feat(issues): respect declined contributor-issue drafts

[galuis116]

Closes #517.

generateContributorIssueDrafts (#462) deduped only against open issues (findDuplicateContributorDraft skips state !== "open"; loadContributorIssueDraftContext loads listOpenIssues). So a maintainer who closed a generated draft saw it re-proposed — and in create mode re-created — on every run while the originating warning persisted (fingerprints are stable for policy:* / upstream:registry_drift topics). That eroded trust in the automation by re-filing declined work.

Changes

• db/repositories.ts: add listClosedContributorDraftIssues(env, repo, markerPrefix, limit=200) — bounded, recent-first, filters closed issues whose payloadJson carries a contributor-draft marker.
• contributor-issue-draft.ts:
  • add findDeclinedContributorDraft(closedIssues, draft, { now?, cooldownMs? }) — matches by the stable marker fingerprint; a wontfix-style label (wontfix/invalid/duplicate/not-planned) suppresses re-proposal indefinitely, otherwise the closure is honored within a 30-day cooldown so a genuine later regression can resurface.
  • new skipped_declined status + declinedBy provenance + skippedDeclined result counter; load declinedIssues in the context and short-circuit after the open-duplicate check.
• Backward compatible: with no closed marked issues, output is unchanged. The tested open-only semantics of findDuplicateContributorDraft are preserved (closed-issue handling is a separate, additive check).

Verification

• New unit tests: findDeclinedContributorDraft (marker / wontfix / cooldown boundary / open / no-marker / other-fingerprint) and a generation test (closed marked issue -> skipped_declined, created: 0, no GitHub POST).
• tsc --noEmit clean; full unit suite green (1330 passed).

[ghost]

Note

Gittensory Gate skipped

PR closed before full evaluation. No late first comment was created.

Signal	Result	Evidence	Action
Gate result	⚠️ Skipped	#518 is no longer open.	No action.

---

Checked by Gittensory, a quiet PR intelligence layer for OSS maintainers.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[JSONbored]

Approving. This correctly closes the trust-eroding re-proposal loop from #517 and is well-tested.

Root cause is accurate: generateContributorIssueDrafts deduped only against open issues (findDuplicateContributorDraft skips non-open; loadContributorIssueDraftContext loaded only listOpenIssues), so a maintainer who closed a generated draft saw it re-proposed — and in create mode re-created — on every run, since policy:* / upstream:registry_drift fingerprints are stable.

The change is clean and additive:

• listClosedContributorDraftIssues is bounded (limit 200), recent-first, and filtered by the draft marker in payloadJson.
• findDeclinedContributorDraft matches by the stable marker fingerprint; a wontfix-style label suppresses indefinitely, otherwise a 30-day cooldown is honored so a genuine later regression can resurface. The closed-issue check is layered after the open-duplicate check, preserving the existing open-only semantics.
• New skipped_declined status + declinedBy provenance + counter are threaded through consistently.

Tests cover the marker / wontfix / cooldown-boundary / open / no-marker / other-fingerprint matrix plus an end-to-end generation test asserting skipped_declined, created: 0, and no GitHub POST. Backward compatible (no closed marked issues -> unchanged output). CI green.

Minor note (non-blocking): findDeclinedContributorDraft keys the cooldown off updatedAt as a proxy for closed-at; if an issue is closed and later edited, updatedAt moves forward and could extend the cooldown. Acceptable for this use, but a closedAt field would be more precise if one is available later.

[ghost]

reviewbot · verdict: comment

Reviewed 3 changed file(s) — two independent AI reviewers:

Reviewer A · llama-3.3-70b-instruct-fp8-fast
This PR introduces a new function to list closed contributor draft issues, allowing the system to respect declined contributor-issue drafts. The code is clean and well-structured. Overall quality is high.

Suggestions

• Consider adding input validation for the markerPrefix parameter to prevent potential SQL injection attacks
• Add documentation for the new function in the repository's README or API documentation

Worth double-checking

• Double-check the performance impact of the new query, especially with large issue datasets
• Verify that the markerPrefix parameter is properly sanitized to prevent SQL injection

Reviewer B · mistral-small-3.1-24b-instruct
This PR adds functionality to respect declined contributor-issue drafts by listing closed issues with a specific marker. The implementation is clean and follows existing patterns.

Suggestions

• Consider adding a comment explaining the purpose of the markerPrefix parameter in the listClosedContributorDraftIssues function.
• Ensure that the limit parameter has a reasonable default and is well-documented.

Worth double-checking

• Verify that the SQL query correctly handles edge cases, such as issues with very long payloads or special characters in the marker.
• Check that the new function is thoroughly tested, especially for performance with large datasets.