Documentation

CLI reference

The commands that initialize, inspect, prepare, review, and maintain a Rationale project.

Commands

Command	Purpose
`init`	Create the `.rationale/` canon, or keep it intact when it exists, then converge agent integration.
`health`	Report project identity, Git revision, provider status, and coverage.
`prepare <target>`	Compile context for a path or symbol.
`review`	Human-review pending proposals.
`review-record <id>`	Mutate an approved Record through its audited lifecycle.
`install-agent`	Add idempotent instructions and MCP registration.
`uninstall-agent`	Revert only what `install-agent` wrote.
`update`	Install the latest preview selected by the local helper.
`serve`	Run the persistent MCP stdio server.

Common options

rationale health --project-root /path/to/project
rationale prepare "src/auth.rs::resolve" --project-root /path/to/project
rationale install-agent --project-root /path/to/project --dry-run

prepare takes the first positional argument that is not an option as its target. init --help and invalid options are side-effect free. init does not accept --project-root; run it from the project root instead. Running init again still detects and configures agents, so a project initialized before Claude Code was installed can recover without deleting .rationale/. --skip-agent-config and RATIONALE_SKIP_AGENT_CONFIG=1 keep that step disabled on both the first and subsequent runs.

Output boundary

Commands that return a machine contract keep JSON on stdout. Human-facing diagnostics and Chestie stay on stderr or are suppressed for non-terminal output. serve never prints a banner to stdout.