The event log now loads only the 50 newest events, limited in its query, and only the newest starts expanded; the rest start collapsed. Paging is removed rather than capped, since 50 events never need a second page: the Previous and Next links, the `page` query parameter and the page number Replay and Resubmit carried back are gone, and a `?page=` left in an old link shows the 50 newest. A webhook with more than 50 events reads "50 most recent of N events" beside the heading. Comments and a README line that called `page` the only query parameter the service reads now name `next` and `notice`.
Model: opus-5-5
An event's row in the event log was a button element, whose text a browser will not let be selected, so an event's ID could not be copied. The row now has the button role instead: focusable, toggled by Enter and Space, with `aria-expanded` giving its state. A click on its text toggles the event after 500 ms, the usual double-click interval; the second press of a double- or triple-click cancels that, so selecting the ID leaves the event as it was. A drag over text does not toggle, and a click on the caret toggles at once. The browser test clicks as a person does, so toggling on the first click fails it.
Model: opus-5-5
A target whose circuit breaker had tripped still showed as Active, and its deliveries sat at a bare "retrying" with no attempts. Its row on the webhook page now says deliveries are paused after repeated failures and until when the cooldown ends, adding that one waiting delivery is then sent to test the target; while half-open it says deliveries are held while one tests it, with no time. Waiting deliveries show "next try no earlier than" the later of the cooldown and their own backoff, with the date when not today. The engine gains one read of a breaker's state and remaining cooldown under one lock, and shares the backoff formula.
Model: opus-5-5
A database target's archive expiry was typed by hand as never or a raw duration such as 720h, and the target list showed it back raw. Adding or editing a database target now offers the new-webhook page's list of choices (never, 1h, 12h, 24h, 30d, 90d, 365d), defined once and shared by all three forms. The edit form starts on the stored expiry, or on the submitted one after a refused save; a stored value outside the choices is listed under its own value, so saving unchanged keeps it. The target list shows the expiry in plain units: "30 days", "12 hours", "never".
Model: opus-5-5
A refused save on the target edit page answered with a bare text page, losing the form and everything typed, and the webhook edit page came back with the stored values instead of the submitted ones. A refused target edit now shows the edit form again with the reason above it and every value submitted, with the same status codes as before; a refused webhook edit keeps the submitted name, description and retention. Target edits use the same validation as new targets, with no second copy; an encoding or database failure stays a logged 500. The browser test covers a refused save on both pages, and its main function is now a plain list of checks.
Model: opus-5-5
The new-webhook page gains an optional HTTP target URL, which creates an http target named HTTP, and an archive checkbox whose pruning choice (never, 1h, 12h, 24h, 30d, 90d, 365d) creates a database target named Archive with that expiry. Both are validated by the add target form's own validation, and the webhook, its entrypoint and its targets are created in one transaction or not at all. A refused form comes back with the reason and every value entered, retention included. The targets can be renamed on the webhook page like any other.
Model: opus-5-5
The target list showed nothing about how a target's deliveries were going. Each target now shows Delivered and Failed, each in total and in the last 24 hours. The totals are the per-target running totals kept for the statistics pane, so retention does not reduce them; the 24-hour figures are one count over the deliveries' final-status index, for all of the webhook's targets at once. Pending and retrying deliveries count in neither. If the event database cannot be read, each row says so instead of showing zeros. The archive details and Download button on database targets are kept.
Model: opus-5-5
The targets section of the webhook page showed its add form open with every field, a URL included for types that use none. It now lists only its targets until "+ Add" is clicked; "+ Add" shows a choice of target type with Next and Cancel on one row, and Next shows the name and only that type's fields. The database and log types show no URL field and the server stores none for them; the slack form gains its retry field. A refused target brings the page back with the form open on its type, the values entered and the reason, and Cancel empties it. An encoding failure stays a logged 500. Target validation returns its message, so the new-webhook page can reuse it.
Model: opus-5-5
Each event now has its own page at /hook/ID/events/EVENTID, behind the login, showing its details, its whole body and every delivery; a resubmitted copy links to its original's page. The recent events on the webhook page link there and expand to show their bodies, only the newest expanded on load. One renderer and one template show a body the same way in the recent events, the event log and the event's page: whole up to 32 KiB, cut there in the two lists with links to the event's page and the download; JSON pretty-printed unless that would grow it past four times plus 1 KiB; over 200 lines in a scrolling box; a body holding NUL or control characters treated as binary and never dumped raw.
Model: opus-5-5
An entrypoint's description was set when it was added and could never change, so renaming one meant deleting it and adding a new one with a new URL every sender had to be given again. Each entrypoint on the webhook page now has an Edit button, in the shared secondary style, that opens its description in place with Save and Cancel and keeps its URL. The save goes through the same login, CSRF and ownership checks as the other entrypoint actions; an empty description shows as "Entrypoint". Activate and deactivate now write only the active column, so they cannot undo an edit. Tests cover each, through the router and the browser.
Model: opus-5-5
Each database target on the webhook page has a Download button that streams its archive as gzipped JSON, archive-WEBHOOKNAME-TARGETNAME-TIME.json.gz, with names made safe by delivery.ArchiveFileName's function. The export reads one consistent snapshot through one cursor in a read-only transaction, so archive writes carry on, and holds the rename lock only while it reads the stored names and opens the file. It extends its write deadline as it writes, so a large archive downloads for as long as the client reads; a failure after the response has started aborts the connection so the browser marks the download failed. The request limit is now the service's own middleware, which no longer writes a 504 over a response already started.
Model: opus-5-5
The HTTP drain at shutdown waited up to ShutdownTimeout regardless of how much of the stop budget earlier hooks had used, so a slow archive sweeper or retention reaper could eat the reserve the hooks after the server need, and the database close was skipped. The drain now waits at most the shorter of ShutdownTimeout and what is left of the budget less TailHookReserve, as the Sentry flush already does. The reserve is documented as derived from the two timeouts. Tests cover earlier hooks having spent part of the budget, on a clock that host speed cannot move, and pin that a drain on the full budget gets all of ShutdownTimeout.
Model: opus-5-5
Two comments named the wrong mechanism: loadResubmitSource credited soft-delete for refusing a reaped event, though the retention reaper deletes event rows outright, and createAndFanOut claimed to be the only path that creates deliveries, though per-delivery replay creates one without an event. Both now say what the code does. The resubmit route's middleware had no tests through the router; new tests drive the production router to pin the refusal without a valid CSRF token, the rate limit, signed-out requests never spending it, and another webhook's event refused by the event lookup while the user's own event is accepted. Each fails with its check removed.
Model: opus-5-5
Many controls were plain coloured text with no sign they could be clicked: Edit, Activate, Deactivate, Delete, Replay, Resubmit, both Add controls and the copy control beside each entrypoint URL. Every clickable control is now a real button or link in one of two shared styles: the buttons in input.css, now with a pointer cursor, and btn-small, a small bordered secondary action in style.css, which the layout now loads. The site name, the footer links and each webhook list card show at rest that they open something, with focus states. The copy control shows "Copied" after a click. Rows on the webhook page and in the event log wrap at phone width. The browser test covers the copy control.
Model: opus-5-5
Follow-ups from an August review of the body cap, each checked against the current tree. One route test now requires an oversized POST, with no session and no CSRF token, to be refused with 413 before CSRF runs, in every page route group with a POST route and in /settings/, so moving a group's body cap after CSRF fails it. The three test router helpers build the server through New with a lifecycle that is never started, so no field is set by hand. The middleware test comment names runMaxBodySize, and the MaxBodySize doc comment says methods other than POST, PUT and PATCH pass uncapped on purpose. The README item was already settled.
Model: opus-5-5
Every page's Content-Security-Policy forbids eval, which the standard Alpine.js build needs, so no directive ran in a browser: both add forms on the webhook page showed open, and events in the event log could not be collapsed. The UI now loads Alpine's CSP build (@alpinejs/csp 3.14.9 in 3p/); the policy is unchanged. Each directive names a property or method of a component registered in static/js/app.js (collapsible, targetForm), and each card holds its own x-data. A browser test, built only with the browser tag, loads the webhook page and the event log under the real headers; make test-browser runs it in Docker. New test-only dependency chromedp, which raises golang.org/x/sys to 0.47.0.
Model: opus-5-5
The receiver route /h/{uuid} has no MaxBodySize middleware, unlike the page route groups; its 1 MB cap is enforced in the handler, which owns the response senders see. Nothing at the route said so, so a reader could take the receiver for uncapped or remove the only bound on the one unauthenticated endpoint. The route registration and the handler's body-reading function now say where the cap lives and why. A routing test sends a body of exactly 1 MB and one a byte over through the production router and pins that only the second is refused, with the handler's 413 and message.
Model: opus-5-5
Each database target now writes its own archive file, archive-WEBHOOKNAME-TARGETNAME-TARGETID.db, named by delivery.ArchiveFileName, in place of one archive per webhook keyed on its UUID. Renaming a webhook or a target renames its archive files (with any -wal and -shm) before the new name is saved, never over an existing file, and moves every one back if a rename or the save fails. The webhook edit, the target edit and target creation share one lock so no two interleave. Deleting a webhook or target leaves its files on disk. Nothing looks for the old archive-WEBHOOKID.db files. The README gives the naming and the recovery steps.
Model: opus-5-5
MAINTENANCE_MODE did nothing but make the healthcheck JSON report maintenanceMode: true; no request was ever served differently, so an operator who set it expecting requests to be refused got nothing. It is removed from the configuration, the startup configuration log line, the healthcheck JSON, the README and the Settings page, together with the uncalled Server.MaintenanceMode method and the healthcheck's dependency on the configuration. A leftover value in an environment is ignored like any other unknown variable.
Model: opus-5-5
The access log's response writer and the metrics middleware's writer hid the writer beneath them, so a handler's flush, hijack or write deadline set through http.ResponseController failed with "not supported" behind them. The access log's writer now has Unwrap. The metrics middleware calls the library's public Measure with a writer of our own that has Unwrap, in place of std.Handler's writer, so the middleware order and what metrics record are unchanged. A test over a real connection sets a write deadline and flushes, metrics on and off, on a global route and in an admin page group, and fails without either Unwrap.
Model: opus-5-5
The delivery duration histogram was registered only when a delivery first ran, so until then /metrics had no series for it and a dashboard or alert on it saw nothing at all rather than zero. Each of the four target types now has its series registered at boot with zero counts, so the histogram is served from the first scrape. Tests pin that all four series are present before any delivery, on the metrics registry and through the production router.
Model: opus-5-5
A new page at /settings, linked from the navigation bar and behind the login, lists every configuration field the server loaded at startup: its environment variable, the README table's description, and the value in effect. METRICS_PASSWORD and SENTRY_DSN show only as set or not set; their values are replaced before rendering and never reach the template. The route is GET only and its group is built like the other admin page groups. Tests set the credentials one at a time so each value shown is checked against its own field and a set secret never appears in the page.
Model: opus-5-5
Every test that builds an fx app with fxtest.New (handlers, server, resetpw, gormlog, config) now passes fx.NopLogger, so fx's own log no longer goes to t.Logf. A hook still running after a start or stop timeout can then no longer write to a test that has already returned, which the race detector reported as a data race. What the tests assert is unchanged, and nothing about the race detector is suppressed.
Model: opus-5-5
Saving, deleting, activating or deactivating a webhook, entrypoint or target, and signing out, now land on their page with a one-line notice such as "Webhook deleted." or "Signed out.". The redirect carries a fixed code that maps to fixed text; an unknown code shows nothing, so nothing from the URL is ever echoed. One partial in the page layout shows the notice on every page, and replay and resubmit now use the same codes and partial. Error pages show no notice.
Model: opus-5-5
A second metrics-enabled router in one process panicked on a duplicate collector registration, because every collector registered on Prometheus's global default registry. metrics.NewRegistry now builds one registry with the Go runtime and process collectors; fx provides it and the delivery metric set built on it. The middleware builds its HTTP recorder once on that registry (NewForTest on a fresh one), the engine and handlers take the metric set from fx, and nothing registers on the global default any more.
/metrics is served from the new registry with the same series names, labels and auth. A test builds two metrics-enabled routers in one process.
Model: opus-5-5
Every 400, 403, 404 and 500 on an admin page now answers with an error page in the normal layout: the status, one fixed line explaining it, and a link back to the webhook list, or to sign-in when nobody is signed in. Unknown paths reach it through the router's not-found handler, a refused form token through the CSRF middleware, and a panic through a recoverer each admin page route group installs first. Status codes are unchanged, and the page always sends Cache-Control: no-store.
If the error page fails to render, the answer is the same status in plain text; if it panics, the answer is a 500. The receiver, the healthcheck and /metrics keep their plain answers.
Model: opus-5-5
A logged-out GET of an admin page now redirects to /pages/login with its path and query in a next parameter, when they fit in 2048 bytes. The login form carries next as a hidden field; a successful login redirects there, and a failed one shows the page again with the same next. A POST still redirects to plain /pages/login.
The value is client-chosen, so every read of it goes through one check: after percent-decoding it must start with exactly one / and contain no backslash or control character; anything else becomes /. gosec's open-redirect rule is suppressed on the two redirects that follow it. The login page's navigation bar no longer links to itself.
Model: opus-5-5
The webhook page and everything under it move from /source/ID to /hook/ID; the list and new-webhook form from /sources to /hooks and /hooks/new; the event log from .../logs to /hook/ID/events; entrypoint URLs, the ones senders post to, from /webhook/UUID to /h/UUID. The old paths are simply gone, per the owner's ruling. Every link, redirect, form action, test, comment and README mention follows; every link to the event log, and its title and heading, read "Full Event Log".
Keeping the UUID out of logs, metrics labels and error reports, and the receiver rate limits, key on the route itself, so they cover /h/ unchanged. Routed tests follow every link and form the pages render to a moved page.
Model: opus-5-5
Unset or empty, TRUSTED_PROXIES now defaults to 10.0.0.0/8, 172.16.0.0/12 and 192.168.0.0/16, so a reverse proxy reaching webhooker from one of those ranges gets per-client rate-limit buckets with nothing set. A set value replaces the default entirely; an unparseable one still fails startup. The startup warning for an empty list is gone.
The README gives the default, one rule (if any client can reach webhooker, or the proxy in front of it, from an RFC 1918 source address, set the list to the proxy's address alone), and where to find that address: the remoteIP field of the http request log line. Loopback is not in the default.
Model: opus-5-5
The build no longer downloads Alpine.js. Its npm package tarball is committed as 3p/alpinejs-3.14.9.tgz, byte for byte the file script/fetch-assets downloaded, with the sha256 that script pinned. script/assets (make assets) extracts package/dist/cdn.min.js to the ignored static/js/alpine.min.js; script/test runs it, so make test, make check, the pre-commit hook and the Dockerfile get the file with no network access, and make build, run and dev run it too.
Removed: script/fetch-assets, its Dockerfile step, static/vendor.sha256 and static/vendor_test.go. The README describes the new flow.
Model: opus-5-5
A new database brings a new session key. A browser still holding the
old session cookie got a 500 on a correct login: Session.Get returned
the cookie's decode error and the login handler answered it with a
500. Get now treats a cookie that does not decode as absent, and
logging in replaces it. gorilla/csrf already did the same for the
CSRF cookie.
A start that creates webhooker.db now logs "created a new, empty
database" at WARN with its path, shortly before the first-boot banner,
so an unexpectedly empty DATA_DIR is noticed.
The codec tests now decode through the store, since Get no longer
reports the codec's reason.
Model: opus-5-5
The static file server was attached with Mount, which registers every
method, so POST, PUT and DELETE on an asset were answered 200 with the
file. It is now registered for GET and HEAD only, inside a /s group
whose method-not-allowed handler answers 405 with Allow: GET, HEAD.
A method chi does not route at all, such as PROPFIND, still gets 405
from the top-level router, without Allow. The inverted test and the
README route table say the same.
Model: opus-5-5
There was no redelivery path anywhere: once a delivery exhausted
max_retries it was failed permanently, even though the event body is
durably stored. Storing an event and being unable to re-send it defeats
the reason it is stored, and the ordinary case is a destination that was
down longer than the backoff ladder.
Adds POST /source/{sourceID}/deliveries/{deliveryID}/replay, inside the
authenticated group so it inherits MaxBodySize, CSRF, NoCache and
RequireAuth. Replay creates a NEW pending delivery against the target's
CURRENT config and hands it to the engine through the same notifier the
receiver uses, so it runs the normal path with the retry ladder, the
SSRF-guarded transport and the circuit breaker. The original delivery's
rows are never touched, and the stored event body is re-sent, never the
recorded response.
Replay is refused, with a distinct message, for a non-terminal delivery, a
deleted target, a deactivated target, and when an earlier replay of the
same event and target is still in flight. Bounded by a per-client rate
limit and by that in-flight check.
The new delivery row is written with Omit(clause.Associations) and with
neither Event nor Target populated, so it cannot upsert a targets row into
the per-webhook event database (#206).
Counted by webhooker_delivery_replays_total on the existing target_type
label. A replay also moves the ordinary attempt, outcome and duration
series, because it is a real delivery.
check / check (push) Superseded by a newer commit; never tested
The receiver had no inbound authentication of any kind: /webhook/{uuid}
was mounted behind a rate limiter alone, so the only thing protecting an
entrypoint was the secrecy of a v4 UUID in a URL path. Inbound headers are
forwarded almost verbatim to the target, so anyone who learned the URL
also chose the headers the downstream service received.
Adds an optional per-entrypoint secret with two schemes: github
(X-Hub-Signature-256, HMAC-SHA256 hex over the raw body) and gitlab
(X-Gitlab-Token, a plain shared token). Comparison is constant-time, the
HMAC is computed over the raw body before any parsing, and rejection
happens before persistence -- an unauthenticated request creates no event
row. An entrypoint with no secret behaves exactly as before, including
every row that predates this change.
The scheme's credential header is stripped from the header map before it
is marshalled into Event.Headers, so the GitLab token reaches neither the
event store nor any delivery target. SchemeInfo.HeaderIsDigest defaults to
false meaning strip, so a scheme added later is protected unless its
header is positively declared a digest.
#160 scrubbed the Sentry body, query, cookies, env and headers but kept
Request.URL, which the SDK builds from the concrete path. On the
receiver that path is /webhook/<uuid> in full — a write capability, not
an identifier: anyone holding it can inject events the operator's
targets then deliver. #146's "2xx and 5xx keep the concrete path" ruling
was reasoned about a log the operator owns and does not transfer to a
tracker with its own retention and access control.
The chi route pattern now replaces the path on every route, reached via
the request the SDK carries on hint.Context. Unconditional, because a
route-conditional rule leaks on any route someone forgets to add, and on
a static route the pattern is the path anyway. The fallback is never the
concrete path.
Also rewrites event.Transaction, which carries the same UUID on the
sibling dispatch and which the issue did not name. Tracing is off today,
so that half is a floor rather than a live fix — and it is why enabling
tracing later needs #185 first, or every transaction collapses into one
bucket.
Independently reviewed. The reviewer ran fourteen adversarial probes —
404 and 405 panics, panics in middleware before and after routing,
direct CaptureException, mounted subrouters, wildcards, tracing on and
off — and found no path where the concrete URL survives, and no third
field carrying it.
Merge note: the final round was a two-comment documentation fix on an
already-passed review, correcting a rationale that called the host
operator configuration when it is the client's Host header. I verified
that amend is comment-only myself rather than spending a fifth review
round on it.
r.FormValue falls back to the query string, so
POST /source/{id}/targets?url=<secret> created a working target from a
value carried on the request line — where proxy logs, browser history
and Referer all record it. Every form read is now r.PostFormValue,
including the login password and both password-change fields, which had
the same defect in a more acute form.
The Sentry leg needed more than the query string. sentryhttp attaches
the whole request to the scope, and ApplyToEvent copies the teed body
into Request.Data with no SendDefaultPII guard — so reading every field
from the body only pointed every credential this change protects at the
one field the first revision did not scrub. Body and query are now
redacted, Cookies and Env cleared, and Headers reduced to an allowlist,
because the SDK's own filter removes four names and would otherwise ship
X-Csrf-Token and the shared secrets senders put on the receiver route.
Also adds json:"-" to Target.Config, APIKey.Key and Setting.Value —
TargetView is the masking barrier for the HTML path only, and the first
handler to marshal a model would serialise a bearer token or the session
encryption key.
Independently reviewed three times. The second review found the Data
leak and proved it with a scratch module; the third disproved the
PR's own claim that BeforeSend gets no request, so the README now
records that redacting unconditionally is a deliberate choice rather
than a limitation — which is what makes #179 cheap to fix.
In the shipped default, any stranger denied the operator the only
administrative path at 5 requests per minute: TRUSTED_PROXIES is empty,
the README requires a reverse proxy, so every login POST shared one
bucket keyed on the proxy.
Credentials are now verified first and only a FAILED attempt spends
budget, so a correct password is never throttled. Failures are counted
per (client bucket, submitted username), bounded. Concurrent Argon2id
verifications are capped at two, and the queue for them at 16 — because
verifying first lets an attacker force a 64 MB hash per request, and
bounding the wait alone bounds nothing.
The issue's own recommendation was insufficient and is rejected here:
keying by username stops an attacker locking out a DIFFERENT account,
but this is a single-admin product with a predictable bootstrap
username, so flooding the operator's own name still locks them out.
This is speculative — it implements a corrected recommendation ahead of
the owner's ruling so the decision can be made by merging or reverting.
Three things are disclosed rather than glossed: online guessing rises
from 5/min to roughly 27/s, because the 429 is a label on the response
and not a gate in front of the hash; the residual exposure is a loss of
login AVAILABILITY, not latency, and a determined flood still denies
login while it runs, at ~400x the cost and clearing the moment it
stops; and the endpoint should be provisioned for ~400 MB resident, not
the 203 MB of live commitment it itemises.
Independently reviewed four times. Reviewers disproved the suspected
FIFO starvation by measurement, then caught two successive memory
bounds the code did not have — the second by parking waiters and
reading the heap rather than checking the arithmetic.
The 8 KB render cap from #135 left storage untouched but no route served
the rest, so a body over the cap was reachable only with filesystem
access to the SQLite files — in a product whose purpose is storing
webhooks so they can be inspected.
GET /source/{sourceID}/logs/{eventID}/body serves the whole body to the
webhook's owner, as application/octet-stream with an attachment
disposition and nosniff. Those are a security control, not formatting:
the bytes come from the public receiver and are handed back inside the
operator's authenticated origin, and the existing CSP would not stop a
stored HTML payload executing there. The truncation marker links to it
only when a body was actually cut.
Accepted deviation, documented rather than glossed: #157's definition of
done asks the route to stream from the row. It buffers whole instead,
because database/sql exposes no incremental handle on a SQLite BLOB and
substr range reads re-materialise the entire column per call — an
earlier revision chunked at 64 KiB and was 11-15x slower for a worse
bound. Three independent reviewers confirmed no streaming path exists.
Independently reviewed three times. Two earlier revisions each asserted
a memory bound the code did not have; the final reviewer measured
2.057x at the ingest cap and pinned the two overlapping allocations from
source — the driver's column buffer and database/sql's convertAssign
clone — confirming the stated "roughly two bodies, and 2x is a floor
not a ceiling" is now accurate, since SQLite's own materialisation sits
outside the Go heap.
fx defaults to a 15s stop timeout and the Dockerfile sets no grace
override, so Docker SIGKILLed at 10s and the bounded shutdown #130 built
— including the log line that tells an operator a component is wedged —
was unreachable in the image this repo produces.
Sets fx.StopTimeout to 5s, and lowers the HTTP drain to 3s so a
full-length drain no longer exhausts the whole sequence budget and skip
every later hook, database close included. The Sentry flush, which runs
in the same hook and honours no context, is clamped to the remaining
stop budget less a 2s tail reserve, so a stalled flush drops Sentry
events rather than the database close.
Also fixes a latent coin flip in the shared stop-hook waiter, which
reported "shutdown timed out" about half the time for a component that
drained cleanly against an already-expired context.
Independently reviewed three times. The final reviewer derived a
stronger invariant than the implementation claims — the server hook's
absolute end is bounded at stopTimeout minus the reserve regardless of
drain length or of time consumed by preceding hooks — and confirmed the
guard's 10ms sweep cannot step over the maximum, since both breakpoints
land on its grid. Both Sentry probe arms, the docker stop demo and every
mutation were reproduced independently.
Known residual, filed separately: the HTTP drain itself is not clamped
by the reserve, so slow preceding hooks can still jointly exhaust the
budget. Demonstrated with a 2.2s sweeper delay.
clawbot