Domain Orchestrator — Productivity

Domain Orchestrator is a skill that helps users build, launch, grade, or schedule Claude Managed Agents (CMAs) in their own Anthropic account. It reads a session goal, routes deterministically to one of five phase sub-skills, and compiles the goal into an execution shape (single-pass workflow, bounded grade-iterate loop, or recurring cron deployment loop). All launches are emitted as BYOK curl commands for the user to run with their own API key.

agent-launcher claude-code

📂 View skill on GitHub ↗ 🚀 How to use

✨ What it does

Reads the per-session goal from ./my-agent/goal.json.
Routes deterministically to one of five phase sub-skills via goal_router.py.
Compiles the goal+phase into an execution shape via loop_compiler.py.
Forks context to keep heavy intake out of the parent thread.
Emits BYOK curl commands; no tool makes API calls.
Enforces pre-flight gates: no goal set, vague goal, no API calls, no key printing.

🎯 When to use it

When the user wants to build a Claude Managed Agent from scratch.
When the user wants to launch an existing agent as a managed agent.
When the user wants to run an agent on a schedule (e.g., nightly worker).
When the user wants to grade an agent against a rubric and iterate.
When the user wants to set up a recurring deployment loop.

🚀 How to use

Trigger: /cs:goal

Trigger by asking to build, launch, grade, or schedule a Claude Managed Agent, e.g., "build me an agent" or "run this on a schedule". The skill expects a session goal (one sentence for one CMA) stored in ./my-agent/goal.json. If no goal is set, run goal_state.py init --goal "..." first. The orchestrator routes to the appropriate sub-skill and compiles the loop. Example prompts:

build me an agent that summarizes my daily emails
launch this as a managed agent and run it every morning at 9am

📄 Output: A compiled plan.v1 execution shape and a digest of the phase outcome, with artifact paths and next steps.

📦 Add this skill to Claude Code

# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills

# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/agent-launcher/skills/agent-launcher-orchestrator .claude/skills/agent-launcher-orchestrator

Skill source: agent-launcher/skills/agent-launcher-orchestrator/SKILL.md

⚠️ Good to know

Requires a clear one-sentence goal; refuses to route vague or multi-job goals. No API calls are made; the user must run the emitted BYOK curl with their own key.

❓ FAQ

What happens if no goal is set?

The orchestrator refuses to proceed and instructs you to run goal_state.py init --goal "..." to set a goal first.

Can the skill make API calls to launch the agent?

No, the skill never makes API calls. It emits BYOK curl commands that you run with your own $ANTHROPIC_API_KEY.

How does the skill handle a vague goal?

The router exits with code 4 (REFUSE) and asks you to provide one sentence naming the one job before routing.

🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.