Skip to content

chore(release): pending release v0.98.3 - #1371

Merged
marcusrbrown merged 9 commits into
releasefrom
next
Aug 11, 2026
Merged

marcusrbrown merged 9 commits into
releasefrom
next

Conversation

@fro-bot

@fro-bot fro-bot Bot commented Aug 10, 2026 •

Copy link
Copy Markdown
Contributor

Pending Release: v0.98.3

This PR tracks changes pending release. Released on the next auto-release cycle (Sunday/Wednesday) or via manual dispatch.

Merge this PR to trigger a release. Releases also run automatically on Sunday/Wednesday at 20:00 UTC, or via manual workflow dispatch.

Commits Since Last Release


Auto-generated by the release pipeline. Updated: 2026-08-11 19:43 UTC

marcusrbrown and others added 5 commits August 9, 2026 19:12
The upstream PR workflow reports what a change introduces by checking
out GITHUB_BASE_REF and diffing against it, and it links results using
the pull request number. A merge queue supplies neither, so the diff
resolved against an empty ref and the printed link carried no number.

Route merge_group to the full scan, which needs no pull request context.
Queue entries now get a complete result instead of a comparison against
nothing.
* ci: fail pull requests that introduce new vulnerabilities

The scanner shipped report-only because two dozen findings would have
failed every pull request on the first run. Those are fixed, so the
pull request path can start enforcing.

Only that path. The upstream PR workflow scans the base ref and the head
ref and reports the difference, so failure there means the change itself
introduced something. The full scan has no baseline to compare against
and runs on pushes, the merge queue, and a weekly schedule; failing it
would break unrelated work whenever a new advisory is published against
a dependency nobody touched. Those findings continue to go to code
scanning.

* test(scripts): guard the scanner's routing and failure policy

Both invariants are one edited line away from breaking, and a wrong
value still produces a green run: pointing merge_group at the diff
workflow compares against an empty base ref, and failing the baseline-free
full scan would break unrelated pushes whenever a new advisory lands.

Each assertion was checked by breaking the workflow and confirming the
matching test goes red.
…#1369)

The module field was normalized once before and drifted back, because the
schema requires it without constraining it. Thirty-four values had
accumulated across ninety documents and twenty-one were used exactly
once, including raw file paths that named a directory rather than a
subsystem.

Map every document onto seventeen values and add a test that rejects an
unknown value, a missing field, or a canonical value that no document
uses. The last case catches a set that has drifted from the codebase
rather than a document that has drifted from the set.

Each assertion was checked by breaking a document and confirming the
failure names it.
@actions/cache catches reservation and upload failures internally and
returns -1 rather than throwing, so the catch block never ran and every
save reported success. A run whose write was denied logged 'Cache saved'
25ms after the platform refused it, then set the state flag that tells
the post-action there is nothing left to do.

Branch on the sentinel at all three call sites. The library cannot
distinguish a policy denial from a benign key collision once it has
swallowed the error, so a collision is now reported as a failure too;
the alternative is continuing to report real failures as success.

Restores the post-action retry, which the false success had been
suppressing.
@fro-bot
fro-bot Bot requested review from fro-bot and marcusrbrown as code owners August 10, 2026 05:40
@fro-bot fro-bot mentioned this pull request Aug 10, 2026
86 tasks
Co-authored-by: fro-bot[bot] <109017866+fro-bot[bot]@users.noreply.github.com>
* fix(cache): let durable storage win over a stale cache

Mention-triggered runs cannot write the Actions cache, so they restored
an older run's cache through prefix fallback and started cold every
time. Their object-store upload had succeeded, but the restore never
looked because any cache hit returned first.

Consult the object store before the cache. A hit requires the main
session database; sidecar files alone stay a miss, as before. Failures
and a disabled store fall through to the cache path unchanged, so this
is inert until durable storage is configured.

* fix(cache): validate storage restored from the object store

A database restored from durable storage went straight into use without
the corruption check the cache path has always run. Check it, and on
failure clean up and fall through to the cache, which may still hold a
usable older copy.

No version check on that path: the marker is written locally and never
synced, so checkStorageVersion treats its absence as compatible and the
call would do nothing. auth.json likewise never enters the store payload,
which carries only the session database files.

Drops a branch in restoreAfterCorruption that became unreachable once the
object store moved ahead of the cache.
marcusrbrown and others added 2 commits August 11, 2026 12:40
…1374)

A platform policy change made Actions cache tokens read-only for
issue_comment and issues runs. The cache client authenticates with
ACTIONS_RUNTIME_TOKEN rather than GITHUB_TOKEN, so no permissions
change could have fixed it and the declared permissions block was
actively misleading.

An unchecked -1 return from saveCache reported the denied write as
"Cache saved" 25ms later, and that false success set the state flag
that suppresses the post-action retry built for this exact case.
Session continuity was lost on the primary human-facing flows for
over a month while logs reported success.

Records the diagnostic that located it: two subsystems disagreeing
under byte-identical printed permissions. Separates what our logs
directly evidence from what is attributed to the platform change.
@marcusrbrown
marcusrbrown merged commit 49eadde into release Aug 11, 2026
1 check passed
@marcusrbrown
marcusrbrown deleted the next branch August 11, 2026 19:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants