[MertensMing]

Summary

Follow-up to the review of #1276 (already merged), for #1233.

A successful reports push in a single-branch or shallow clone can leave origin/teamai-reports stale. After local deltas are cleared, contribute and recall index rebuilds then miss the delivered votes. Record the accepted branch commit in the tracking ref after a successful push, using compare-and-swap to preserve concurrent updates. The index still reads a confirmed commit, never an unpushed worktree HEAD.

Also address the remaining review notes:

• Preserve unchanged same-member v1 entries' legacy scores when applying another document's local deltas.
• Keep vote filenames in parse-error diagnostics and remove redundant scoped votesDir arguments/probes.
• Capture the confirmed ref and local deltas under the votes lock, then release it before reading member blobs.
• Make the extensionless test hook work under both CommonJS and ESM package boundaries.

Type of Change

• Bug fix

Test Plan

Final head 018ac61 on main e45baeb, Node 20.19.5:

• npm run build
• npx tsc --noEmit
• npm run lint
• git diff --check
• 153 relevant unit/integration tests passed across 10 files.
• 38 real-CLI tests passed with retries disabled:

npx vitest run --config vitest.e2e.config.ts src/__tests__/e2e/index-local-votes-1233.test.ts src/__tests__/e2e/votes-sync-retry-1027.test.ts src/__tests__/e2e/self-mode-worktrees-808.test.ts src/__tests__/e2e/reports-writer-sync-561.test.ts --retry=0

The real-CLI regression uploads feedback to an isolated local Git origin, then rebuilds through contribute and through recall with a missing index. Full, single-branch and depth-1 clones now all retain 1.3 votes; before this follow-up the latter two fell to 0.3 until pull fetched again. Original rejected-push and pushed-but-unconfirmed cases still pass. The unconfirmed case now runs its hook beneath a type:module package. The hook was also checked independently in both module modes.

Unit regressions cover unchanged legacy scores, corrupt filenames with and without local deltas, and a writer adding feedback during blob reads: it takes the released lock successfully, the current index retains its coherent snapshot, and the next rebuild includes the new feedback.

Before rebasing this follow-up onto e45baeb, the full unit suite completed with 11,490 passed, 25 skipped, and one failure: project-hook-rules.test.ts, ZCode registered-entry check. That failure was independently reproduced from an unmodified archive of upstream e45baeb on this machine. The full suite was not repeated after the rebase; the final-head checks above were. Other provider/agent environments remain for CI.

Related Issues

Follow-up to #1233 and the #1276 review.

Notes for Reviewers

Tracking-ref maintenance is in the shared side-branch push helper so it records a confirmed delivery before the existing onDelivered callbacks clear vote deltas. Failed pushes do not update the ref. Retry and delivery-ID accounting remain unchanged. No added network fetch or changes to the upload protocol.

[github-actions]

No confident findings in the specified PR diff.

The PR description satisfies the testing requirement: it records final-head build, type-check, lint, and 38 real-CLI tests, including narrow-clone regressions. Deferring other provider/agent environments to CI is acceptable.

Review was read-only; I did not run, build, or install anything from the PR.

[SaulMoro]

This was generated by AI during triage.

@MertensMing Approving with notes. We built this head and main, gave feedback, delivered it with pull, then rebuilt the index without a fetch, and checked the recall verdict.

Clone	main	this head
full	RELEVANT score=8.6	RELEVANT score=8.6
--single-branch	NOT_RELEVANT score=2.5	RELEVANT score=8.6
--depth	NOT_RELEVANT score=2.5	RELEVANT score=8.6

A push the remote rejects does not move the tracking ref. The new narrow-clone e2e cases fail on main and pass here. Reading the ref and the local deltas in one lock hold, then reading blobs from that pinned commit, neither double-counts nor misses a delta.

[P2 non-blocking] src/utils/branch-worktree.ts:654-655 swallows every update-ref error. With a leftover refs/remotes/origin/teamai-reports.lock, the push succeeds, update-ref exits 128 with cannot lock ref, nothing is printed, and a rebuild without fetch drops the vote again: 1.3 before, 0.3 after. main loses it too, so this is not a regression. Also, whenever the clone's fetch refspec covers the reports branch (full clones, and a single-branch clone of that branch), git push already moves the ref, so the compare-and-swap always fails there and the comment's "another updater" reason is not what happened. Re-reading the ref after the push, calling update-ref only when it differs from sent, and logging a warning on failure would cover both.

[P3 nit] indexableVotesDir now runs inside aggregateVotes, which buildIndex wraps. An unexpected checkOwner failure therefore no longer aborts the rebuild: it continues without votes and logs a warning. That fits how vote reads already degrade, and it's worth one line in the description.

[P3 nit] src/viz.ts:344 is a no-op change, because aggregateVotes ignores votesDir when voteScope is set. src/votes.ts:584 re-parses content that was already validated and casts it with as UserVotes.

#1280 also edits src/utils/branch-worktree.ts in adjacent hunks. A trial merge of the two heads is clean and typechecks. This PR owns the tracking-ref write in the push loop, readIndexVoteFiles and the votesDir removals.

[github-actions]

No confident findings in the specified PR diff. The earlier review raised no findings, so there are none to mark resolved.

The PR description satisfies the testing requirement with 38 reported real-CLI tests, including narrow-clone regressions. The record names 018ac61, rather than the reviewed head 67fef9d; under the review rules, this is a note, not a blocking finding. Deferring other provider/agent environments to CI is acceptable.

Review was read-only and limited to the specified diff. I did not run, build, or install anything from the PR.

[SaulMoro]

This was generated by AI during triage.

@MertensMing CI is red on 67fef9dd. Every Lint & Test job and the Windows job fail on windows-test-list.test.ts and codex-team-hooks.test.ts. Those failures came from main, not from this PR, and #1287 fixed them in c15fc0db. The Node 24 macOS job also failed a timing check in env-loader.test.ts (expected 104.33 to be less than 100), which this PR does not touch.

Please rebase or merge current main so CI runs again. The code review from the earlier approval still holds, and we'll approve again once CI is green. Note that #1285 adds a test at the end of the same describe block in search-index-local-votes.test.ts. Whichever PR lands second keeps both tests.

[SaulMoro]

This was generated by AI during triage.

@MertensMing Approving again so this can land. The red CI on 67fef9dd came from main's Windows test list, which #1287 fixed. We merged this branch onto current main (c15fc0db) locally: tsc and lint pass, and so do the vote, confidence and Windows-list tests. No rebase is needed on your side.