crh review command flags reference

Every flag accepted by crh review is documented here, including its type, default value, and behaviour. Two flags — --no-sandbox and --config — are parsed by the CLI but their underlying functionality is not yet complete in v1.

Flags

--provider

string

required

The code hosting provider to fetch the merge proposal from. Only launchpad is supported in v1. Passing any other value causes the command to exit with an error.

--provider launchpad

--pr

string

required

The Launchpad API URL for the merge proposal to review. Must be a git merge proposal URL in the form https://api.launchpad.net/devel/~<user>/+git/<repo>/+merge/<id>. Bazaar merge proposal URLs are explicitly rejected.

--pr "https://api.launchpad.net/devel/~user/+git/repo/+merge/123"

--sink

string

default:"stdout"

Where the completed review is emitted.

stdout — writes a single JSON line to standard output. Use this for local dry-runs or scripted pipelines.
launchpad — posts inline comments for each finding and a summary comment with a verdict vote back to the merge proposal. Requires valid Launchpad OAuth credentials (LP_ACCESS_TOKEN, LP_ACCESS_SECRET, LP_CONSUMER_KEY).

--sink launchpad

--model

string

default:"anthropic:claude-sonnet-4-6"

The LLM to use for the review session, in provider:model-id format. The provider prefix determines which API client and key environment variable CRH uses. See Model selection for the full list of examples and the API key naming convention.

--model anthropic:claude-sonnet-4-6

--model-base-url

string

Override the API base URL for the chosen model. Only valid with OpenAI-compatible models (openai:* or a registered provider whose API type is openai-completions / openai-responses). Passing this flag with a non-compatible model causes the command to exit with an error.

--model openai:gpt-5.4-mini --model-base-url "http://localhost:8080/v1"

--debug

boolean

Dump all raw session events and sub-agent events to stderr as newline-delimited JSON. Useful for diagnosing unexpected agent behaviour or inspecting the full event stream during development.

--debug

--no-sandbox

boolean

Accepted by the CLI parser and stored as an option, but sandbox lifecycle enforcement is incomplete in v1. Passing this flag does not currently change runtime isolation behaviour.

--config

string

Accepted by the CLI parser as a path to a configuration file, but config file loading is incomplete in v1. Passing this flag has no effect on the running session.

--no-sandbox and --config are accepted by the CLI in v1 but are not fully implemented. They will not change agent isolation or load any configuration file. Their behaviour will be defined in a future release.

Complete example

The following command combines multiple flags to post a review using a locally running OpenAI-compatible model with debug output enabled:

OPENAI_API_KEY="$YOUR_KEY" \
LP_ACCESS_TOKEN="$LP_TOKEN" \
LP_ACCESS_SECRET="$LP_SECRET" \
LP_CONSUMER_KEY=crh \
crh review \
  --provider launchpad \
  --pr "https://api.launchpad.net/devel/~user/+git/repo/+merge/123" \
  --sink launchpad \
  --model openai:gpt-5.4-mini \
  --model-base-url "http://localhost:8080/v1" \
  --debug