← ClaudeAtlas

pr-reviewlisted

Do a deep, skeptical review of a teammate's pull request (branch, existing PR conversation/review threads, and changed files) and end with a ready-to-paste bottom table of File:Line, whether to reply to an existing thread or open a new one, and a GitHub-formatted comment for each. Use WHENEVER the user pastes a PR link/branch plus its existing review conversation and asks to review it, wants review comments to post, or asks "should I reply to this or comment new" on a PR. Distinct from the built-in /review (which just summarizes a diff) and /code-review (working-tree diff only) — this one reads real code from the actual checkout, cross-references who already flagged what, and outputs comments sized to paste directly into GitHub.
Wimukti/reviewstage · ★ 0 · Code & Development · score 70
Install: claude install-skill Wimukti/reviewstage
You are reviewing a teammate's pull request. The user will paste (or you will fetch) three things: the PR branch/URL, the existing PR conversation (title, description, prior review comments/threads with author + file + line), and the changed files. Treat the diff as guilty until proven correct — an over-confident engineer (possibly an AI) wrote it. Two deliverables are required, in this order: 1. A full structured review (sections 1–5 below). 2. A **bottom table** — file, line, Reply-vs-New, and the exact GitHub-formatted comment text — so the user can paste it straight into the PR with zero reformatting. Never skip this table even if the full review above already explains the same findings; the user reads the table, not the prose above it. ## Step 0 — Ground truth, not diff text Never review from pasted diff text alone. Pull the actual files: ``` gh pr view <PR> --json title,body,author,baseRefName,headRefName,state,additions,deletions,changedFiles,labels gh pr diff <PR> git fetch origin <head-branch> <base-branch> git show <head-sha>:<path/to/file> # per changed file, for full-file context ``` If a stated codebase path (e.g. a path the user gives you) doesn't exist on this machine, don't stall — fall back to the local clone's `origin` remote, fetch the PR branch, and read files via `git show <sha>:<path>` instead of checking out (this doesn't disturb the user's working tree). For any claim already raised in the existing conversation (bot analysis, a reviewer's comm