[ydflow]

Summary

Add the rejected-push regression requested in the #1324 review, on top of the merged #1323 fix. The existing SessionStart case covers a slow successful push; this case verifies that a delayed rejection retains unacknowledged usage and session snapshots, releases both report and sync locks, and later publishes exactly once when the remote accepts writes.

Tests only: one E2E file, using its existing real-CLI SessionStart helper. The local bare remote delays its first rejection by seven seconds, exceeding pull's five-second wait; subsequent writes reject immediately so the case does not spend seven seconds on every report category. All original receiver modes keep their behavior.

Type of Change

• Tests only

Test Plan

• npm run build
• npx tsc --noEmit
• npm run lint
• git diff --check
• Built-CLI E2E, with retries disabled:

npx vitest run --config vitest.e2e.config.ts src/__tests__/e2e/report-timeout.test.ts -t 'session-start|retains a rejected report' --maxWorkers=1 --retry=0

3 passed, 18 not selected: the existing direct-pull rejection/retry case, existing slow SessionStart success case, and new slow SessionStart rejection/retry case. The new case completed in about 26 seconds locally under the unchanged 60-second test deadline.

Before narrowing the fixture delay to the first rejection, the broader selection including git/claude slow direct-pull also passed (4 cases). That first version delayed every rejected report write and took about 54 seconds for the new case; the final fixture preserves the same data/lock/retry assertions with less idle waiting.

Windows / Node 24, Claude + generic Git provider, real CLI and Git subprocesses against an isolated local bare remote. No remote credentials or model calls. The full provider/agent matrix and complete unit suite were not run locally for this test-only change.

Related Issues

Follow-up to #1323 and #1322; extracts the failure-path test from #1324. The runtime fix is already in main through #1323. This PR adds regression coverage for its rejected-push path.

[github-actions]

No findings in the specified diff.

This is a tests-only change, so no end-to-end record is required under the review rules. The PR description nevertheless documents a representative real-CLI run with retries disabled: 3 passed, including the new rejection/retry case. The documented testing is sufficient.

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

[jeff-r2026]

LGTM. Reviewed the diff; merged locally onto main together with #1289/#1299/#1348/#1342, then ran tsc, build, the affected unit tests and e2e (redact, dashboard-collector, ai-client-openclaw, report-timeout, openclaw-ai-import): all passed.