The five internal/app partitions ran end to end inside one job, so the app
shard's wall clock was the sum of all five: 780s in CI, of which the schema
partition owned 357s. Each partition is now its own matrix shard, so they run
concurrently and the shard's wall clock is set by its slowest partition rather
than by their total. Every partition shard still selects the same single
internal/app package, so the impacted-package query maps the shard name back to
app and the partition only chooses which tests run.
The helper gains a partition argument and a list-partitions mode. APP_PARTITIONS
is the single source of truth for the set, and the discovery pass still runs in
every job, so each one independently verifies that the partition patterns cover
every top-level test exactly once before running the one it was asked for.
Two fail-closed checks guard the split, because the helper's own coverage check
can no longer prove the whole package ran once the partitions are separate jobs:
- The helper cross-checks APP_PARTITIONS against the coverage counters in both
directions, so a counted partition that nothing dispatches and a dispatchable
partition with no counter both fail instead of silently skipping tests.
- TestCIAppRacePartitionMatrixMatchesHelper pins the workflow's app-<partition>
shards to list-partitions output in both directions, so a partition cannot
lose its job while every job stays green.
The discovery loop variable is renamed from partition to spec: it would
otherwise shadow the partition requested on the command line, which run mode
reads after the discovery pass completes.
The workflow contract pinned the focused path by literal: the job name
`Test (changed packages)`, the unsharded
`list "$TEST_BASE_REF" "$TEST_HEAD_REF"` call, and a single
`go test -timeout=15m` line standing in for internal/app's package-level
headroom. Sharding the job changed all three literals, so `Test (workflow
and release contracts)` failed on this branch even though every shard
selection test passed.
Each invariant the contract guarded still holds, so the assertions are
updated to the new shape rather than relaxed:
- the focused job must still exist, now as the matrix job, named the way
the contract already names `Test (race: ${{ matrix.shard }})`;
- package selection must still derive from the authoritative synthetic
merge base/head, now with an explicit shard argument, so pointing it at
any other ref still fails the contract;
- internal/app's headroom is asserted through the process-isolating
helper and the per-shard budgets, mirroring the assertions already
applied to test-race. That is stronger than the old single -timeout: it
pins the mechanism that keeps the suite inside its budget rather than
the number alone. release-scripts membership is asserted too, because
its dedicated job only runs at full-suite or release-sensitive scope,
so losing it here would silently stop testing test/scripts changes.
The shard comparisons in the focused job are quoted so that job reads
verbatim like test-race's.
Ablating the implementation one change at a time turns the contract red
in all five cases: removing the app helper call, dropping release-scripts
from the matrix, selecting from HEAD~1, collapsing the matrix back to a
single unsharded job, and dropping the cli/smoke timeout budget.
The focused path tested every impacted package in a single job with a
plain `go test -race`, so internal/app ran inside one long-lived process
alongside all of its reverse dependencies. That is exactly the shape
scripts/ci/run-app-race-tests.sh exists to avoid: a single app test
process retains every constructed command tree in framework registries,
so the run grows to 900s and the job stays alive long enough to be
reclaimed by the runner. Recent focused runs failed with SIGTERM after
9-10 minutes without a single test failure, and one earlier run failed
at `internal/app 902.651s`, 2.65s past the package timeout.
Fan the same package plan across the shard matrix test-race already
uses, and run each shard the way test-race runs it: internal/app through
the process-isolating helper, cli/smoke with their wider package budget,
release-scripts without race and with archive tooling.
changed-test-packages.sh gains `list-shard`, which intersects the
impacted set with scripts/ci/test-packages.sh shard membership so shard
definitions stay single-sourced — and so an unknown shard name aborts
there rather than reporting an empty selection, which would let a
mistyped shard skip every test while reporting success.
release-scripts is in the matrix on purpose: its dedicated job only runs
at full-suite or release-sensitive scope, so omitting it here would stop
testing test/scripts changes altogether. A test pins that the shard
selections partition the impacted set exactly, so shard-plan drift
cannot silently shrink focused coverage.
The Coverage context was the PR critical path (~17 min end to end):
coverage-current re-ran the whole suite serially (-p 1, ~13 min) and
coverage-baseline re-ran it again at the merge-base (~13 min) although
that profile is a pure function of the base commit.
- coverage-current now owns only the scoped (standard-tier) profile;
full-suite candidate profiles come from a 5-way shard matrix
(app/cli/generators/helpers/remaining) that keeps -p 1 inside each
shard on isolated runners. scripts/ci/test-packages.sh list-coverage
defines the shards and verify proves the union equals the previous
single-run package set exactly once.
- the aggregate Coverage job reassembles the disjoint shard profiles
into coverage.txt before make coverage-gate, failing closed when a
shard file is missing, so gate semantics (100% changed-code +
scope-matched overall non-regression) are byte-compatible.
- coverage-baseline restores the merge-base full-suite profile from an
exact-key cache (merge-base SHA + resolved Go version) written by the
last green main push; any miss falls back to recomputing in the
merge-base worktree. Exact key only - no prefix fallback, a near-miss
profile would compare the candidate against the wrong commit.
- new contract tests pin the shard matrix, the assembly step, the
exact-key cache pair, and the absence of restore-keys; the package
plan test also covers the coverage shard partition.
release_version was interpolated into an awk regex, where '.' matches any
character. Version 1.0.1-beta.1 therefore also admitted
.changes/released/1x0x1-betaX1/, letting the archive drift from the
CHANGELOG version while every other seal assertion still passed and
breaking the documented audit trail.
Compare the archive prefix with index() and split the basename off with
substr(), matching the literal-comparison idiom already used throughout
check-changelog-pr.sh. Only the basename, whose character class is fixed,
stays a pattern.
Git records no diff entry for a directory itself, so adding
.changes/foo/bar.md only surfaced the nested path, which the single-level
trigger regex skipped. The entry validation and the renderer were both
bypassed, letting a nested directory reach main and break every later
fragment render with 'unexpected directory'.
Trigger the top-level tree validation on any .changes change outside
.changes/released/ (which keeps its own immutability and release-seal
checks), and assert .changes itself is still a tree so replacing it with a
blob or symlink cannot empty the child listing unnoticed.
Re-rendering stays keyed on fragment changes so a README-only edit does
not fail on an empty fragment set.
Bind each accepted marker to the exact workflow run attempt, immutable artifact, source comment, and current PR head so a historical successful run cannot authorize a different payload.
The fragment gate only ran validation when the changed path matched the
legal fragment name pattern, so `.changes/Foo.md`, `.changes/notes.txt`
and a symlinked fragment slipped through untouched and then broke the
next PR that added a legal fragment. The trigger now fires on any
top-level `.changes/` change other than README.md and rejects every
entry that is not README.md, released/, or a 100644 blob named
^[a-z0-9][a-z0-9._-]*\.md$.
The renderer had the same hole from the other side: `find -type f`
is false for symlinks, so a symlinked fragment was silently dropped
from the rendered notes, and the `[a-z0-9]*.md` glob only constrained
the first character so `chat reply.md` passed. It now walks every
top-level entry and fails on symlinks, unexpected directories,
non-regular files and illegal names. Both scripts pin LC_ALL=C so the
ASCII ranges cannot match uppercase under a different collation.
Adds regression coverage for illegal names, non-markdown entries,
symlinks and executable modes on both the gate and the renderer.
Address P1 lint finding: structured eval-dispatch comments could be
forged by unauthorized users. Add three-layer consumer-side validation:
1. comment.user.login == 'github-actions[bot]' (platform-enforced identity)
2. comment.performed_via_github_app.slug == 'github-actions' (App signature)
3. payload.run_id verified against actual successful workflow run via API
Also adds:
- eval_poll_validate.py: consumer validation module (in-repo, auditable)
- test_eval_poll_validate.py: unit tests proving forged comments are rejected
- Go security contract test updated to assert run_id and validate reference
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- /eval on one's own PR may omit sha=: the guard auto-pins the
dispatch-time head (commenter == PR author leaves no third-party
swap window); dispatching another author's PR still requires the
explicit reviewed SHA (keeps the P1-2 TOCTOU remedy where the
threat lives)
- cases= is now validated structurally per git check-ref-format
semantics (leading/trailing//double slashes, '..', dot-leading
components, .lock suffixes) and rejects '-'-leading values to
prevent git fetch option injection (review P2)
chichuan