Request: PR Review Closure And Thread Sync Skill

Request Summary

Add a dedicated skill plus hook support for closing PR review loops with less manual gh api plumbing and fewer missed unresolved threads.

Problem Statement

Current PR-review follow-ups still require manual composition of:

• review-thread fetch commands,
• unresolved thread filtering,
• per-thread reply/resolution commands,
• and post-push verification.

This is repeated often enough to justify a standard skill and one helper hook.

Evidence

• Recent PR follow-up flow required direct gh api repos/<owner>/<repo>/pulls/<number>/comments usage to inspect inline Copilot comments, then manual reply posting and confirmation.
• Existing hooks cover CI snapshots and bounded PR wait, but not unresolved-thread inventory and closure planning.
• Tracker evidence shows repeated PR review/checkpoint loops across rounds, including manual verification that review threads are resolved.

Proposed Significant Change

New user-invocable skill

• Path: .github/skills/pr-review-closure-loop/SKILL.md.
• Purpose: run a deterministic loop for review-thread triage and closure.
• Core procedure: 1. Gather unresolved review threads and comments. 2. Map each thread to code change, doc clarification, or already addressed. 3. Apply narrow fix or post evidence reply. 4. Re-check unresolved count after push.

New hook script

• Path: Scripts/SkillHooks/Get-PrReviewThreads.ps1.
• Purpose: output unresolved/resolved thread inventory in JSON.
• Required output keys:
• pullNumber
• totalThreads
• unresolvedThreads
• threads[] with id, path, author, state, latestComment, url

Priority, Impact, And Confidence

• Classification: Now.
• Impact: high (faster and safer PR review closure loop).
• Effort: medium.
• Confidence: 91%.

Status

Deferred

Triage note (2026-05-31): keep repo skill count within 8-10 and avoid adding a second PR-delivery skill. The canonical .github/skills/pr-review-delivery/SKILL.md now covers end-to-end PR handling. Keep this request scoped to hook-script enhancements only.

Acceptance Criteria

• Skill can run end-to-end on a real PR and produce an explicit unresolved-thread delta before/after fixes.
• Hook output is stable JSON and can be consumed by the skill without custom parsing hacks.
• The skill defaults to conservative checks (single refresh) and only escalates polling when explicitly needed.

Follow-Up

• Optionally add a second helper script later for generating templated review replies from evidence (test command + result + file path).