Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/goulinkh/code-review-harness/llms.txt

Use this file to discover all available pages before exploring further.

Code Review Harness (CRH) automates AI-powered code review for merge proposals. Point it at a Launchpad merge proposal URL, choose your LLM backend, and CRH will orchestrate a review session — delegating file inspection to parallel sub-agents and posting the resulting findings to stdout or directly back to Launchpad as inline and summary comments.

Quickstart

Run your first AI code review in under five minutes

Installation

Install via npm, pnpm, or build from source

CLI Reference

Every flag and option for the crh review command

Programmatic API

Embed CRH in your own TypeScript pipelines

​
How it works

CRH follows a three-phase workflow for every review:
1

Prepare workspace

CRH fetches PR metadata, CI status, diffs, and existing comments from the provider (Launchpad) and writes a deterministic workspace tree to a temporary directory. It also clones the target repository so the agent can read source files.
2

Run the review session

An orchestrator agent reads the workspace, groups changed files into scopes, and fans out review work to parallel sub-agents via delegate_review. Each sub-agent inspects its assigned files and returns structured findings.
3

Emit results

The orchestrator calls submit_review with the aggregated findings. The configured sink — stdout or Launchpad — receives the review and posts inline comments and a summary verdict.

​
Key features

Multiple LLM backends

Works with Anthropic Claude, OpenAI, Kimi (Moonshot), OpenRouter, and any OpenAI-compatible server

Parallel sub-agents

Fan out large diffs across concurrent sub-agents — each with its own context window — for fast, thorough reviews

Structured findings

Reviews return typed severity levels: blocker, major, minor, and info — no freeform text blobs

Pluggable providers & sinks

Swap out providers (where code lives) and sinks (where reviews go) via a clean TypeScript interface

​
Explore the docs

Concepts: workspace model

Understand the deterministic workspace CRH builds before running the agent

Concepts: agent architecture

Learn how the orchestrator and sub-agents collaborate on a review

Launchpad provider

Configure the Launchpad merge proposal provider

API reference

Full TypeScript API reference for programmatic usage

Build docs developers (and LLMs) love

Get started for freeTalk to us