From pr
Read all review comments on a GitHub PR using the gh CLI, summarize the issues, categorize them as code-fixable vs needs external config/admin, and produce a prioritized fix plan. Trigger when the user runs /pr:summarize with a PR URL or number, or asks to summarize/plan fixes for PR review comments.
How this skill is triggered — by the user, by Claude, or both
Slash command
/pr:summarize <PR URL or PR number><PR URL or PR number>The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Given a PR URL or number, fetch all review feedback and produce a clear fix plan.
Given a PR URL or number, fetch all review feedback and produce a clear fix plan.
Parse the argument for a PR URL (e.g. https://github.com/org/repo/pull/42) or a bare number (e.g. 42 or #42).
OWNER, REPO, and PR_NUMBER from it.OWNER and REPO from the current git remote:
gh repo view --json owner,name
gh pr view --json number,url,headRepositoryOwner,headRepository
Run both calls with --paginate to avoid missing anything on busy PRs:
# Inline code comments (attached to specific file lines)
gh api repos/OWNER/REPO/pulls/PR_NUMBER/comments --paginate
# Top-level review bodies (overall review text)
gh api repos/OWNER/REPO/pulls/PR_NUMBER/reviews --paginate
Skip reviews where state is APPROVED and the body is empty — those add no signal.
Group comments by theme or code area (e.g. "workflow permissions", "author login matching", "pagination"). For each group write:
Be concise — this is a map, not a transcript.
Split every issue into one of two buckets:
Fixable in the repository — changes to source files, YAML, prompts, configs checked into the repo.
Needs external action — GitHub org/repo settings, secrets, admin permissions, third-party service config. For each, say what needs to happen and who likely needs to do it.
List the repository-fixable issues in priority order (significant → moderate → minor). For each:
Follow with a short "External actions needed" section listing anything requiring config or admin access, so the user knows what to pursue in parallel.
Keep the whole output skimmable. No padding, no restating the reviewer's comments verbatim.
Guides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub vialytics/claude-skills --plugin pr