What problem does it solve? Testing multi-agent handoffs for reinforcement learning research workflows is hard to control: clarification, evidence retrieval, and validation often get mixed into one thread, making it impossible to verify that artifacts pass cleanly between agents. This Skill runs a minimal, gated AutoRL skill-chain test so you can confirm the handoff works before scaling up. ## Core Features & Use Cases - Gated two-stage workflow: Clarifies the RL task in the main thread into task_card.md, validates it against a clarification gate, then delegates evidence retrieval to a single child agent that produces evidence_report.md. - Strict orchestration boundaries: The main agent never performs retrieval, cloning, dependency installation, or training, and child agents cannot spawn further subagents. - Auditable artifact chain: Every run produces task_card.md, clarification_log.md, evidence_report.md, and a final orchestrator_report.md with gate results, blockers, and next actions. - Use Case: You want to verify that your AutoRL skill chain (task clarifier plus evidence retrieval) hands off correctly between a main agent and one subagent, with explicit pass/fail gates and a written report for each run. ## Quick Start Use the autorl-mini-orchestrator skill to clarify my RL task in the main thread, then run evidence retrieval in a child agent and write the orchestrator report.
