Start puts a per-attempt nonce in the returnTo it sends Heimdall and keeps the attempt (provider,
member flag, local return path, expiry) in a Data Protection cookie named for that nonce. Heimdall's
completion page sends the browser to /auth/heimdall/complete with a code in the URL fragment; the
page posts it back, Bifrost redeems it at Heimdall, requires the redeemed returnTo to name the
cookie's nonce, verifies the access token for the redeemed account, and signs that browser in.
Deleted: the callback route, the wait route, the server-side attempt store, the pending state and
the backend_callback handoff. Added: LocalReturnPath for return targets, Heimdall:AppSecret sent as
x-heimdall-app-secret (member starts are refused without it), a short timeout on the key fetch, and
a key lookup that does not hold the fetch gate for cached keys. Refusals at the completion route all
answer one redirect and log a fixed reason.
Tests are rewritten against a fake Heimdall with start and redeem routes. Not compile-checked by this
commit alone; later commits record the verification.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The forgery and wrong-claim tests now assert which rule refused, start failures,
blank attempt ids, capability shapes and date extremes are covered. A token whose
exp is the largest representable date no longer overflows the clock-skew sum, and
the unreachable blank-return-url fallback is removed (Program defaults it).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The verifier was registered as a typed client, which is transient, so each request
would have started with an empty key cache. It is now a singleton that takes its
HTTP client from the factory per fetch. Test hosts share one in-memory database
root so EF keeps a single service provider.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The callback verifies the EdDSA access token against Heimdall's JWKS and takes the
account, display name and member access from its claims; attempts expire, are
bounded, are completed only if issued, and are redeemed once with a per-attempt
binding cookie. Not yet built or tested; the first verification run follows.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Drops antiforgery from the anonymous vote flow: no antiforgery cookie, no hidden token.
CastRequestGuard is the one decider; VoteModel.OnPageHandlerSelected is the one gate and
answers 4xx when no handler is selected (was a NullReferenceException 500).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Empty, NUL and unreadable paths, malformed UTF-8 and lone surrogates refuse
with the command's exit 1. The manifest is bounded (1 MiB, 1000 ships) and
duplicate JSON properties are refused. Titles refuse format, line and
paragraph separator characters and edge whitespace. Naming an already
retired key is a recorded no-op, so a rerun converges.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
UseConceptTallyCors replaces UseCors and appends Vary: Origin to every
response under the export route, preflights and 404s included, keeping any
Vary already set. A single-origin ASP.NET policy never adds it itself.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GET /public/concepts/{collection}/tallies returns per-live-subject up, down and
member counts only, anonymous, Cache-Control public max-age=60. CORS policy
ConceptTallies is built from ConceptVotes:AllowedOrigins, GET only, no
credentials, required on this route alone. Production startup refuses an
allowed origin that is not an exact https origin.
Not yet built or tested when committed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ConceptSubjectRegistry is the only writer of ConceptSubject. ConceptKey is the
one key grammar (anchored with \z), shared with ConceptAppraisalService.
Not yet built or tested at this commit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The snapshot is regenerated; EF also reordered two pre-existing properties
(BridgeAction.TargetSurface, one ProjectId) into its canonical order.
Migration discovery now lists thirteen ids.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ConceptAppraisalService is the only appraisal writer; tallies are derived from
rows. VoterIdentity.For decides voter key and class; NetworkPrefix is the only
prefix computation (HMAC tag). NetworkTagSecret is required in production.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
UseForwardedHeaders runs first with KnownProxies exactly loopback and
ForwardLimit 1; Data Protection always uses application name "Bifrost";
the startup validator refuses production without DataProtection:KeysPath.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A failed entry stamped ahead of the clock (host clock stepped forward then
corrected) is eligible now instead of waiting out the skew; the next attempt
rewrites the stamp. The WATCHDOG_NOTICE_NOW_MS env var is deleted: the reader
reads Date.now(), and tests fix it with a --import preload (tests/fake-clock.mjs).
The unknown-entry tests now rerun far past the backoff so a retried unknown fails.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The finally-sends test expected six attempts on all three journal rows, but the
closure entry is first tried in the run that completes its opening, so it has
one. Test-only fix; the previous commit's test failed on this line.
The previous commit (69e12e3867) builds; its
suite has one failing test, fixed here.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Ruling notice-retry-policy. The five-attempt cap and its terminal
"exhausted" state are gone. A failed entry is eligible again once
min(2^(attempts-1), 60) minutes have passed since its last journal
update; a run inside the delay skips it without spawning the bridge.
Every attempt keeps the same nonce. The exit status is non-zero while
any entry is unknown or has failed 3 or more times, and retries go on.
status shows nextEligibleAt for failed entries.
The clock is one now() that WATCHDOG_NOTICE_NOW_MS can override, like
the bridge timeout override, so tests drive the backoff without sleeping.
The process summary field "exhausted" is renamed "failing". A journal
entry whose updatedAt is not a date is now refused as
journal-entry-invalid, since the delay is measured from it.
All commits build.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Tests only, plus wording. No change to what the reader or bridge does.
- A spawn timeout and a spawn error (ENOBUFS from a bridge that floods
stdout) record unknown, are never re-posted, and block the closure.
WATCHDOG_NOTICE_BRIDGE_TIMEOUT_MS overrides the 60 s timeout so the test
reaches that branch; the default is unchanged.
- A 500/502/503/429 answer to the message POST exits 75 from the bridge, and
through the reader is retried with the same nonce until it completes.
- allowed_mentions {parse: []} is asserted in the posted body; status exits
non-zero while an entry is unknown and zero when none is.
- The bridge comment and docs/watchdog-notice.md no longer call a non-2xx
answer "definite": it is retried with the same nonce, relying on
enforce_nonce inside Discord's window, and may double-send if Discord acted
on a 5xx outside it.
Builds. Two tests fail on Windows (SIGKILL status), as before; they pass on
Yggdrasil.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Fix batch for Soul pass s1 (spec r3):
- Delete prune-on-absence and its test. Absence from Idunn's store no longer
deletes a journal entry, so an empty or restored store never re-sends.
- discord-dm exits 75 only when no message can have been sent (arguments,
token, DM channel open, non-2xx on the message POST). The reader records
failed for 75 only; any other non-success without a message id is unknown
and is never re-posted.
- Subject grammar is Idunn's require_id; the notice puts it in backticks.
- Times are bounded to the Date range; notice text is built before the
running entry is written.
Builds and passes on Linux. On Windows the two tests that SIGKILL the reader
cannot pass (process.kill does not deliver a signal there).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The egress test searched for the substring CANARY, which the recipient double
also carries, so it failed on its own input.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
tools/watchdog-notice.mjs is a one-shot reader. It reads Idunn's incident
store unlocked with SingleFileMessagePackBackingStore.pullAll, sends one opening
and one closing DM per incident through bifrost-bridge.mjs, and keeps a
bifrost.watchdog_notice_execution.v1 journal: a running entry becomes unknown
and is never re-posted, a failed one is retried with the same nonce up to five
times, terminal entries are pruned once Idunn retires the incident. The exit
status is the delivery-failure signal.
discord-dm accepts --nonce (at most 25 chars) and sends nonce with
enforce_nonce so a retry inside Discord's window returns the original message.
The tests run the CLI against the store bytes Idunn wrote (fixture from
GameCult/Idunn 7f7528b3) with a counting bridge double, and once through the
real bridge with fetch stubbed. docs/watchdog-notice.md states the CultLib
>= 8cb3b728 floor.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>