
+// database copy is created owner-only (0600), even under a lenient 022
+// umask that would otherwise leave a fresh file world-readable.
 //
 //nolint:paralleltest // syscall.Umask is process-global; parallel tests would clash
-func TestCopyFileExportCopyMode(t *testing.T) {
+func TestCopyDatabaseExportCopyMode(t *testing.T) {
 	restore := syscall.Umask(0o022)
 	defer syscall.Umask(restore)
 
+	ctx := context.Background()
 	dir := t.TempDir()
 
 	src := filepath.Join(dir, "index.sqlite")
 
-	err := os.WriteFile(src, []byte("index data"), 0o600)
+	db, err := database.New(ctx, src)
 	if err != nil {
 		t.Fatalf("creating source index: %v", err)
 	}
 
+	err = db.Close()
+	if err != nil {
+		t.Fatalf("closing source index: %v", err)
+	}
+
 	dst := filepath.Join(dir, "snapshot.db")
 
 	sm := &SnapshotManager{fs: afero.NewOsFs()}
 
-	err = sm.copyFile(src, dst)
+	err = sm.copyDatabase(ctx, src, dst)
 	if err != nil {
-		t.Fatalf("copyFile: %v", err)
+		t.Fatalf("copyDatabase: %v", err)
 	}
 
 	info, err := os.Stat(dst)

internal/snapshot/index_and_lock_files_test.go

@@ -0,0 +1,65 @@
+package snapshot_test
+
+import (
+	"context"
+	"testing"
+
+	"github.com/spf13/afero"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/snapshot"
+)
+
+// A backup of the directory that holds vaultik's index and PID file
+// stores neither, with no exclude pattern set, and still stores the
+// files beside them. See https://git.eeqj.de/sneak/vaultik/issues/421.
+func TestScanSkipsIndexAndLockFiles(t *testing.T) {
+	t.Parallel()
+
+	const dataDir = "/home/user/.local/share/vaultik"
+
+	indexAndLockFiles := append(database.IndexFiles(dataDir+"/index.sqlite"),
+		dataDir+"/vaultik.pid")
+
+	fs := afero.NewMemMapFs()
+	for _, p := range append(indexAndLockFiles, dataDir+"/notes.txt") {
+		require.NoError(t, afero.WriteFile(fs, p, []byte("data"), 0o600))
+	}
+
+	db, err := database.NewTestDB()
+	require.NoError(t, err)
+
+	defer func() { _ = db.Close() }()
+
+	repos := database.NewRepositories(db)
+
+	scanner := snapshot.NewScanner(snapshot.ScannerConfig{
+		FS:               fs,
+		ChunkSize:        64 * 1024,
+		Repositories:     repos,
+		MaxBlobSize:      1024 * 1024,
+		CompressionLevel: 3,
+		AgeRecipients: []string{
+			"age1ql3z7hjy54pw3hyww5ayyfg7zqgvc7w3j2elw8zmrj2kg5sfn9aqmcac8p"},
+		IndexAndLockFiles: indexAndLockFiles,
+	})
+	require.NotNil(t, scanner)
+
+	ctx := context.Background()
+	createSnapshotRecord(ctx, t, repos, "test-snapshot")
+
+	result, err := scanner.Scan(ctx, "/home/user", "test-snapshot")
+	require.NoError(t, err)
+
+	require.Equal(t, 1, result.FilesScanned)
+
+	stored, err := repos.Files.GetByPath(ctx, dataDir+"/notes.txt")
+	require.NoError(t, err)
+	require.NotNil(t, stored, "a file beside the index is stored")
+
+	for _, p := range indexAndLockFiles {
+		f, err := repos.Files.GetByPath(ctx, p)
+		require.NoError(t, err)
+		require.Nil(t, f, "%s is not stored", p)
+	}
+}