Rook CLI Reference
Use rook to start the interactive terminal, or rook <command> from your shell. This reference was checked against the public Rook 0.1.13 CLI on October 23, 2026. It groups command syntax, examples, RCA, state changes, diagnostics, and output handling on one page. Commands that invoke the target or spend credits still require appropriate approval.
Using Claude Code or another coding assistant? Describe your goal in chat with the Rook skill; the assistant handles these commands. This page is the explicit CLI reference, not a list of commands you must manually repeat while using a skill.
Find a Command
- Terminal and Help:
rook,ask,guide,help,docs,clear,exit. - Account and Authentication:
login,auth,whoami,logout,plan. - Projects and Discovery:
project,explore,agent. - Profiles and Scenarios:
profile,generate,scenarios. - Synchronization, Runs, and Results:
status,sync,run,runs,report,ui. - Environment and Diagnostics:
env,mcp,doctor,update; export diagnostic logs. - RCA for an existing run: approval, analysis, and interpreting the saved result without rerunning tests.
- Shared flags and structured output, exit codes, environment variables, and interactive keys.
First Journey
login → project → explore → agent → profile add → generate → scenarios → sync → run → ui
Follow the tested quickstart for a runnable sample. A normal timeline run needs sync first; run --test deliberately stays off the timeline and keeps its evidence local.
Discover Syntax From the CLI
rook --version
rook --help
rook run --help
rook profile add --help
Inside the terminal, use /help run. Do not paste slash commands into a normal shell.
The screenshots below were captured inside the Rook 0.1.11 interactive TUI, using /help <command> in the quickstart workspace. They show command help, not completed paid operations. Only the Rook content is captured; no desktop, window title, or browser chrome is included.
Terminal and Help
rook
Use rook to start the interactive terminal in the workspace whose agent material and testing state you want to use.
Syntax
rook
rook --no-animation
rook --version
rook --help
The TUI keeps the active project, agent, profile, credits, command duration, progress, and permission questions visible. Use --no-animation for recordings, slow terminals, or a static startup.
First Start
- Change to the workspace containing your checked-out agent, PRD, or test specification.
- Run
rook. - Use
/loginif the session is not authenticated. - Choose a project with
/project. - Enter
/guidefor the workflow or/helpfor every command.
Run Rook from the intended workspace. The current directory selects the local .testmuai/rook/ store; project content is kept below projects/<project-id>/. The workspace is also the base for relative source, command, certificate, and evidence paths.
rook ask
Use rook ask to ask a question or describe a testing task in natural language. This is Rook's own model-backed operation and can spend Rook credits; it is different from asking a coding assistant that has loaded the Rook skill.
Syntax
rook ask <prompt...>
rook ask <prompt...> --verbose
rook ask <prompt...> --json
| Option | Purpose |
|---|---|
--verbose | Show subagent activity, tool activity, and credits while the request runs. |
--json | Return machine-readable output for this command. |
In an attended terminal, Rook can suggest a command, ask Run it?, and execute it after confirmation. In headless mode or with --json, it returns the answer and any suggested command without executing that suggestion. The JSON can include answer, command, and blocked_by. Review any proposed operation's spending, target effects, and permissions separately.
rook ask "Which agent is active and is its tree synchronized?"
rook ask "Run only the boundary scenarios with the staging profile"
rook ask "Explain the latest failures" --verbose
For deterministic automation, prefer the explicit command and flags. Natural language is useful for attended work and one-off requests, but it is not a stable machine interface.
/guide
Use /guide when you know you want to test an agent but do not yet know which command comes next.
Syntax
/guide
The shell form is rook guide.
The guide covers this sequence:
login → project → explore → agent → generate → profile → sync → run
A normal timeline run requires an upstream project version, so the guide puts sync before run. In a new project, synchronize the reviewed tree before the first normal run; use run --test only when the draft run should stay off the shared timeline.
The guide also lists the commands that spend credits (explore, generate, profile add, profile fix, run, ask, and report --rca), where local files live, what leaves the machine, and how to inspect status at any point. It reads command metadata and state; it does not invoke the target or spend credits.
/help
Use /help to list the current command surface or inspect one command in full.
Syntax
/help
/help <command>
From a shell:
rook help
rook help <command>
The overview groups commands into the testing sequence, workspace operations, and session/product operations. Command-specific help lists subcommands separately from cumulative options.
Help, slash-command completion, and shell parsing are derived from the same command registry. A renamed flag therefore changes all three surfaces together.
/docs
Use /docs to print and open the Rook documentation site.
Syntax
/docs
/docs --no-open
The shell form is rook docs with the same option.
--no-open prints the URL without launching a browser. The command does not require a selected project, invoke an agent, or spend credits.
/clear
Enter /clear in the TUI to clear retained command-output state and return to the prompt.
/clear
This does not delete projects, agents, scenarios, profiles, runs, evidence, credentials, variables, permissions, or authentication. The startup context remains because it belongs to the current TUI session. Output already committed to terminal scrollback can remain visible until the terminal itself is cleared or Rook is restarted.
There is no rook clear shell command.
/exit
Enter /exit or /quit at an idle TUI prompt to close Rook.
/exit
/quit
Rook waits for pending job-end records to settle and stops any local evidence viewer owned by the session. Exiting does not log out or delete project files.
While a command is running, press Esc to request an orderly interruption. Completed scenario evidence is preserved. Pressing Ctrl+C exits the TUI and also runs the exit bookkeeping path.
There is no rook exit shell command.
Account and Authentication
/login
Use /login when Rook has no stored credential or the existing token is invalid.
Syntax
Interactive:
/login
Headless launcher:
rook login
Choose the Environment
Public packages default to production. Use ROOK_ENV=prod with the hosted Web UI. Browser sessions and CLI credentials are separate; use the same account and environment in both.
Unattended Authentication
Rook accepts LT_USERNAME and LT_ACCESS_KEY from the shell or CI secret manager. When both are present, operations use them ahead of any stored browser login. Supplying only one is an error.
rook login also accepts --username, --access-key, and --oauth. Prefer secret-manager environment injection over literal command arguments. To use a stored OAuth account consistently, unset both LT variables in that terminal; forcing OAuth login does not stop exported credentials taking precedence in later commands.
Step-by-step
- Run
/loginorrook login. - Complete the TestMu AI flow in the browser.
- Return to the terminal.
- Verify the identity with
/auth statusorrook whoami.
State and security
Successful login stores credentials in the global Rook home, not the project directory. Do not copy that state into a repository or share it between users.
If a browser cannot open, follow the URL or instruction printed by the command. Do not paste login callbacks or tokens into tickets or screenshots.
Common problems
- Browser opens with the wrong account: sign out there or use a separate browser profile, then retry.
- Status still invalid: check
ROOK_ENVand whether exported LT credentials override the stored login before signing in again. - Controller unreachable: diagnose network and environment with
rook doctor.