Anthropic’s launch-your-agent is a Claude Code skill to help founders build Claude Managed Agents, from interview to v0 scope, launch, grading, iteration, and optional scheduling. It includes a build folder and CMA primitives reference, licensed Apache-2.0 since 2026.
What it is
A Claude Code skill that helps a technical founder build Claude Managed Agents (CMA) — an internal worker, a piece of their product, a customer-facing agent. It interviews you about what you want to build, scopes a v0, launches it in your own account, grades it against your definition of done, iterates, and (if it should run on a clock) puts it on a scheduled deployment — with everything bigger laid out as an explicit v1/v2 plan.
Reference implementation. Not maintained and not accepting contributions. Licensed under Apache 2.0.
How it works
The repository contains a Claude Code skill with four phases: interview, stage & launch, grade & iterate, and run without you. It provides a set of CMA primitives and an examples bank, plus a companion wrap-up skill for status checks. The CMA documentation is cited as the source of truth for API details.
Getting started
Quickstart
git clone https://github.com/anthropics/launch-your-agent.git
cd launch-your-agent
claude
Then type:
/launch-your-agent
The skills in .claude/skills/ are picked up automatically when you run Claude Code inside this folder — nothing to install.
/wrap-up
### What you need
- Claude Code installed and signed in.
- An Anthropic API key for your own account (you'll create it during the flow at platform.claude.com → API keys; it goes into a local `.env` file, never into the chat). Runs cost cents.
### What you walk away with
- A live managed agent in your Console (agent + environment + graded run, plus a scheduled deployment if your task recurs).
- A `my-agent/` folder: the build sheet, the exact API payloads, a resumable launch script, an eval scaffold, an overview page, and a NEXT-DIRECTIONS.md laying out v1/v2.
## Repo layout
- `.claude/skills/launch-your-agent/` — The main skill: 4 phases (interview → stage & launch → grade & iterate → run without you) + references (interview mapping, verified API call shapes, examples bank, overview template)
- `.claude/skills/wrap-up/` — Companion skill: explicit close-out / status check for a built agent
- `cma-primitives.md` — Inventory of CMA primitives and limits, from the public docs
- `interview-to-config.md` — Background: how interview answers map to CMA primitives
- `examples-bank.md` — Sourced example agents and production proof points
- `ui/` — Example overview page + build sheet
The CMA documentation is the source of truth for the API: https://platform.claude.com/docs/en/managed-agents/overview