What problem does it solve? Coordinating multiple AI sub-agents without a clear framework leads to context bloat, lost state between invocations, and uncontrolled execution. This Skill provides a structured orchestration framework for delegating tasks, managing memory across agent boundaries, and enforcing human oversight gates. ## Core Features & Use Cases - Delegation Sub-Agent Pattern: Define structured JSON payload contracts for worker agents with isolated context, clear boundaries, and machine-parseable handoffs. - Supervisor Execution Loops: Fan-out/fan-in dispatch of specialized workers with definitive state signals (COMPLETE, BLOCKED, ERROR) for controlled coordination. - Session State Management: Persist task state in task.md checklists and distill long-term knowledge into learnings.json injected on subsequent startups. - Human-in-the-Loop Gates: Mandatory approval gates before execution and recovery gates after repeated failures. - Use Case: When a user asks to "add OAuth and secure it", the Supervisor dispatches an authentication worker to build the implementation and a security-auditor worker to review it against OWASP, then synthesizes results before committing. ## Quick Start Ask the agent to break down a complex feature request into specialized sub-agent tasks with isolated context and tracked progress in a task file.
