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.

The crh review command kicks off an AI-powered code review session against a Launchpad git merge proposal. It fetches the proposal metadata and diff, runs an agent session against the configured LLM, and writes the resulting review to the sink you choose — either standard output for a local dry-run, or directly back to Launchpad as inline comments and a summary vote.

​
Syntax

crh review --provider launchpad --pr <url> [options]

​
Required flags

​
--provider
string
required
The review provider to use. Only launchpad is supported in v1. Passing any other value causes the command to exit with an error.
​
--pr
string
required
The Launchpad API URL for the merge proposal you want to review. Must be a git merge proposal — Bazaar merge proposals are rejected. Example: https://api.launchpad.net/devel/~user/+git/repo/+merge/123

​
Optional flags

FlagDefaultDescription
--sinkstdoutWhere to emit the review. stdout prints JSON to standard output; launchpad posts inline comments and a vote back to the merge proposal.
--modelanthropic:claude-sonnet-4-6LLM backend in provider:model-id format. See Model selection.
--model-base-url—Override the API base URL for OpenAI-compatible providers. Requires an OpenAI-compatible model.
--debug—Dump all raw session events to stderr for troubleshooting.
--no-sandbox—Accepted by the CLI; sandbox enforcement is incomplete in v1.
--config—Accepted by the CLI; config file loading is incomplete in v1.
For full flag details see the flags reference.

​
Examples

crh review \
  --provider launchpad \
  --pr "https://api.launchpad.net/devel/~user/+git/repo/+merge/123" \
  --sink stdout
When using --sink launchpad you must supply valid OAuth 1.0a credentials via LP_ACCESS_TOKEN, LP_ACCESS_SECRET, and LP_CONSUMER_KEY. LP_CONSUMER_KEY defaults to crh when omitted.

​
What happens when you run it

1

Workspace preparation

CRH fetches the merge proposal metadata, description, CI results, and numbered diff from Launchpad and writes them to a deterministic local workspace. Progress is streamed to stderr.
2

Agent session

The configured LLM agent reads the workspace files and runs a code review using read-only built-in tools plus CRH-specific tools. Mutating tools (bash, edit, write) are not exposed to the agent.
3

Sink emission

The agent calls submit_review when it finishes. CRH forwards the structured output to the configured sink — either a single JSON line on stdout or inline comments and a vote on the Launchpad merge proposal.

Model selection

Choose the LLM backend and configure API keys.

Flags reference

Complete reference for every flag accepted by crh review.

Build docs developers (and LLMs) love

Get started for freeTalk to us