I

status: complete size: medium branch: fix/restore-ai-linter-rules

Restore GitHub AI lint reviews

Status: Complete. Latest main is merged, iterate[bot] proved the rule end to end, and the temporary probe is removed. Historical events replay with the new suggestion policy, production policy v8 is live, and the full local and GitHub test/static gates are green.

Goal#

Restore iterate[bot] reviews for this repository and add an AI lint rule that rejects metaphorical uses of lane, door, and seam unless a nearby comment explains why the word is literal or justified.

Assumptions#

  • Root rules/**/*.md remains the canonical rule source for this repository.
  • The rule should target identifiers and prose in code, not genuine traffic lanes, physical doors, or actual joined/material seams.
  • A short comment immediately above an intentional non-literal use is the explicit escape hatch.
  • Metaphor diagnostics explain the unclear concept but do not propose a rename or publish a GitHub suggested-change patch.
  • The task is complete only when a deliberately offending variable in this PR causes a visible iterate[bot] review comment.

Checklist#

  • Reproduce the missing AI lint run from a fresh PR and capture the failed stage. Production had no /repos/iterate route after the August 8 recreation; restoring it then exposed the obsolete SDK's exact durable processor error.
  • Repair the stale rule-loading/configuration path without hiding failures. Restored /repos/iterate; config commit f059bea pins the current subscription API, recreates the source link after future restores, and produced project/worker-updated plus an active review-bot subscription.
  • Add the banned-metaphor rule to the canonical rules and hosted linter config. Added terminology/no-metaphorical-lane-door-seam to root policy, the default template, and production config policy v7.
  • Add one obvious offending variable solely as an end-to-end review probe. configs/default/worker.ts temporarily named the rule-path array reviewLane without an exception.
  • Prove iterate[bot] comments on that variable, then preserve the proof in the PR. Review 5016262887 and inline comments 3850695732/3850695742 flag reviewLane under the new rule.
  • Prove the metaphor diagnostic is explanation-only, even if the model emits a fix. Policy-v7 review 5029789862 and comments 3862225098/3862225102 explain the unclear role without a replacement or suggestion fence; publisher regression coverage also strips a model-supplied patch.
  • Preserve replay compatibility for events written before suggestion policy existed. The event schema defaults missing suggestions to allowed; a processor regression test replays the historical payload shape.
  • Remove the deliberate offense after the live proof. Renamed reviewLane to githubAiLinterRulePaths and refreshed the generated config template.
  • Run focused tests and checks for every changed code/config surface. After merging latest main, the full tests, typecheck, lint, Knip, format check, template typecheck, and focused linter/config tests all pass.
  • Move this task to tasks/complete/ once the live proof and CI are green. Completed after production policy v8 was ingested and the full test/static gate passed locally and in GitHub.

Implementation log#

  • 2026-08-25: Started from origin/main in a dedicated worktree. Initial inspection found that configs/default/worker.ts still configures the deleted rules/structure/no-small-single-use-helper.md, while the canonical root rule was replaced by four newer structure rules in PR #2305.
  • 2026-08-25: GitHub history showed the last iterate[bot] review on August 6. Production project prj_f80006b6f56e466592117d896d0eaec0, recreated on August 8, contained only /repos/config; the linter router therefore had no route for repository ID 1057144703 (iterate/iterate).
  • 2026-08-25: The GitHub installation remained connected and could access the repository, so no integration reinstall was needed. Recreated and linked /repos/iterate at commit d40cf451; the link event then durably failed with Cannot read properties of undefined (reading 'app-review-bot#review-bot').
  • 2026-08-25: The production config package was pinned to pre-redesign commit bdfb9a; OS had moved to named subscriptions in ffe683 on August 6. Pushed config commit f059bea with compatible package commit bde72e, policy v6, the new rule, and project-specific repo reconciliation. Production imported, built, and served that exact commit and installed active review-bot.
  • 2026-08-25: The compatible SDK reveals that the config repo's unimported voice-agent.ts still uses removed consumesEphemeral types. It does not enter the project-worker/linter build; migrate it separately before the next voice agent deployment rather than mixing that large change into this repair.
  • 2026-08-25: PR #2514's ready_for_review delivery created the linter child, loaded policy v6 from config commit f059bea, reported two terminology/no-metaphorical-lane-door-seam diagnostics, and published comment-only review 5016262887 plus a neutral Check Run. The review-bot connection subscription finished at zero lag with no recorded error.
  • 2026-08-26: Added a generic per-rule suggestions policy. Rules default to allowed, while the metaphor rule uses forbidden; the agent prompt omits a fix and the mechanical publisher refuses to render one even if supplied.
  • 2026-08-26: Production config commit b03c198 pins the linter package from this PR and activates policy v7. OS published project/worker-updated for that exact commit; the refreshed review-bot subscription is active.
  • 2026-08-26: Analysis 5982 used policy v7, prompt v4, and config b03c198. Its two durable diagnostics contained no fix; iterate[bot] review 5029789862 published the same explanation-only comments with no suggestion fences.
  • 2026-08-27: Merged latest main, which also renewed the expired mobile parked-test marker. Cursor Bugbot exposed that historical analysis requests predated the required suggestions field; added a replay regression test and made the field default to allowed at the event contract boundary.
  • 2026-08-27: Tightened prompt v5 so forbidden-suggestion diagnostics only describe what is unclear and never tell the author what to change. Removed the deliberate reviewLane probe after its live proof and completed the full local gate successfully.
  • 2026-08-27: Production ingested private config commit c80df28, which pins this PR's current package and activates policy v8. Resolved every historical proof thread and the Cursor replay report before the final current-head bot pass.

Was this page helpful?

Powered by Docsbook