Conversation
- keep one Linear incident per workflow and report failed jobs - email first failures and recoveries from trusted main runs
- run Linux and macOS browser projects against isolated test services - keep Safari OPFS available with per-test persistent profiles - sanitize failure reports before artifact upload
- isolate one credential-free diagnostic from Nightly alerts - record safe readiness and first-query categories for the runner failure
- provide the gitignored browser worker asset before the isolated startup case
- keep the one startup case while starting the original server set
Semgrep Security ScanNo security issues found. |
|
Preview environment destroyed 🧹 Stack |
PR Metrics
Updated Fri, 25 Sep 2026 15:27:19 GMT · run #3129 |
|
There was a problem hiding this comment.
🔭 thunder-deep-review (advisory)
Complements the other bots — surfaces only what they did not flag. Never approves, never requests changes, never gates merge.
head: c2383316b558 · mode: single · deferred 0 item(s) already reported by other bots (best-effort dedup)
- Keep the test matrix accurate as the suite grows - Align local startup checks with CI readiness
- Keep diagnostic branching out of the initialization sequence - Preserve finite labels and background readiness observation
There was a problem hiding this comment.
🔭 thunder-deep-review (advisory)
Complements the other bots — surfaces only what they did not flag. Never approves, never requests changes, never gates merge.
head: 7cd119b6b37c · mode: single · deferred 0 item(s) already reported by other bots (best-effort dedup)
Additional notes (couldn't anchor to a diff line)
.github/workflows/nightly.yml:191— 🚫 Blocking — Dropping thenotifyjob removes all failure alerting from Nightlyvite.config.ts:191— 📐 Convention —forwardConsoledoesn't appear to be a real Vite server optione2e/db-diagnostic.ts— 📐 Convention — Label grammar is defined twice with nothing keeping the two copies in sync
…resh Main's THU-884 nightly fixes (#1302, #1303) rewrote and deleted parts of the testing doc our branch had moved to docs/internals/. Mirrored main's version at the new path rather than preserving the older port: - keep the deletion of the old docs/development/testing.md path - take main's note on the separate `bun test --cwd=e2e` step, its shorter Nightly E2E section, and its nightly-WebKit OPFS wording - drop the "CI failure and recovery alerts" section, which main removed now that alerts are paused - name e2e/db-diagnostic.test.ts in AGENTS.md's test-scope sentence, which main's new e2e step made stale
Summary
Reason
The original WebKit first-query rejection has not been reproduced. The installed PowerSync query path already awaits readiness, so this PR collects evidence without a speculative database fix. Uploaded diagnostic JSON contains only allowlisted outcomes, categories, error names, test status, and retry number; it contains no raw error text, stack, URL, or credentials.
Validation
bun run testpassed; the second phase included the two new parser tests (318 pass, 0 fail).bun run type-check,bun run e2e:check-collected(22 specs),actionlint, Prettier, the global Oxlint hook, andgit diff --checkpassed.readylabels on failure and retry; it timed out later at the sign-in Continue click. No full CI Nightly run was dispatched from this PR.Follow-up
Merge this PR to affect scheduled runs. After reviewing the diagnostic artifacts, restore the Nightly notifier and its heartbeat dependency manually, and remove the temporary instrumentation. There is no automatic cutoff in code.