rook CLI
A CLI that tests AI agents end to end: derives scenarios from code, runs them live and separates unverified from failed.
- 81 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Adversarial scenarios run against the live agent and its real actions can have side effects.
- Runs spend paid-plan credits and send data to the TestMu AI service.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/rookDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Adversarial scenarios run against the live agent and its real actions can have side effects.
- Runs spend paid-plan credits and send data to the TestMu AI service.
This is third-party code. Review the repository files before installing.
What it does
rook, a TestMu AI (formerly LambdaTest) product, tests how an agent behaves across workflows, tools and actions. It derives the tests itself: `/explore` reads the code and finds agents, `/generate` creates scenarios and `/run` executes them on the live agent. A `/profile add` profile describes how to invoke the agent: curl, a command, HTTP or MCP. Scenarios split into functional, non-functional (token economy, quality) and adversarial: prompt injection, jailbreak, data and PII exfiltration, hallucination and hijacking. Adversarial scenarios map to the OWASP Top 10 for LLM and agentic applications and to MITRE ATLAS. rook does not treat an agent account of its own actions as proof: it checks it against tool calls and side effects that your hooks report.
Who it is for. Teams preparing an AI agent for release who want to check its behaviour and prompt-injection resistance in CI.
Good fit when
- The agent creates tickets and writes files, and you check that it actually did, not just said so.
- Before release you need adversarial scenarios for jailbreak and personal-data leakage.
- You compare the pass rate between two runs and see which scenarios started failing.
- The check must plug into CI with a clear exit code.
Not a fit when
- You have no TestMu AI account with an Agent Assurance plan.
- The agent triggers real paid or irreversible actions and there is no isolated test environment.
- You need model-judged text quality without side-effect checks.
Example request
Use rook to test my agent against its refund policy and against prompt injection, then show the unverifiable criteria.Limitations
Without a TestMu AI account on an Agent Assurance plan rook refuses before the first model call. Exploring, generating, writing profiles and running all spend plan credits. On a plan without the cloud rook works locally: `rook sync` is refused and `rook run` keeps results with you. The npm package needs Node.js 20 or newer, while Homebrew and the install script bring their own runtime. Scenarios run against the live agent, so target actions must be approved in advance, as well as the credit spend.
How to disable. Uninstall the CLI with `npm uninstall -g @testmuai/rook`, `brew uninstall lambdatest/rook/rook` or by deleting `~/.local/bin/rook` and `~/.testmuai/rook-*/`. Sign-in and settings live in `~/.testmuai/rook/`, so delete that directory separately.
FAQ
How do I install rook?
The first option in the README is `npm install -g @testmuai/rook@latest`. There is also `brew install lambdatest/rook/rook` with the full tap name, and an install script that verifies the checksum.
What does an unverifiable result mean?
A criterion that rook could not confirm with evidence. It is counted separately, as neither pass nor fail, and a scenario with one cannot count as passed.
How do I connect rook to Claude Code or Codex?
Run `npx @testmuai/rook-skill`: the skill is installed user-wide for Claude Code, Codex and Gemini CLI. Then ask your agent to test your agent with rook.
How do I run a check without the cloud?
Skip `rook sync` and use `rook run --test`. On a plan without the cloud, such as Starter, this is the only form and results stay local.
Which scenarios does rook generate?
Three classes: functional, non-functional (token economy, quality) and adversarial. Together they make sixteen categories, from prompt injection to PII leakage.
Discussion
No questions or reviews yet.
Related
CLI toolsA local debugger for TypeScript agents: run step trees, CI trajectory checks and portable evidence bundles, with no account.
Snapshot tests for AI agents: records tool calls today and tells you when they silently change after an edit
Vercel CLI for testing coding agents on your framework: runs Claude Code, Codex and others and measures the success rate
A skill and MCP server for the Rhesis platform: the agent designs, runs and analyzes test suites for your own AI agent
A CLI for testing and red teaming LLM applications: compares models, runs automated CI checks and finds vulnerabilities
An MCP server that checks whether the agent is really done: drives the real app and returns pass, fail or couldn't tell with the file and line
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| rook (this card) | no | 81 | ||
| agentevals | no | not required | 169 | scores agent behaviour from OpenTelemetry traces; rook runs scenarios on the live agent and checks claimed actions against side effects |
| agent-qa | no | 856 | a QA agent for web and mobile apps; rook tests AI agents themselves, including adversarial scenarios mapped to OWASP and MITRE ATLAS |
README in short
A terminal tool with `/explore`, `/generate`, `/profile`, `/sync`, `/run` and `/ui` commands. Results appear per criterion with a quote, trend between runs and a list of what could not be checked.