feat(#836): no-LLM duplicate-issue detection + challenge + 1-day auto-close (#843)

* feat(#836): no-LLM duplicate-issue detection + challenge + 1-day auto-close

Adds a deterministic (no-LLM) duplicate-issue governance lifecycle:

- scripts/issue-dedupe.cjs: pure, unit-tested module (tokenize, Sørensen–Dice
  title similarity, scoreCandidates, renderChallengeComment, shouldClose) with
  fail-safe destructive-action guards.
- duplicate-check.yml (issues:opened): scores new-issue title against open
  issues, posts a challenge comment + applies the pending `possible-duplicate`
  label on a clear match.
- duplicate-sweep.yml (daily cron): closes possible-duplicate issues whose
  challenge comment is >24h old with no human reply and no 👎 veto; honors
  exempt labels; re-checks the label immediately before close (TOCTOU guard);
  strips the label on close to avoid reopen loops.
- remove-duplicate-label.yml (issue_comment:created): clears the label and
  applies needs-maintainer-review when any human responds.
- bug_report.yml / docs_issue.yml: add the required "I searched existing
  issues" preflight checkbox so all five forms force a pre-search attestation.
- docs/agents/triage-labels.md: document the label + lifecycle.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* chore(#836): add changeset fragment for duplicate-issue detection

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Tom Boucher
2026-06-07 20:35:43 -04:00
committed by GitHub
parent 988024c1a3
commit d32b8db635
9 changed files with 1117 additions and 0 deletions

View File

@@ -9,6 +9,7 @@ Maps the five canonical triage roles to the actual label strings in `open-gsd/gs
| `ready-for-agent` | `confirmed` | Bug verified + fully specified — AFK agent can pick up |
| `ready-for-human` | `approved-enhancement` / `approved-feature` | Enhancement/feature approved by maintainer — human codes it |
| `wontfix` | `wontfix` | Will not be actioned |
| `possible-duplicate` | `possible-duplicate` | Applied by the Duplicate check workflow when a new issue's title closely matches existing open issues. The reporter (or a maintainer) replies justifying why it is not a duplicate within 24h, or the Duplicate auto-close sweep closes it. A reply clears this label and applies needs-maintainer-review for human adjudication. React 👎 to the bot comment to veto auto-close. |
## Notes on this repo's label model
@@ -17,3 +18,12 @@ Maps the five canonical triage roles to the actual label strings in `open-gsd/gs
- There is no separate "ready-for-human" vs "ready-for-agent" distinction for enhancements — both flow through the same `approved-*` labels. If the work requires human judgment (design decisions, external access), note it in the issue body.
- `needs-triage` is removed when any other state label is applied.
- `needs-reproduction` is used instead of the generic `needs-info` — be specific in triage comments about what reproduction steps or information are missing.
## Duplicate detection lifecycle
The `possible-duplicate` label is managed by three GitHub Actions workflows that together form a self-service deduplication loop:
1. **Detect on open** — When an issue is opened, `duplicate-check.yml` scores its title against all other open issues using Dice-coefficient similarity. If any match clears the threshold, the bot posts a challenge comment listing the similar issues and applies `possible-duplicate`.
2. **Challenge comment + reporter window** — The reporter (or a maintainer) has `DEFAULT_WINDOW_HOURS` (24h) to reply explaining why the issue is not a duplicate. Reacting 👎 to the bot comment also signals the reporter objects to auto-close.
3. **Daily sweep auto-close** — `duplicate-sweep.yml` runs at 07:00 UTC daily. For each open issue with `possible-duplicate`, it checks whether the window has elapsed, whether the reporter replied, and whether a 👎 reaction exists. Issues with exempt labels (`priority: critical`, `pinned`, `confirmed-bug`, `confirmed`, `fix-pending`) are never auto-closed. Issues that pass the close check receive a closing comment and are closed with `state_reason: duplicate`.
4. **Reporter reply clears label** — `remove-duplicate-label.yml` fires on every new non-bot comment. If the issue still carries `possible-duplicate`, it removes that label and applies `needs-maintainer-review` (the value of `HUMAN_REVIEW_LABEL` in `scripts/issue-dedupe.cjs`), routing the issue to a maintainer for manual adjudication.