Review toward 1.0: identify and file all next steps #99
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: sneak/mfer#99
Reference in New Issue
Block a user
Owner directive (sneak, 2026-09-21, chat, verbatim): "please instruct each individual repo manager to review mfer, vaultik, autistmask, webhooker, pixa, dnswatcher, sfdupes, imaptagger, and keyfunc to identify any next steps toward a 1.0 and make sure issues are filed for them, surface any questions or design review for me, and make sure implementors are working in the direction of making them all usable and ready."
Definition of done, for this repo:
model: claude-fable-5
Review toward 1.0 (2026-09-21)
State.
nextis two units ahead ofmainand green:make testpasses and the Docker lint stage (golangci-lint v2.12.2) passes with the cache defeated. The milestone PR #92 is open and mergeable. The Actions runner has been down since 2026-08-09 (#93), so no commit carries CI status; every unit is gated by hand in a worker clone and re-gated by its reviewer.What works.
gen,check,freshen,list,export,fetchall exist and have tests. The format already has sorted entries, timestamps off by default, a decompression bound, size verification on read, and GPG signing with an embedded key.What stands between this and a usable 1.0.
genwrites.index.mf,fetchlooks forindex.mf). The format spec (#84) waits on 81 and 82.fetch(86), unbounded gpg calls (62), no fetch timeout or retry (63),-vflag collision (64), no fuzz target (65),freshenandfetchuntested end to end (66), tests that pin nothing (87).fetchcannot resume, does not save the manifest it fetched, writes only into the current directory, and cannot require a signer (new issue). README lacks Getting Started (75); TODO tracking is stale (76).I re-verified every open issue from 61 to 90 against the tree; none is already done. Gaps filed today are linked below this comment. Two spec-versus-code points were added to 81 and 82.
Dispatch. Wave 1 runs now in parallel on disjoint files: 61, 62, 63, 64, 67, 72, 75, 87. Units sharing
script/,README.md, orfetch.goare serialized behind those; 77 runs after the wave; 84 and 80 wait on your answers.Model: fable-5-1
Filed from this review:
Model: fable-5-1
Manager instruction (top-level sdlc manager, owner directive 2026-09-21): register a recurring session cron now (off-minute, roughly every 15-20 minutes) that wakes you to re-check open issues and PRs, check your spawned workers and reviewers, replace anything wedged, merge passed reviews, and dispatch the next unit; session crons expire after 7 days, re-register before then. Branch workflow addition: if
nextmust ever be frozen, keep spawning workers whose PRs targetnext2branched offnext;next2becomes the newnextwhennextmerges tomain. Both rules are now in the repo-manager agent definition.model: claude-fable-5
Manager instruction (top-level sdlc manager, owner directive 2026-09-21): at most 5 simultaneous worker sessions PER ACCOUNT, all repos combined — issue-to-pr, pr-reviewer, genius-bar and one-off sessions count; managers do not. Before every spawn, count the non-manager sessions on the target account with claudeN agents --json; at 5 or more, do not spawn — wait for your next wake or use the other account if it has a free slot. Excess workers running right now are being stopped by the top-level manager; re-dispatch their units one at a time through free slots, preferring reviews and rework of already-pushed PRs over new implementations.
model: claude-fable-5
Blocked on worker slots, for the top-level manager. Since 16:20 every wake has found both worker accounts at or over the cap of 5 (latest:
claude312 andclaude211 non-manager sessions), so mfer has dispatched nothing. Queued, in order: rebase of #108 and #105 (both passed review, conflict inTODO.mdonly), then reviews of #107 and #106. Needed: the accounts brought back under the cap, or one slot reserved for mfer.Model: fable-5-1
Still blocked on worker slots, for the top-level manager; new fact since the last note. At 20:01 both accounts hold 6 live workers, none of them mfer's:
claude2has three pixa sessions, one vaultik, one AutistMask and one homoicon;claude3has three vaultik sessions and an audit of vaultik's issue 73, one pixa, one AutistMask and one homoicon. Seven of them were started between 19:59 and 20:01, so other repo-managers are spawning while their account is already at 5. mfer has had one slot since 16:20 (used to rebase #108, which now waits for a fresh reviewer). Needed: a per-repo share of the cap, or one slot reserved for mfer.Model: fable-5-1
Repo-manager plan, 2026-10-03
Picked up after the 2026-09-29 halt; state read from the tracker and the branches.
Order of work
nextinTODO.mdonly: #106, #107, #105. Each is rebased by a worker, re-gated by a fresh reviewer on the currentnext, then squash-merged, one merge at a time. Theissue-87-pin-cli-errmsgsbranch has a commit and no PR: a worker finishes it and opens the PR.go_packagechange ruled there), #103, #77; then #100 (option A);fetchwork in sequence after 86: #63, #101, #66; tooling in sequence after 106: #102, #71, #70, #68, #73, #74, #90; #110 last among CLI changes.Until #76 is answered, every unit adds a
TODO.mdentry, so every merge forces a rebase of the others; merges stay one at a time.Standing rules for every worker and reviewer
mktemp -d /srv/code/tmp/ROLE-mfer-N.XXXXXX, thengit clone git@git.eeqj.de:sneak/mfer.gitinside it. Never under$HOME, never a shared checkout, never git worktrees.next; the PR targetsnext, nevermain. Pullnextbefore starting; immediately before pushing, fetch, rebase onto the currentnexthead, resolve any conflict yourself (aTODO.mdCompleted Steps collision is resolved by keeping every entry, yours on top), and re-run the checks.make checkandscript/cibuildonly (nevergo build/go test/go vet/linters directly). All linting runs in Docker. Wrap each container build in the shared lock:flock -w 1800 /srv/code/tmp/.docker-build.lock script/cibuild;flock -wexits 1 without running anything on timeout, so a 1 with no build output is a timeout, not a red gate: retry. Remove every container you start; nothing RAM-heavy, no container memory flags.trap, foregroundwait, recorded PIDs). Neverpkill,killall, or anything picked fromps.sed -i,perl -pi,awk, a python replace or a shell splice. Read the file and edit it by hand; formatting only throughmake fmt. Markdown changes are formatted withmake fmtbefore commit.TODO.md: follow its Workflow section (your Completed Steps entry in the same commit as the work) until #76 is answered.(closes #N); body is plain prose, at most about 120 words, then a blank line andModel: opus-5-5. PR body at most about 250 words, endingModel: opus-5-5. Never mention any AI company or assistant; no Co-Authored-By or session trailers or links.needs-review/needs-rework/needs-rebase/needs-checks, assignedclawbot. Never assignsneak.make checkresults. A failure is a summary: what, where, what acceptable looks like. Full links for every issue/PR mention (https://git.eeqj.de/sneak/mfer/issues/Nor/pulls/N), never bare#N. Escape raw angle brackets as HTML entities in comment bodies; no 4-byte emoji. Every comment endsModel: opus-5-5.nexthead immediately before gating; ifnextmoves before you post, re-gate. Pass = one comment: "Review passed." plus one-line disclosures only; leave the labelneeds-review, the manager merges. Fail = one comment with the findings as short summaries, labelneeds-rework. Never merge, never push to the PR branch.cd /srv/code/tmp, then onlyrm -rfof your temp dir.Model: opus-5-5
Handoff, 2026-10-03 16:30 UTC
This manager is being stopped with the primary account. The plan and standing rules above (#99 (comment)) still hold.
Landed today on
next: #86 and #115 (critical), #89 (built--no-cache, decided by the shared policy), #87. Runner issue #93 closed.Open units, in order (each PR carries a state note):
make checkon the shared host runs golangci-lint outside Docker, and its shared cache has given workers a false lint failure. New today: #116 (deprecated linter, small).Merging with
TODO.mdstill per unit: a reviewer gates the branch rebased onto the currentnextand names thatnextin the verdict; a worker then pushes that rebase; the manager checks withgit range-diffthat the pushed commit differs from the reviewed one only inTODO.mdcontext, and merges beforenextmoves. Ifnextmoves first, the PR needs a fresh re-gate.Model: opus-5-5
Handoff, 2026-10-04 20:10 UTC
Every open unit not waiting on sneak has landed on
next(#92 lists it). This manager is stopping.Waiting on sneak
simplelog).mainthere; #102 (Go 1.27) follows it, since the current lint pin cannot lint Go 1.27.For the next manager: merges stay one at a time; a PR reviewed against an older
nextis gated again on the rebased tree before it merges.Model: opus-5-5
Handoff, 2026-10-06 19:20 UTC
Every unit not waiting on sneak has landed on
next(#92 lists it), including the re-vendor fromsneak/prompts(#159, which also closed #116), Go 1.27 (#102), and sneak's two format rulings:atimedropped (#158) and file mode recorded (#161). This manager is stopping.Waiting on sneak
checkandfetchdo with a recorded mode; built to the recommended reading) and B (mimeTypeandctimeare never written).For the next manager:
https://git.eeqj.de/sneak/mfer/pulls/92is mergeable but not yet labelledmerge-ready; label it and assign sneak once its checks are green. Host-only steps (make test) never take the shared docker lock.Model: opus-5-5
Review toward 1.0, 2026-10-07
nextis green. Everything open that does not wait on sneak is filed; the units below run in this order, each one PR tonextthrough an independent review.--require-signatureaccepts a manifest signed by a key other than the required one.fetchandcheckread from a server.Still with sneak: #81, #82, #83, #147; the rest of #84 follows 81 and 82, and #80 waits for a release he declares.
Model: opus-5-5
Every unit in the review above has landed on
next, plus one more defect found since: #177 (checkandfreshenused the current directory instead of the manifest's).Waiting on sneak, nothing else open:
gen DIRwrites its manifest and how it lists the files.Model: opus-5-5
Landed on
nextsince the note above, each through an independent review:genwrites its manifest (#178).gpg(#181).genandfreshenlisted nothing for a directory reached through a symlink, andfreshenemptied the manifest there (#185, critical); the library's single-path scan given a file (#182).Waiting on sneak, nothing else open: the remaining wire format questions (#81), the API surface and determinism defaults (#83), and the logging library (#147). The format specification (#84) follows the wire format answers; the release mechanics (#80) wait for a release he declares.
Model: opus-5-5
Repo-manager plan, 2026-10-12
Picked up after the 8 to 12 October outage; state read from the tracker and
next.Everything sneak ruled on #81 (drop
atime, the file mode field), #82 and #178 is onnext, except the signature part of the format document, now #187:docs/FORMAT.mdstill describes the signed string and thesignerfield differently from the code. It is in progress.Also running: one audit of
nextfor real defects, mainly in the new OpenPGP signing code and in the recent path changes togen,check,freshenandfetch. Confirmed defects are filed as their own issues and built in order of severity, one PR each through an independent review. The worker rules from #99 (comment) still hold, except that there is noTODO.md.Waiting on sneak: the rest of #81, #83 and #147. The rest of #84 follows the wire format answers.
Model: opus-5-5