← ClaudeAtlas

review-prolisted

One-command AI code review. Runs triage -> relevant specialist reviewers -> synthesis on the current branch and returns the verdict + report. Stack-specific signals are loaded automatically from the repo's .review-pro/ directory. Use to review a branch or PR with review-pro.
tufantunc/review-pro · ★ 4 · Code & Development · score 77
Install: claude install-skill tufantunc/review-pro
# Review-Pro (one-command review) You are the **orchestrator**. Run the entire pipeline on the current branch in ONE pass and return the final verdict + report. Do everything with your own native tools — shell for git, Read/Glob/Grep for files. **Do NOT ask the user to run any scripts.** Do not hand off between stages. ## Procedure ### 1. Prep (native — you do this, not the user) - **Base branch:** `main`, falling back to `master` if `main` doesn't exist. - **Changed files:** run `git diff --name-only <base>...HEAD` in your shell. Read each changed file's full contents with Read. (git already excludes gitignored/generated paths from the diff.) - **Installed stacks:** `Glob .review-pro/*/manifest.json`. Each match is a stack the user installed (via `npx review-pro`). These are the repo's **active stacks**. If `.review-pro/` is absent or empty, reviewers run on their core rubric only. ### 2. Triage (you, inline) Follow the `review-pro-triage` skill. Classify the changed files, detect concern relevance, and produce a **dispatch plan**: which reviewers to run + each one's scoped context (per `core/shared/context-policy.md`). Be conservative — when in doubt, dispatch. ### 3. Fan-out — reviewers (subagents, parallel) For each reviewer in the dispatch plan: 1. **Gather its stack signals.** For each installed stack, Read `.review-pro/<stack>/<reviewer>.md` **if it exists**. Concatenate the ones you find — this is the reviewer's `### Stack signals` content. (The subagent auto-lo