plugins/claude-core/skills/pr-thread-resolver/SKILL.md
Fetch, fix, and resolve unresolved GitHub PR review threads end-to-end with the gh CLI and Claude Code's edit/verify loop. Use when addressing PR review comments or reviewer feedback, fixing or resolving review threads, clearing unresolved or blocking conversations before merge, or applying GitHub suggested changes. Pulls fresh review-thread state via gh GraphQL, verifies each finding against current code (skips stale or already-fixed), fans independent threads out to parallel Task subagents, applies minimal fixes, runs repo-native verification, makes scoped conventional commits, pushes, then resolves only threads backed by a committed and verified fix. Never resolves on head drift, failed verification, or skipped threads. Not for local Codex/Zen/manual review notes (use review-remediation), passive PR or CI watching until merge (use babysit-pr), or CI-log-only failures with no review thread.
npx skillsauth add bjornmelin/dev-skills pr-thread-resolverInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Turn unresolved GitHub PR review threads into committed, verified, resolved fixes — natively, with
gh + Claude Code's edit/verify loop. No external orchestrator.
Use for hosted PR review threads. Route away:
review-remediation.babysit-pr.gh workflow-log remediation.AGENTS.md/CLAUDE.md; git status --short; capture the head SHA
(gh pr view <pr> --json headRefOid -q .headRefOid).gh api graphql into a compact worklist
(see references/gh-graphql.md): thread id, path, line, isResolved, isOutdated, body, any
suggestion hunk.Read the current line. Skip with a reason if already-fixed or stale
(isOutdated + code no longer matches); else queue a minimal fix.suggestion block verbatim only on an
exact hunk match; otherwise write a hand-minimal fix./verify and /code-review as the backbone).references/closeout.md).resolveReviewThread only threads
backed by a committed+pushed+verified fix (or already-fixed at current head).When ≥3 threads touch disjoint files, dispatch one read-and-fix Task subagent per file/cluster.
Subagent contract: fix + verify + report only (file:line, change, verification result) — it does
not commit, push, or resolve. The parent serializes all staging, commits, pushes, and
resolveReviewThread mutations. Workers own disjoint files and never touch another's.
Resolve a thread ONLY when it maps to a committed, pushed, and verified fix (or is already fixed at current head). Never resolve when: the PR head drifted unexpectedly since you pushed; verification failed; the thread was skipped, ambiguous, or unmatched. Do not auto-reply by default — reply only when a thread cannot be fixed cleanly or the user asks for a comment.
Worklist · verified-fix / skipped summary (file:line evidence) · verification commands + results ·
commit SHAs · thread→commit closeout map · terminal status completed | blocked | no-op.
references/gh-graphql.md — copy-paste gh api graphql queries + the resolveReviewThread mutation.references/closeout.md — Conventional Commit grouping, forbidden process-wording, head-drift guard.development
Pre-PR multi-model review, parallel opus and codex exec adversarial lanes, then adversarial verification of merged findings. Read-only. Use before shipping nontrivial diffs.
tools
Independent gpt-5.6 diff review via the Codex CLI, normal or steerable adversarial with JSON findings. Use before shipping nontrivial changes.
development
Delegate implementation, investigation, or bulk work to gpt-5.6 codex via pinned codex exec. Use for clear-spec builds, migrations, debugging, or any task MODELS.md routes to codex.
development
Adversarial pre-mortem: imagine the plan failed, work backwards to surface risky assumptions + irreversible bets, then harden them. Proactively offer it (after the current request; confirm first) before a hard-to-reverse or one-way-door call (API, schema, framework, a hire), an all-upside plan, or unvalidated assumptions. Also on request.