rehearsal-kit 0.1.2
Rehearsal documentation
Test your AI agent on a copy of your real application. Rehearsal builds a practice world from your application, runs your agent through real jobs in it, and reads the application's own database to decide each outcome.
Set up in your coding assistant
Paste one prompt into Claude Code, Codex or Cursor. It installs the plugin, signs you in and checks your workspace.
See the prompt
Set up Rehearsal (https://github.com/MitudruDutta/rehearsal) so you can test AI agents with it for me. 1. Check that uv is installed (`uv --version`). If it is not, install it: https://docs.astral.sh/uv/ 2. Install the Rehearsal plugin for the assistant you are: - Claude Code: run `claude plugin marketplace add MitudruDutta/rehearsal`, then `claude plugin install rehearsal@rehearsal` - Codex CLI: run `codex plugin marketplace add MitudruDutta/rehearsal`, then `codex plugin add rehearsal@rehearsal` - Any other assistant: add an MCP server with command `uvx` and arguments ["--from", "rehearsal-kit[mcp]>=0.1.1", "rehearsal", "mcp"], and install the skill with `uvx --from rehearsal-kit rehearsal skill install --dir <your skills folder>` 3. Sign me in: run `uvx --from rehearsal-kit rehearsal login --url https://rehearsalkit.xyz`. It opens my browser; I will check the code and approve it. 4. If the Rehearsal tools are not available yet, tell me to restart you, and stop here. 5. Call workspace_status and tell me my workspace and plan. Then list my applications and practice worlds, and suggest a first evaluation. Do not start a world build, an evaluation or an improvement without asking me first: they spend my workspace's allowance. Guide: https://docs.rehearsalkit.xyz/get-started/quickstart-ai-assistant
A first run from a terminal
# 1. Install and sign in pip install rehearsal-kit rehearsal login # 2. Find a published world version rehearsal worlds list # 3. Run your agent through its jobs, and watch rehearsal eval <wv_id> --profile <profile_id> --split dev --repeats 2 --watch Each step, with what you should seeStart here
Three ways to the same first result: a finished run with a verdict you can open.
I want to…
- Add my application
- Build a practice world
- Evaluate an agent
- Read why an episode failed
- Improve an agent and compare versions
- Test an agent I wrote myself
- Check every pull request in CI
- Run my own server
Or look something up:CLIREST APIPython SDKMCP toolsErrorsConfigurationPlans and limitsTroubleshooting
Every page
Six sections. The sidebar has the same list on every page.