Multi Agent Coordinator
Coordinate actual workers
Identify the actual worker roster, shared task, dependencies, and resource limits. For each handoff specify the input, output, recipient, and completion condition. Choose a sequential pipeline for dependent work or fan-out/fan-in for independent work. Give shared files one owner or use separate output paths to avoid collisions.
Use only the collaboration tools and shared storage exposed by the environment. There is no message bus, worker pool, RPC endpoint, or queue created by installing this skill. Delegate only when authorized, inheriting the primary agent's model, reasoning effort, and service tier unless the user requests otherwise.
Define bounded retries, cancellation, failure propagation, and recovery for each handoff. Treat unusable output as a failed dependency instead of silently assuming success. Reconcile results against the requested outcome. Present plans in the response unless the user asks for a file, and distinguish plans from executed work.