[quality] the bundle e2e suite never drives ok-to-test on pull_request or in the sweep, nor the sweep's listing pagination and failure arms

[hivecommons-hive]

Finding

The bundle e2e suite (__tests__/bundle/bundle.test.ts, spawning the committed dist/index.js against fakeGithub.ts) never exercises the two non-comment entry points of the ok-to-test trust marker, nor the listing arms of the sweep cron job:

• okToTestOnPullRequest (src/issueComment/trigger.ts:173) — the pull_request handler registered in pullRequestHandlers (src/pullReq/handlePullReq.ts:18) that approves the runs waiting on a new head on synchronize/reopened. The only bundle routes for GET …/actions/runs and POST …/actions/runs/<id>/approve are in the issue_comment /retest and /ok-to-test cases (bundle.test.ts:743-800); no pull_request case routes them.
• approveIfTrusted (src/cronJobs/sweep.ts:140) — the ok-to-test step of sweepPullRequest, including its 403 (cannot approve workflow runs: grant \actions: write` to the workflow`) and listing-failure arms being collected per pull request.
• recentlyUpdatedPulls (src/cronJobs/sweep.ts:172) — the page-2 read when a full page is still inside the window, the stale-tail stop, the zero-candidate run, and the could not list the open pull requests failure. The existing schedule sweep job block (bundle.test.ts:1816) answers only page=1.

Evidence

• Unit: npx vitest run --coverage on main @ c48bd6d (Node v26.10.0, @vitest/coverage-v8 5.0.1) → src/cronJobs/sweep.ts and src/issueComment/trigger.ts 100% lines; All files 99.82% lines. These paths are unit-covered (__tests__/cronJobTest/sweep.test.ts, __tests__/issueCommentTest/trigger.test.ts).
• End-to-end: the bundle suite's coverage of src/ is not machine-captured ([quality] the bundle e2e suite's coverage of src/ is never captured — runBundle.ts drops NODE_V8_COVERAGE and dist/ has no source map #235, fix in test(bundle): opt the node --check spawn out of NODE_V8_COVERAGE so the suite survives coverage.autoAttachSubprocess #236), so this is established by inspecting the suite's routed requests directly: grep -n 'actions/runs' __tests__/bundle/bundle.test.ts matches only lines 747-797 (the comment commands), and the sweep block's routeList serves page === '1' ? prs : []. Not an inference from the unit coverprofile.
• Limitation: unit and e2e evidence cannot be merged until test(bundle): opt the node --check spawn out of NODE_V8_COVERAGE so the suite survives coverage.autoAttachSubprocess #236 lands; analyzed separately.

Recommendation

Add a new bundle file __tests__/bundle/sweepOkToTest.test.ts driving dist/index.js through:

• schedule jobs: sweep — fork pr carrying ok-to-test with a pending run (approve + merge path), with nothing pending, without the label (no runs read), approve 403 collected as the step's failure while tide still runs, a 500 on one pr's runs listing not stopping the next, zero candidates, a full first page followed by page 2 with a stale tail, and a refused listing failing the run before any pr read
• pull_request synchronize / reopened with ok-to-test approving run on the head (tide skipping synchronize; reopened also running require_matching_label and tide), opened and label-less synchronize reading no runs, approve 403 failing the run

Disjoint from every open bundle PR: #238 (/cc …/reopen), #240 (/test, lgtm cancel), #242 (cron jobs input arms), #244 (/meow), #246 (fixed/prefixed labels), #248 (label-sync inputs), #250 (/assign, /milestone), #252 (dispatcher arms, /hold aliases). Disjoint from the unit-level #225 (trigger/tide residual fallbacks) and #219 (evaluateMerge).

Priority

• Impact: medium — the sweep is the only write path for fork pull requests; its ok-to-test step and listing pagination are what a busy repository hits first, and neither is ever run end to end
• Effort: low

---

Filed by quality agent (hold-gated mode)

---

🐝 Hive Agent: quality | Instance: hosted-available-lke648397-260827-5q9t | SHA: c48bd6d

— hive: agent=quality backend=copilot model=claude-fable-5.1 copilot=1.0.88

[github-actions]

Please add a kind label with /kind failing-test or /kind cleanup.

[hivecommons-hive]

task-list sweep: 0 of 2 items ticked. Not closing yet — outstanding boxes remain.

Outstanding items:

• 🔲 schedule jobs: sweep — fork pr carrying ok-to-test with a pending run (approve + merge path), with nothing pending, without the label (no runs read), approve 403 collected as the step's failure while tide still runs, a 500 on one pr's runs listing not stopping the next, zero candidates, a full first page followed by page 2 with a stale tail, and a refused listing failing the run before any pr read
• 🔲 pull_request synchronize / reopened with ok-to-test approving run on the head (tide skipping synchronize; reopened also running require_matching_label and tide), opened and label-less synchronize reading no runs, approve 403 failing the run

Merged PRs referencing this issue so far:

• test(bundle): drive ok-to-test on pull_request synchronize and in the sweep through dist/index.js #260 — test(bundle): drive ok-to-test on pull_request synchronize and in the sweep through dist/index.js

This comment is edited in place by the task-list sweep on every cycle; it is not duplicated.

[hivecommons-hive]

Verified on main@ec76a3b: #260 landed the two wiring cases the maintainer kept — synchronize of a pull request carrying ok-to-test approving the runs waiting on its head (bundle.test.ts L1025) and the sweep approving a fork pr's pending runs before tide reads it (L1990). The pagination/failure arms were cut on review as duplicates of sweep.test.ts and handlePullReq.test.ts; closing as resolved at the scope the maintainer accepted.

— hive: agent=quality backend=copilot model=claude-fable-5.1 copilot=1.0.88