← ClaudeAtlas

independent-reviewlisted

Spawn an independent reviewer agent that re-reviews the current branch with no implementer context. Use after the implementing agent has written its own review.md, or as part of the /ship-feature flow. Catches the class of bugs that single-agent self-review systematically misses.
renewisepunk/wisepunk-agentic-engineering-framework-v2 · ★ 2 · Code & Development · score 63
Install: claude install-skill renewisepunk/wisepunk-agentic-engineering-framework-v2
# Independent Review The implementing agent reviews its own work in `/ship-feature`. That's checking your own homework. This skill spawns a second agent that re-reviews the diff with no implementer context — no worklog, no implementation chat — and produces a second review. ## When to use - After `/ship-feature` Step 4 has written `review.md` but before Step 5 (compound) and Step 7 (close). - On any PR where you want a second pair of eyes before merging. - As a standalone audit on a recently-shipped change. ## What the reviewer gets - `git diff origin/main...HEAD` — the actual changes - `ai/runs/<run>/plan.md` — what was supposed to be built - `ai/STANDARDS.md` — rules to enforce - `ai/checklists/review.md` — the categorical checks - `ai/knowledge/pitfalls/` — listing (so it can grep for relevant ones) ## What the reviewer does NOT get - `worklog.md` — would anchor it to the implementer's narrative - `review.md` (the implementer's) — would anchor it to the implementer's conclusions - Conversation history from the implementing session - Any context about why decisions were made This isolation is the whole point. The reviewer reads only what a fresh code reviewer would read. ## Step 1 — Gather inputs ```bash RUN_DIR=$(ls -dt ai/runs/*/ | head -1) # or use the path the user gave you DIFF=$(git diff origin/main...HEAD) PLAN=$(cat "$RUN_DIR/plan.md") STANDARDS=$(cat ai/STANDARDS.md) CHECKLIST=$(cat ai/checklists/review.md) PITFALLS=$(ls ai/knowledge/pitfalls/) ``` ##