Review Runner — Development agent for Claude Code
Opus worker that runs pr-review steps 1–4 (snapshot, repo trap KB, review pass incl.
How to install Review Runner
Installs to ~/.claude/agents/mdaaaao-ai-baton-review-runner.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/MdaaaaO/ai-baton/HEAD/agents/review-runner.md -o ~/.claude/agents/mdaaaao-ai-baton-review-runner.md Restart Claude Code, or start a new session, for it to be picked up.
What Review Runner does
name: review-runner description: Opus worker that runs pr-review steps 1–4 (snapshot, repo trap KB, review pass incl. the --deep lenses, independent verification) for ONE pull request, writes the triage sheet to $CTX/triage.json and returns only the overview block (≤3K tokens), so the diff never enters a long-lived prefix. Never posts anywhere and never asks the user. Trivial-PR --auto passes go to auto-runner. metadata: version: "6" updated: "2026-09-26" reviewed: "2026-09-24" model:
Alternatives in Development
- Ref Sheet Artist — 웹툰 캐릭터 레퍼런스 시트 아티스트 305 ★
- After Identifying Or Executing A Chain — findings.sh add chain " " --score \ --steps " step2 -> step3>" --mitre " " findings.sh update chain --status < 213 ★
- Chain Executor — Executes delegated chain steps with context isolation 186 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Acceptance Runner
Plays for real the walkthrough steps an agent can play — browser, requests, commands — and keeps evidence for
Khadgar
Tests a brief's "Done when" steps in the running app, as a user would, and reports what it finds. Use on a pul
Sense Check Opus
Hunts self-defeating logic in anything a stranger has to get through — a site, a form, an onboarding flow, a s
Dr Removed Behavior
Deep-review finder angle. Audits everything a diff deletes or replaces - guards, validations, error paths, tes
Inbound Triage
Maintainer-only. Use when triaging inbound GitHub issues and pull requests on skyf0xx/hedgehog — "triage the i
Pre Mortem Runner
PROACTIVELY use this subagent whenever product planning calls for a dedicated pre-mortem deep run, typically a
Related Skills
Lovable Cloud To Supabase Migration
Migrate from Lovable Cloud to your own Supabase. Official export path, 33 steps, every trap documented. Claude
Prompt Snapshot
Vitest-for-prompts. File-based prompt eval runner with snapshots and assertions — no SaaS, no dashboard, CI-na
Agentrouter Guide
FREE GPT-6 Astra (Sep 2026) + GPT-5.5 + Claude Opus via AgentRouter — OG credits, 1.05M ctx, one API key. Defi