pact.MCPGrant gains BackendUserID, the backend account the consent page
showed. pier gives an OAuth identity a backend subject (on resolution and
after a link event) only while the user's link names that account; a link
to anyone else, or no recorded account, gives none, so a link changed after
the consent never re-targets the grant.
The custom-element-in-form rule that keeps Esc, / and ? out of a plugin
field type's closed shadow root also applies to a focusable widget that
is not an input. The header comment and isEditableTarget now state the
trade-off and why the rule is not narrowed to plugin field types.
The README said a revoked personal API token fires the backend subject
event. admin:token:revoke runs in its own process, where no MCP endpoint
listens, so it fires none; its token's sessions end at the next
verification. Say so in the events entry and the command table.
An action whose Applies failed was dropped with no signal, so a section
whose Status succeeded showed a normal card with a button silently
missing. The section is now also sent with error: true, and My account
shows its error line. The error field's description, the cabana README
and the My account docs say so; the OpenAPI document and the SPA schema
types are regenerated.
Browsers strip tab and newline from URLs, so "/\t/host" passed the
SPA's same-origin check and resolved off-site. The check now refuses any
ASCII control character first, as Go's url.Parse does on the server, and
an absolute URL must parse with a host, so both sides apply one rule.
validatePluginTokenScope checked the trimmed name with
pact.ValidTokenScopeName while compileTokenScopes registered the
untrimmed one, so " blog:read" booted, was listed under its padded name
and could never be granted, and the duplicate check depended on
registration order. The name is now validated exactly as registered,
which is the shared rule pact.ValidTokenScopeName states.
The save shortcuts blurred document.activeElement before clicking. With
focus on an input inside a shadow root that is the host, and blur() on a
host that is not itself focused does nothing (spec, Firefox), so a plugin
field type committing on the native change event lost the typed value.
The shortcut now focuses the button first, as a pointer press does, which
unfocuses the inner input in open and closed roots, and gives focus back
to the innermost element reached through open shadow roots.
For a store without Reserve (only a test wrapper reaches this; both shipped
stores reserve), serveNew now calls Put right after the Count check and
before the server is built, and a failed connect or initialize deletes the
record again. The session is therefore never in the local map without its
record, so a reaper pass in between cannot close it as expired. The cap of
this fallback stays check-then-act, as its comment now says.
The built-in backend guard's principal answers errScopeUnavailable, not a
nil principal, when an admin personal token's stored scopes all fail to
resolve (a scope's plugin was removed). MCPDispatch passes it through and
dispatch maps it to the fixed tool error "pier: the token's scopes no longer
grant this route" instead of claiming the administrator's account is
unknown or deactivated; the listing check still treats it as a refusal, not
as a failed check.
The stateful GET stream and the stateless subscriptions/listen stream
capture the client address when they open and resolve every periodic
re-check with towel.WithClientIP, as the bearer middleware does, so a
failing re-check (a revoked admin token, for example) is logged with the
remote address. TestStreamRecheckClientIP records the address the verifier
sees on every verification of both streams.
reserveScript reads the record keys of the binding set's members, which it
cannot declare in KEYS, and its declared keys hash to different slots, so
Redis Cluster refuses it and a key-pattern ACL must allow the whole
<prefix>mcp: range. The script's comment, the pier README's
mcp.sessions.store row (which also no longer says the store opens on the
first request) and docs/services/mcp.md say so.
lazyStore's and newStore's doc comments no longer claim that nothing dials
at mount: Mount starts a background warm-up that dials Redis with a backoff
and logs each failed attempt. The comment records the decision to keep the
warm-up for commands that mount pier without serving, such as route:list:
it only dials and logs until the process exits.
demo_pins counts the mismatch lines it accepts and refuses when
check-pins.sh exits non-zero without one, instead of passing on the single
empty line of an empty output. The self-test plants a silent failing
check-pins.sh (pins-silent-failure).
demo_pins records every plugin submodule it accepts ahead of its go.mod pin
by non-test commits (UNRELEASED_PINS, and PHASE197_PINS_FILE for run_all).
The consumers stage then ends with "phase19.7 consumers passed WITH
UNRELEASED PINS: <names> (the release pins are not yet proven)", and --all
ends with a notice plus "Phase 19.7 gate passed WITH UNRELEASED PINS: the
release pins are not yet proven (<names>)" instead of the unqualified
"Phase 19.7 gate passed". The exit status stays 0 so the pre-release
verification can run; only a PHASE197_RELEASED=1 run proves the pins.
The self-test pins that accepted unreleased drift is recorded, test-only
drift is not, and gate_verdict is unqualified only with no recorded pin.
promptHandler logs a failing Render's error with the prompt name and
answers the fixed JSON-RPC internal error "pier: the prompt could not be
rendered", as dispatch does for tool calls (A-WR-08), so database or
provider text never reaches the agent or its transcript.
TestPromptRenderError renders a prompt that fails with a database error
naming its host and asserts, on both paths, that the client sees only the
fixed text and the cause is in the log.
Each stateless request now carries an entry-use record on its context;
getServer counts the entry it hands out as active (under the path lock)
and ServeHTTP releases it when the request ends, so a subscriptions/listen
stream keeps its entry active for its whole life. Capacity eviction walks
the LRU from the back and drops only idle entries, so it no longer ends a
quiet listener or cancels a tool call in flight, and a revocation still
finds every open stream. While every entry is busy the LRU stays above
capacity until requests end.
TestStatelessFanOut's LRU subtest now asserts that the listener's server is
spared when the binding past capacity arrives (the oldest idle server goes
instead) and that a revocation then still ends the stream.
catalog.allowedReport returns the allowed lists together with the first
provider error. Widening (stateful) and the stateless refresh leave the
cached server and its recorded fingerprint untouched on an error, so the
next request retries instead of the tools staying hidden until the next
rotation or event. A fan-out recompute still applies its fail-closed diff
but records no fingerprint, so the binding's next request re-filters.
TestWideningGuardError fails the guard during a widening on both paths and
during a permissions recompute, then asserts that the next request lists
every tool once the provider answers again.
The identity fingerprint no longer includes token ids: guards decide on
scopes, subjects and the ceiling, and an OAuth access token's id changes on
every refresh, so a rotation alone no longer re-filters a cached server.
The stateless path's refresh only adds tools and prompts for a fingerprint
new to the entry (remembered until a fan-out recompute) and never removes:
one server serves every live token of a binding, and removing on each
alternation between a wider and a narrower token pulled a tool from under
the wider token's call. The receiving middleware still filters every list
and call per request; fan-out recomputes still narrow. The stateful path
keeps its exact widen, a session serves one client.
TestFingerprintIgnoresTokenID pins the fingerprint; TestStatelessSharedBinding
blocks the wide token's call after its server was chosen, sends a request
of the narrow token and asserts the call still succeeds (it failed with
"unknown tool" before).
Fan-out matching snapshots the sessions (and stateless entries) under the
router-wide lock, releases it, and reads each identity through a small
identity box that is never held across I/O, so an event no longer waits on
a session that is re-filtering while it holds sr.mu or sp.mu.
Widening and the fan-out recomputes run their guard checks with
guardContext: not cancelled with the triggering request, but bounded by
mcp.call_timeout, so a provider that never answers cannot hold a session
lock forever. README and docs/services/mcp.md describe the wider meaning of
mcp.call_timeout.
TestLockScopeUnderWidening blocks a widening guard check, fires a
permission event for that session and asserts that another session's call
still finishes, and that a never-answering check ends at call_timeout.
B-IN-05 (deferred, the shared confirm primitive), B-IN-06 (no fix
needed), the admin:token:revoke event, the OIDC decision (resolved), the
local go.work retarget, the demo pins, the release order with
PHASE197_RELEASED, the lenient link code parse and one offline-build
flake; the plan 05 test-commit entry is superseded by the release.
T-19.7-42..T-19.7-65 join the register, the eight corrected rows carry
their fix commits, the audit trail records the gap closure; threats_open
stays 0 (every high threat is fixed or verified).
Corrects the verdicts the code review disproved (T-19.7-22 first, then
T-19.7-03, -08, -17, -18, -23, -31, -35 and cross-cutting findings 1 and
6), adds the Method step that a high or critical control is verified only
by a test that fails without it, the revert-checks of the three critical
fixes, and the register rows T-19.7-42..T-19.7-65.
- USER_SHORT_TESTS (TestMemberTokenScopesBoot, GOWORK=off -short) in the
plugins stage; USER_TESTS, OAS_TESTS, JOURNAL_SHORT_TESTS and DEMO_TESTS
gain the gap-closure tests and the plan 08 smokes
- consumers stage: the demo site's vitest run with VUE_FILES
(app/lib/linkCode.test.ts) required, through detect_vitest
- docs stage: cooler's KeyPrefix and keyPrefix carry their own doc
comments (A-IN-02)
- self-test cases for the demo site's required file
A request's state is persisted after its answer is streamed; one of the
instance that served the last request may finish after the expiry and
close that copy itself (the record is gone). The reaper is required to
close the copy of the instance that did not serve it.
A reaper pass that cannot read the store keeps the copy by design, so the
Redis case reaps until the copy is gone and still requires exactly one
close per instance; both copies are asserted before the expiry.
- check-phase19.7.sh: workspace_check (framework to this checkout, the OAuth
server plugin's user plugin to USER_PLUGIN) in the plugins, consumers and
docker stages; consumer_with_user_plugin builds fonoteka.go and sm-bm-app
against USER_PLUGIN through a scratch go.work; demo_pins accepts
unreleased commits its canonical checkout holds and refuses any drift
with PHASE197_RELEASED=1; self-test cases for each, one PASS line each
- TestSessionReaper pins A-CR-01, A-IN-01 and A-IN-06 (memory and Redis
stores, store errors, the GET stream re-check, the ephemeral label)
- TestAuthTime pins auth_time at mint and through every refresh path
- GO_TESTS and GO_TAGGED_TESTS require them with the plan 06/07 smokes
