fix(windows-ci): land the win-ci fixes (kibitzer ticket sharing, mailbox journal, deny ordering, applyEdit, parent wake, downloader probe, git bound, test removals) - #8961
Merged
Conversation
github-actions
Bot
added
utils
code-yeongyu
added
ci:full-matrix
code-yeongyu
force-pushed
the
ci/win-ci-integration
branch
from
September 27, 2026 14:33
41cfb4c to
cb725a9
Compare
github-actions
Bot
added
memory-core
code-yeongyu
force-pushed
the
ci/win-ci-integration
branch
2 times, most recently
from
September 27, 2026 17:03
996be76 to
3ac3306
Compare
github-actions
Bot
added
the
isolation-core
code-yeongyu
force-pushed
the
ci/win-ci-integration
branch
2 times, most recently
from
September 27, 2026 17:27
f53771d to
0b4c2f1
Compare
…t command (#8950) The confirm auto-deny test asserted < 50 ms of wall clock across a socket, which a loaded Windows runner exceeded (55.2 ms) while the deny itself was correct. It now proves the real contract: once the client has ingested the UI request, the deny is on the wire before the client's next command.
…ocess (#8979) The test cold-started one Bun process per platform under a 10 s abort; a loaded Windows runner exceeded it on one of the six (exit 143). The downloader reads process.platform/arch per call, so one probe process now covers every target and each target keeps its own assertions.
…romptAsync hold (#8951) A parent wake that met its own promptAsync post-dispatch hold always retried after a fixed 2 s, the whole hold length. The hold expires lazily on Date.now(), so a retry timer that fired a tick early (coarse Windows timers) still saw the hold and waited a second full hold. The gate's reserved result now carries the hold's expiresAt. The first meeting with a hold keeps the full 2 s back-off; meeting the same hold again (same expiresAt) waits only what is left of it.
… bounded git reads
…out instead of a real stalled git (#8997) The stand-in stalled process left a Windows grandchild holding the fixture directory past teardown (EBUSY, seen in integration run 36336999917 attempt 2). The deadline and tree teardown of a real stalled git stay covered in command.test.ts; this test now proves only the retry contract.
code-yeongyu
force-pushed
the
ci/win-ci-integration
branch
from
September 27, 2026 18:19
0b4c2f1 to
c13d35e
Compare
…i/win-ci-integration
…n' into ci/win-ci-integration # Conflicts: # packages/omo-senpi/plugin/extensions/omo-task.js
…' into ci/win-ci-integration
…' into ci/win-ci-integration # Conflicts: # packages/omo-senpi/plugin/extensions/omo-task.js
…wn' into ci/win-ci-integration
code-yeongyu
force-pushed
the
ci/win-ci-integration
branch
from
September 27, 2026 23:10
0dd4d46 to
3586a3e
Compare
code-yeongyu
changed the title
code-yeongyu
marked this pull request as ready for review
September 27, 2026 23:11
This was referenced Sep 27, 2026
This was referenced Sep 28, 2026
timothybrush
pushed a commit
to timothybrush/oh-my-openagent
that referenced
this pull request
Sep 28, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This lands the remaining Windows-CI fixes from the win-ci effort as ONE merge, instead of cycling eight PRs through a moving generated bundle. Each fix keeps its own commits; its original PR has the full write-up, RED/GREEN evidence and focused Windows soaks, and is closed as superseded by this one. The generated omo-senpi bundles are regenerated once on linux/amd64 over the merged sources.
Already merged separately: #8971 (lane A, #8932 foreign-owner liveness over the named pipe, #8942) and #9033 (#9032, the #8877 Windows stdin-pipe fake).
QA & Evidence
windows-flake-soak.ymlruns on windows-latest (at least 10 iterations each), recorded in its superseded PR.test (windows-latest, 1/2),test (windows-latest, 2/2)andsenpi-compatibility (windows-latest)actually executed each time. Attempt 2 failed only on the team-core recycled-pid lock flake, since fixed on dev by fix(team-core): acquire team locks without fsync and reclaim stale ones exactly once #9034 (Windows CI: team-core withLock can wait on a delete-pending stale lock after reaping a recycled-pid owner (2s test budget) #9030)..skipor.only, deletes no test, and widens no timeout. The wall-clock assertions it touches are replaced by event or ordering proofs.Fixes #8953
Fixes #8948
Fixes #8950
Fixes #8954
Fixes #8951
Fixes #8979
Fixes #8997
Fixes #9025
Refs #8324, #8250, #8942