Guide · Agent skill
How to use the Stage and Launch Managed Agent agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The stage-launch skill is the second step in setting up a Claude Managed Agent. It takes a validated build sheet from the planning phase and produces the exact JSON payloads the Anthropic API expects, plus a launch.sh script that creates the environment, agent, session and first message in order. After this guide you will know how to install the skill, generate and validate the payloads, and run the launch script with your own Anthropic API key.
Stage and Launch Managed Agentalirezarezvani/claude-skills · MIT Download the skill Quick start
- 1Install the skill. Unzip the downloaded ZIP into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills.
- 2Prepare the build sheet. Confirm you have a validated build sheet JSON file from the planning phase.
- 3Generate payloads. Run payload_generator.py with --sheet and --out-dir to create the four ordered payload files.
- 4Write the launch script. Run launch_script_writer.py with --out-dir to create launch.sh.
- 5Validate. Run payload_validator.py --dir and fix any FAIL, especially a key_leak finding.
- 6Set your key. In your own shell, export ANTHROPIC_API_KEY. Do not paste it into chat or a file.
- 7Run launch.sh. Run ./launch.sh and watch the first poll. If it stops, re-run it to resume from the last created ID.
What the stage-launch skill does
An agent skill is a folder of instructions and scripts that an AI assistant such as Claude can load and follow when your request matches the skill's description. This one covers Phase 2 of building a Claude Managed Agent: turning a plan into something runnable.
The skill has three scripts. payload_generator.py turns your build sheet into four ordered payloads. launch_script_writer.py writes launch.sh. payload_validator.py runs a pre-launch check that includes a scan for accidentally embedded API keys. The skill itself makes no network calls; you run launch.sh yourself.
- Generates payloads for environment, agent, session and kickoff
- Sets agent toolset permissions to always_allow and MCP toolsets to always_ask
- Writes a launch script that reads $ANTHROPIC_API_KEY at runtime and never embeds it
- Resumes on re-run, skipping any step whose ID file already exists
Install the stage-launch skill
Download the ZIP for this skill from the directory page and unzip it. You can install it for yourself or for one project.
For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. Alternatively, upload the skill in the Claude apps under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project only: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload
Check what you need before starting
You need Python 3 (standard library only), a Bash shell for launch.sh, and an Anthropic API key set as ANTHROPIC_API_KEY in your own shell. You also need a validated build sheet JSON file from the earlier planning phase, and access to the Claude Managed Agents beta, because the script sends a managed-agents beta header.
Running the launch step uses your own API key and will incur Anthropic API usage costs.
- Python 3
- Bash shell
- ANTHROPIC_API_KEY set in your shell
- A validated build-sheet JSON file
- Access to the Claude Managed Agents beta
Generate and validate the payloads
Run the three scripts in order from your project folder. The generator writes four files into a payloads folder. The validator must pass before you launch; a FAIL blocks the launch, especially a key_leak finding, so fix it and re-run.
- Generator output: payloads/01-environment.json, 02-agent.json, 03-session.json, 04-kickoff.json
python3 scripts/payload_generator.py --sheet ./my-agent/build-sheet.json --out-dir ./my-agent
python3 scripts/launch_script_writer.py --out-dir ./my-agent
python3 scripts/payload_validator.py --dir ./my-agentSet your key and run launch.sh
Set your key in your own terminal, not in the chat. You can check whether it is already present without printing it. Then run the launch script. It creates the environment, agent, session and kickoff in that order, and chains the IDs between steps.
If a run stops partway, run launch.sh again. It continues from the last created ID. Watch the first poll in the foreground before treating the launch as successful.
- Never paste the key into chat or write it to a file
- Launch order: environment → agent → session → kickoff
[ -n "$ANTHROPIC_API_KEY" ] && echo "key present" || echo "export ANTHROPIC_API_KEY=... first"
export ANTHROPIC_API_KEY=...
./my-agent/launch.shTrigger the skill with natural requests
You do not need special syntax. The skill's description matches requests such as "launch it", "deploy the agent" or "create the agent now". It is also used when the orchestrator routes the phase to stage-launch. Ask in plain words once your build sheet is ready.
The skill is separate from the planning phase (interview) and from the outcome loop (grade-iterate), so ask for planning or evaluation work through those skills.
- "Launch it" after your build sheet is validated
- "Deploy the agent now"
- "Create the agent now"
Safety notes and tips
The key never enters chat, a file, a payload or a log. launch.sh reads it from the environment, and payload_validator.py scans for sk-ant- patterns and fails the check if it finds one. Keep the MCP permission setting at always_ask unless you have a specific reason to change it.
Remember that the launch step calls the API under your account, so usage is billed to you. Review the payloads and the validator output before you run anything.
- Keep MCP toolsets on always_ask
- Run the validator again after any edit
- Do not run this skill without a validated build sheet
What people use it for
- Deploying a planned customer-support agent with a fixed set of tools and an approval step for MCP tools
- Replacing hand-written curl commands with a repeatable launch process you can re-run
- Checking a build sheet for accidentally embedded API keys before sharing the project folder
Frequently asked questions
How do I install the stage-launch skill?
Download the ZIP from the skill page and unzip it into ~/.claude/skills/ for personal use or into .claude/skills/ inside a project. You can also upload it in the Claude apps under Settings → Capabilities → Skills.
Is the stage-launch skill free?
The skill is licensed under MIT. Running the launch step calls the Anthropic API with your own key, so API usage is billed to your Anthropic account.
What do I need before using stage-launch?
You need Python 3, a Bash shell, an Anthropic API key set as ANTHROPIC_API_KEY, a validated build sheet JSON file from the planning phase, and access to the Claude Managed Agents beta.
Does stage-launch make network calls?
No. The skill's tools make no network calls. You run launch.sh yourself, and it sends requests to the API using your key from the environment.
Is my API key safe with this skill?
The launch script reads ANTHROPIC_API_KEY at runtime and never embeds it. payload_validator.py scans for sk-ant- patterns and fails the check if it finds a leaked key.
What should I do if launch.sh stops partway through?
Run launch.sh again. It resumes from the last created ID, because each step skips if its ID file already exists.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use stage-launch if you have already planned a Claude Managed Agent, have your own Anthropic API account, and are comfortable running shell commands in a terminal. Pick something else if you have no validated build sheet yet, want to avoid API usage costs, or prefer not to run scripts. Planning and outcome evaluation are handled by separate skills.
Download the skill