Gh Pr Comment Fixer

Use this on a writable PR branch when actionable review feedback exists.

If you only need to detect whether review feedback, CI failure, stale head state, or mergeability needs attention, start with the read-only babysitter snapshot instead:

uv run python scripts/dev/pr_babysitter_snapshot.py <pr-number> --expected-head-sha <sha> --json

Use this skill after that snapshot recommends process_review_comment or when the user explicitly asks to fix known review comments.

Scope

• Fetch PR context for current branch.
• Read docs/context/issue_relationships.md and the linked issue's current native relationship state while fixing comments. PR bodies do not mirror those links.
• Collect review threads and top-level PR comments.
• Apply only fixable-now items that stay inside PR scope.
• Run proof at the tightest practical level, then push and resolve threads.

Workflow

1. Confirm PR for current branch and branch match.
2. Pull requested comments and review threads; keep thread IDs for actionable items. Re-read the linked issue's native Parent/Blocked by/Blocking state before any relationship-related edit.
3. Classify requests:
   • fixable now,
   • needs clarification (ask before editing),
   • out-of-scope (defer).
4. Apply minimal edits grouped by concern.
5. Run appropriate validation (prefer targeted first; full suite only when needed by risk).
6. Commit and push.
7. Rebuild the final PR title/body from the post-fix diff and validation, then run the REST reconciliation helper with the final title and body. A no-op is valid; any metadata change invalidates prior final-state review evidence and requires a fresh exact-head review trailer. Do not infer or silently change graph edges while applying review fixes.
8. Re-query unresolved review threads after the push and metadata reconciliation before resolving anything. Bots may add fresh findings once a draft PR becomes ready or after the first fix commit.
9. Resolve only addressed threads using the resolveReviewThread mutation via gh api graphql.
10. Re-check thread state; report any unresolved items with blocker reason (permissions, rate limits, or external dependency).

Anti-Loop / Retry

• Do not rerun unchanged failing validation more than twice without code change or clarified input.
• If a review request cannot be fixed safely, leave a precise blocker note and stop.

Artifact and Race Rules

• Inspect output/-class artifacts before commit and document disposal/durability.
• Verify PR head before applying fixes; avoid resolving threads not yet satisfied by code or commit.
• Do not force-push or mutate thread state for unresolved/ambiguous feedback.
• Do not create native issue links from review prose. Relationship changes require an explicit issue declaration, a fresh read, and native-link readback from the writable worktree.

Required Output

• Which comments were fixed,
• validation command and result,
• commit SHA pushed,
• post-push unresolved-thread re-query result,
• resolved thread IDs,
• remaining open threads/blockers.

When to use

Use this skill for the scope named in its frontmatter description and registry metadata.

Guardrails

• Stay within the skill scope declared in .agents/skills/skills.yaml.
• Prefer repository scripts and canonical docs before ad-hoc commands.
• Record blockers and validation gaps instead of overstating completion.

Output

Return the schema named by the output_schema frontmatter field, or a compact equivalent when the caller does not require YAML.