gh-process-review_skill

Overview

This skill automates processing GitHub pull request review comments by fetching review threads to a local JSON file, guiding planning, implementing each fix, and tracking progress. It minimizes context pollution by storing reviews locally, commits each fix separately, and starts in planning mode by default. Use the optional "continue" argument to skip fetching and resume from an existing reviews file.

How this skill works

When invoked, the skill runs a single entry-point script from the project root to fetch or list review threads into a .scratch/reviews JSON file. It guides you through a planning phase that produces a self-contained plan, then enforces a one-commit-per-fix workflow: get thread details, implement the fix, commit, push, reply with the commit, and mark locally resolved. The script supports fetch, list, get, reply, and mark subcommands and uses jq-compatible JSON for parsing.

When to use it

• You need to address PR review comments without polluting agent context.
• You want an audit trail of each individual fix (one commit per review thread).
• You prefer offline, local review processing and step-by-step confirmation.
• You have an existing reviews file and want to resume processing (continue mode).
• You need to produce a clear plan before changing code.

Best practices

• Always run from your project root and call the entry script; do not cd into the skill directory.
• Start in planning mode (default). Only skip planning on explicit instruction like "skip planning".
• Make one commit per review thread and push immediately before replying and marking.
• Resolve all paths to absolute script paths in the approved plan because context clears after planning.
• Do not resolve threads via the GitHub API; use local marking only and let reviewers close threads.

Example use cases

• Run the script to fetch unresolved review threads for the current branch and create a plan of fixes.
• Resume an interrupted session with: /gh-process-review continue, which lists threads from the existing JSON file.
• For a single thread: fetch details, implement fix, commit, push, reply with the commit hash, then mark locally.
• Generate proposed memory or documentation updates after finishing all threads to prevent repeat issues.
• Use the list full option to inspect raw JSON when complex thread context is needed.