What problem does it solve?
Turn a pile of unresolved GitHub PR review threads into a small set of cluster-aware fix plans and automated actions, reducing noise and surfacing systemic issues before making code changes. It prevents ten one-off replies from obscuring two or three real concerns and avoids unnecessary churn by triaging handled vs new threads and grouping related feedback by category and spatial proximity.
Core Features & Use Cases
- Fetch & Triage: Retrieve unresolved review threads via GraphQL and classify them as new, handled-not-replied, handled-and-replied, or outdated.
- Cluster Analysis: Group new threads into one of eleven fixed categories and by spatial proximity to surface systemic issues and produce per-cluster fix hypotheses.
- Dispatch & Safety Gates: Dispatch parallel pr-comment-resolver subagents for safe fixes, gate ambiguous changes for confirmation, batch manual/taste decisions for humans, and post advisory replies without code changes.
- Modes & Artifacts: Support interactive, autofix, report-only, and headless modes and write run artifacts and manual-decision briefs for cross-invocation continuity and auditing.
Quick Start
Use the resolve-pr-feedback skill to fetch unresolved review threads for PR 123 in owner/repo and produce a clustered plan in interactive mode.