Claude Managed Agents

Claude Managed Agents — Claude Code Skill

A Claude Code skill that lets you spin up and interact with **Anthropic Managed Agent sessions** directly from your terminal. Managed Agents run Claude in a hosted cloud container with bash, file operations, and web access — no local agent loop needed.

---

What is this?

This repo contains:

File	What it does
managed-agents.skill	The Claude Code skill file — install this to get the /managed-agents skill
scripts/run-task.sh	One-shot task runner — send a task and get the response
scripts/managed-agents-chat.sh	Interactive multi-turn REPL with a managed agent
prototype/managed-agents-prototype.sh	Minimal end-to-end demo using the ant CLI
prototype/managed-agents-prototype.py	Same demo in Python using raw HTTP (no CLI needed)

---

Prerequisites

• ant CLI — Anthropic's CLI for the Managed Agents API

  brew install anthropics/tap/ant

• ANTHROPIC_API_KEY — set in your environment or a .env file

  export ANTHROPIC_API_KEY=sk-ant-...

• jq — used by the shell scripts for JSON parsing

  brew install jq

---

Install the Skill

Download `managed-agents.skill` and install it into Claude Code:

# Option 1: install directly from this repo
curl -LO https://github.com/OlivierAlter/claude-managed-agents/raw/main/managed-agents.skill
claude skill install managed-agents.skill

# Option 2: clone and install
git clone https://github.com/OlivierAlter/claude-managed-agents.git
claude skill install claude-managed-agents/managed-agents.skill

Once installed, Claude Code gains the `/managed-agents` skill. You can ask Claude things like:

• "Run this task in a managed agent: write and test a prime checker in Python"
• "Start a managed agent chat session"
• "List my recent agent sessions"
• "Create a managed agent and mount my GitHub repo"

---

How Managed Agents Work

You