script/cibuild was a plain docker build ., so on a tree Docker had
seen before every check step came from the build cache and the build
still passed. It is now the org model from sneak/prompts, byte for
byte: script/bootstrap, script/check, then docker build --no-cache
with the git describe version as the VERSION build argument.
The workflow puts ~/.local/bin, where bootstrap links what it
installs, on the step's PATH. Bootstrap now installs its pinned node
when the installed one is older than 22.12.0, the oldest the
frontend's dependencies accept (puppeteer-core's engines field), as
it already does for Go against backend/go.mod.
Model: opus-5-5
golangci-lint v2.12 deprecates gomodguard, which the org .golangci.yml
reached through "default: all", so every lint run printed a
deprecation warning. backend/.golangci.yml is now the current copy
from sneak/prompts, fetched unedited: gomodguard is disabled and
gomodguard_v2 enabled with the org block list. The new file also
turns depguard on with its test-support rule, which forbids
net/http/httptest outside test code. netwatch has no test-support
packages of its own to add to that rule, so the file is identical to
the canonical one. backend/script/lint checks the new sha256. The
backend raises no findings under the new rules.
Model: opus-5-5
nginx sent none of the security headers REPO_POLICIES.md requires.
security-headers.conf now sets all six with always, included at server
level and again in /assets/, whose own add_header would otherwise drop
them. nginx hides the copies netwatch-server sets, so /api/ and the
health check carry each header once. The content security policy
allows no inline script or style; the host row's status dot took its
grey from a style attribute, now a class. connect-src is * because
several probed hosts redirect to other hosts and the browser checks
every redirect against it. Referrer-Policy is no-referrer, as the
backend already sends.
Model: opus-5-5
The request log wrote the URL, User-Agent, Referer and other
request-supplied strings with no length limit, and the server accepts
headers up to 1 MiB, so one request could put about 1 MiB per field
into a log line. Every string the request log takes from the request,
including the request ID chi copies from X-Request-Id, is now cut to
the 128-byte bound the report handler already used. That bound and
its helper moved from the handlers package to the logger package so
both use the one copy.
Model: opus-5-5
nginx trusted X-Forwarded-For from every RFC1918 address, so a client
reaching it from one could write a new address on each request and
get a fresh rate-limit allowance. The container's TRUSTED_PROXIES now
names the reverse proxies nginx trusts, none by default.
bin/entrypoint.sh makes each entry a CIDR, checks it with the new
"netwatch-server check-cidr", which runs the server's own
TRUSTED_PROXIES parsing, and writes one set_real_ip_from line per
entry into /etc/nginx/trusted-proxies.conf, which nginx.conf includes.
The backend is started with TRUSTED_PROXIES=127.0.0.1/32, since nginx
is its only client. The viewport test mounts an empty file there.
Model: opus-5-5
Report files were named by a millisecond timestamp and created with
O_EXCL, so two flushes in the same millisecond, such as a flush for
size and the final flush at shutdown, got the same name and the second
failed, losing its reports. Each name now carries a number after the
timestamp that goes up by one for each file the server starts to
write, so names still sort by time and never repeat within a run. A
failed write uses up its number, leaving a gap if the file could not
be created and otherwise a file under that number that may be
incomplete.
Model: opus-5-5
The image's HEALTHCHECK requests /.well-known/healthcheck through
nginx on the port from PORT, so it fails unless both processes answer.
The backend reads PORT and DEBUG with strconv instead of viper, which
turned a bad PORT into 0 and a bad DEBUG into false. Those, and a
BIND_ADDRESS that is not an IP address, now stop the start with an
error naming the variable; the TRUSTED_PROXIES error names it too.
bin/entrypoint.sh also refuses a container PORT outside 1 to 65535,
or 8081, where the backend listens, naming PORT. README.md gains
"Running under upaas". Its first-run steps create the host directory
owned by uid 1000, so the image changes no ownership.
Model: opus-5-5
nginx.conf is now a template the nginx image renders into conf.d at
container start. bin/entrypoint.sh sets PORT to 8080 when unset or
empty, and stops with an error before starting anything when PORT is
not digits only: nginx would take a value such as localhost or
unix:/tmp/x.sock as an address and start anyway. NGINX_ENVSUBST_FILTER
limits the rendering to PORT, so $uri, $host and every other nginx
variable pass through unchanged. server_tokens off drops the version
from the Server header and error pages. script/frontend-viewport-test
renders the template the same way. EXPOSE still documents 8080; the
backend stays on 127.0.0.1:8081.
Model: opus-5-5
POST /api/v1/reports stays unauthenticated but is bounded. Each client
address, as the trusted-proxy logic resolves it, may send
REPORTS_PER_MINUTE reports a minute (default 60, counted by
go-chi/httprate over a sliding minute); past that it gets 429 with
Retry-After. reportbuf refuses a report that would take the report
files past DATA_DIR_MAX_BYTES (default 1 GiB), counting the files
already in DATA_DIR and unwritten reports at their uncompressed size;
the handler answers 507. CORS adds nothing unless CORS_ALLOWED_ORIGINS
lists origins. A limit that is not a positive number, or an origin
that is not a plain scheme://host[:port], stops the server from
starting.
Model: opus-5-5
The root Dockerfile builds the only image; Dockerfile.backend is gone.
Its stages: lint, a Go stage that runs the tests and builds
netwatch-server, the node stage, and an nginx runtime. nginx serves
dist/ on 8080 and proxies /api/ and /.well-known/healthcheck to the
backend on 127.0.0.1:8081. bin/entrypoint.sh starts both, turns TERM or
INT into a stop of both, and exits non-zero when either exits on its
own. The backend runs as user netwatch and keeps reports on the /data
volume. New setting BIND_ADDRESS (empty: every interface). STOPSIGNAL is
SIGTERM, since the nginx image's SIGQUIT would miss the entrypoint.
script/docker is the org model verbatim.
Model: opus-5-5
Root make check, and with it the pre-commit hook, now gates the Go
backend too. The backend's Makefile targets are shims over
backend/script/*; script/cibuild builds both images and is the
workflow's only build step. Root make test runs both halves within one
30-second timeout.
Root make lint runs golangci-lint only in Docker, by building the lint
stage of Dockerfile.backend without the cache; the .golangci.yml drift
check moved into backend/script/lint. script/bootstrap installs no
linter: it reuses a Go at least as new as backend/go.mod asks for,
otherwise installs the pinned, hash-verified release, linked into
~/.local/bin without replacing anything it did not create. With VERSION
unset or empty, the backend version falls back to git describe inside a
git checkout, then to dev.
Model: opus-5-5
A Reporter beside AppState POSTs each host's unreported, non-paused
samples to /api/v1/reports every reportInterval (default 60s).
buildReport is an exported pure function of host state; init() runs only
when #app exists, so a test can import the module. A per-host mark
advances only on a delivered POST. At most one report POST is pending at
a time and it is abandoned after half the interval, so a slow POST never
overlaps the next report and a mark never moves backwards. The samples of
an abandoned POST are sent again at the next interval, so a backend that
stored them but answered late receives them twice. Failure is quiet and
never blocks probing. vite.config.js proxies /api for yarn dev.
Model: opus-5-5
A report the buffer refuses now returns 500 instead of a false `ok`.
Reports reach disk later, so a failed disk write is still answered 200
and shows in the log, and at shutdown as a failed stop with a non-zero
exit. An over-limit body returns 413; malformed JSON stays 400. A
MaxBodyBytes middleware caps every route at 1 MiB; a route group can
only lower that limit. The raw geo blob is no longer logged; client_id,
timestamp and decode error text are cut to 128 bytes before logging.
Panic recovery logs the panic value and stack through slog.
Model: opus-5-5
The old backend/.golangci.yml declared version "2" but used v1 schema
keys, so under v2 it never validated and its thresholds were inert: the
linter ran at defaults. Replace it verbatim with the org-standard file,
repin the Dockerfile.backend lint stage to golangci-lint v2.12.2, and
assert the config's sha256 as the first step of the backend lint target
so it cannot silently drift again -- a local hash check, no network.
The standard config surfaces findings only in the tests: the repeated
IP literals in middleware_test.go become named constants (goconst) and
its request switches to NewRequestWithContext (noctx). reportbuf.go's
gosec suppression gains a plain justification comment. The rest of the
backend, including the fx-based server lifecycle, is already clean.
TODO.md updated.
Model: opus-4-8
The server ran os.Exit at the end of its own goroutine, racing fx's
teardown and sometimes killing the process before reportbuf's OnStop
flushed — silently losing a full flush window of telemetry on every
restart, at exit 0. Shutdown now goes through fx.Shutdowner, so every
OnStop runs in order.
The http.Server is built synchronously in OnStart before the serving
goroutine, so shutdown can no longer race or nil-deref it. A listen
failure exits non-zero via fx.ExitCode(1). reportbuf's OnStop is guarded
by sync.Once. writeTimeout now exceeds the chi per-request budget so that
budget is reachable. Dead startupTime, exitCode, and cancelFunc fields
are gone. A new test asserts a buffered report reaches disk after the
lifecycle stops.
Model: opus-4-8
`make frontend-viewport-test` builds `dist/`, serves it from the same
digest-pinned nginx image and nginx.conf the shipping container uses, and
drives a digest-pinned headless Chrome over CDP. Viewport widths are derived
from the app's own @media breakpoints rather than a list of phone models: each
breakpoint is tested one pixel below, on, and above, plus four anchor
viewports. Assertions are on computed layout — horizontal overflow, off-screen
elements, clipped text, 44x44 tap targets, host-row reflow — not screenshots,
and each check declares the minimum elements it must find so a stale selector
fails instead of passing blind against a page it is not measuring. Kept out of
`make check`: it needs Docker and takes minutes. Proven able to fail before
being trusted.
Model: opus-4-8
Add ReadHeaderTimeout and IdleTimeout to the http.Server as named constants beside the existing timeouts. Add a SecurityHeaders middleware (HSTS, a JSON-API CSP of default-src 'none'; frame-ancestors 'none', X-Frame-Options DENY, nosniff, Referrer-Policy, Permissions-Policy), registered before CORS so preflight responses carry it. Resolve the client IP from X-Forwarded-For / X-Real-IP only when the direct peer is in the trusted-proxy allowlist (loopback plus RFC1918 by default, configurable via TRUSTED_PROXIES); an untrusted peer's forwarded headers are ignored. Uses net/netip; no new dependency.
Model: opus-4-8 (implementation and review); claude-fable-5 (merge)
clawbot