Kapso plugin
Leeroo plugin for Claude Code and Codex: experiment campaigns scored by a judge the agent cannot tamper with.
- 153 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- A campaign launches agents on its own that edit code on separate git branches and run commands on the machine.
- One campaign can run for hours and consume `claude` and `codex` subscription limits.
- The `deploy` verb ships the winner, and the run itself reads keys from `.env`.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/kapsoDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- A campaign launches agents on its own that edit code on separate git branches and run commands on the machine.
- One campaign can run for hours and consume `claude` and `codex` subscription limits.
- The `deploy` verb ships the winner, and the run itself reads keys from `.env`.
This is third-party code. Review the repository files before installing.
What it does
Kapso runs campaigns: the system proposes a candidate, implements it on its own git branch, scores it with a judge, keeps the best and repeats. The agent has no access to the scoring, and every step is traced. The plugin brings ten commands, six skills and an MCP server for steering a campaign from a Claude Code, Codex or Cursor session. `/kapso:setup` installs and verifies the environment, and `/kapso:build <what you want>` launches the first campaign. The `kapso-build`, `kapso-evaluation`, `kapso-harness`, `kapso-review` and `kapso-setup` skills cover stating the goal, evaluation, a harness for your domain and reviewing a result. `kapso-mcp` gives the session web research, repository memory and experiment history.
Who it is for. ML engineers and developers with a measurable goal, such as a validation metric, who are ready to hand solution search to agents.
Good fit when
- A metric needs improving and an independent judge can score it.
- A campaign runs for hours and someone follows it, telling a slow run from a dead one.
- A campaign winner is reviewed and ready to deploy.
- Your own domain calls for a custom harness with evaluation.
Not a fit when
- The task is an ordinary code edit with no numeric goal.
- You lack subscriptions or access to the `claude` and `codex` CLIs: there is no direct API fallback.
- No budget or time is set aside for campaigns that last tens of minutes to hours.
Example request
/kapso:build Raise validation classification accuracy from 0.81 to 0.85, scoring with the `eval.py` script.Limitations
Every model call goes through an agent CLI: `claude` and `codex` must be installed and logged in, with no direct API fallback. Node.js 18 and Python 3.10 are required. The package is `leeroo-kapso`: the PyPI package named `kapso` is an unrelated WhatsApp tool that shadows the command and must be removed. The `LEEROOPEDIA_API_KEY` key connects a paid knowledge base, with a $20 starting credit on new accounts. A campaign runs from tens of minutes to hours.
How to disable. Remove the plugin with `claude plugin uninstall kapso@kapso` and the package with `pip uninstall leeroo-kapso`. Stop active campaigns with `kapso stop`.
FAQ
How do I install the Kapso plugin in Claude Code?
Run `claude plugin marketplace add Leeroo-AI/kapso && claude plugin install kapso@kapso`, then `/kapso:setup` verifies the environment.
How do I install Kapso in Codex?
With `codex plugin marketplace add Leeroo-AI/kapso && codex plugin add kapso@kapso`. In Codex use `$kapso-setup` and `$kapso-build` instead of slash commands.
How do I follow a campaign?
Launch with `--detach`: the command returns the pid, log path and a `kapso watch` line at once. `kapso inbox` helps answer a campaign that waits on you.
Does Kapso work without Leeroopedia?
Yes, the key is optional. It connects a hosted knowledge base of ML frameworks and is enabled with `--experience leeroopedia`.
How do I check the setup?
`kapso doctor` reports what your config needs and names the exact fix. `kapso doctor build` narrows the check.
Discussion
No questions or reviews yet.
Related
PluginsA Claude Code plugin with hooks and 26 skills: measures what the agent knows, blocks edits until code is read, keeps memory
cap-evolve edits an agent's prompts, tools and skills from failed traces and keeps an edit only if it wins on a held-out split
Multi-agent orchestration for Claude Code: teams, roles, autopilot and task routing across skills and subagents
A plugin that syncs Claude Code, Codex CLI and Antigravity on one project and lets them hand work to each other.
A Claude Code plugin that runs one task across several external models and surfaces disagreements before you ship
A Claude Code plugin that hands coding tasks to Codex agents, monitors them and verifies the results independently
README in short
Reports results on the RelBench and MLE-Bench benchmarks, then covers installing the plugin, the agent CLIs and the `leeroo-kapso` package. Next come the `kapso doctor` check, usage examples, campaign structure and documentation links.
README badge
Original SKILL.md
--- name: kapso description: Operate Kapso (PyPI leeroo-kapso), which turns coding-agent output into verified outcomes against a measurable objective, from a coding session — launch a build and hand it off, follow it with kapso watch, answer a campaign that is WAITING ON YOU, tell a slow run from a dead one and stop or resume it, learn from a finished campaign into your bank, serve a build or this session the experience it should start from, deploy the winner, and run the harness and experience verbs. Use when the user mentions Kapso or kapso build, watch, inbox, resume, stop, learn, experience, harness, deploy or doctor, asks how a campaign is doing, or says a run died. Do not use for ordinary code edits, for writing the objective and its evaluation (kapso-build does that), for reading a finished run's result (kapso-review does that), or inside a Kapso campaign session, which never launches another campaign. --- # Kapso Kapso runs experiment campaigns: propose a candidate, implement it on its own git branch, score it with the judge, keep the best, repeat. These facts are written for leeroo-kapso 0.8, the version this plugin pins. Prefer them and the docs links at the end over reading the package source; a session that greps it runs out of turns. ## Facts that prevent the common mistakes - Install with `pip install leeroo-kapso` (or `uv tool install leeroo-kapso`). The PyPI package named `kapso` is an unrelated WhatsApp tool that shadows the `kapso` command. Python 3.10+. The `kapso-setup` skill does the whole install. - Every model call runs through a coding-agent CLI that must be logged in: `claude` (ideation, implementation, judging) and `codex` (research, utilities). There is no API-key fallback. `OPENAI_API_KEY` is optional and only ever embeds