onchainyaotoshi

Agent Review Pipeline — Data skill for Claude Code

Data community

Autonomous code review pipeline for Claude Code.

How to install Agent Review Pipeline

This entry records only its repository, not the path inside it, so there is no exact command to give. Open onchainyaotoshi/agent-review-pipeline and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agent Review Pipeline does

Autonomous code review pipeline for Claude Code. Asymmetric dual-engine dispatch — Codex (correctness + adversarial) + Gemini (/ce:review compound-engineering pipeline). Fingerprinted consensus, bounded auto-fix, loop-thrash kill switch, safe-off defaults.

Alternatives in Data

Privacy & Data Handling — This repository contains: - a plugin package (plugins/compound-engineering) made of markdown/config content 10.8k ★
Autonovel — An autonomous novel writing pipeline, by Hermes Agent 1.5k ★
Claude Deep Research Skill — Enterprise-grade deep research skill for Claude Code with 8-phase pipeline, source credibility scoring, and au 1k ★

README

agent-review-pipeline

Autonomous dual-engine code review pipeline for [Claude Code](https://claude.ai/code). **Asymmetric dispatch** — [Codex](https://github.com/openai/codex-plugin-cc) runs ARP's dual-framing (correctness + adversarial) while [Gemini CLI](https://github.com/google-gemini/gemini-cli) runs its own `/ce:review` compound-engineering pipeline. 3 parallel perspectives per iteration. Dedups findings by confidence, auto-fixes inline, escalates unresolvable findings. Verification delegated to your CI.

What It Does

Asymmetric Dual-Engine Review — 3 parallel dispatches per iteration: Codex × correctness, Codex × adversarial, Gemini × /ce:review. Codex gets ARP's framing discipline; Gemini uses its own multi-persona compound-engineering pipeline (P0-P3 tiering) because it already has one. No redundancy.
Confidence-Weighted Consensus — Findings fingerprinted by file:line:severity:normalize(issue):sha1(fix_code[:200]). Multi-source agreement boosts confidence by +0.15 per extra source. Findings below 0.60 confidence are dropped.
Bounded Auto-Fix Loop — Applies fix_code inline, re-runs until PASS or maxIterations (1-10, default 3). Unlimited loops are intentionally not supported.
Loop-Thrash Kill Switch — If a fingerprint reappears after its fix was applied, the fix didn't work. The finding is escalated to human review instead of looping forever.
Safe Defaults — autoCommit and postPrComment default to false. --dry-run previews findings without editing. Dependency precheck fails fast if Codex / Gemini CLI missing.
Agreement Telemetry — Per-run .arp_session_log.json records Codex↔Gemini agreement rate. Tune dual-engine cost/value over time.

How It Works

flowchart TD
    Start([/arp PR]) --> Pre

    subgraph Stage0["Stage 0: Pre-flight"]
        Pre[deps · flock · gh auth · PR resolve
+ PR conversation context fetch]
    end

    subgraph Stage1["Stage 1: Review (dispatch →