What problem does it solve? Managing a self-improving harness learning subsystem requires careful coordination between CLI commands, proposal payloads, and user approval flows. This Skill bridges the moai harness CLI and the orchestrator's AskUserQuestion interface so Tier 4 auto-update proposals are surfaced, approved, applied, or rolled back safely. ## Core Features & Use Cases - Proposal Payload Production: Runs moai harness apply to fetch pending Tier 4 auto-update proposals and emits a structured JSON payload (proposal_id, target_path, field_key, new_value, confidence, recommended_action) for orchestrator consumption. - Apply/Rollback Orchestration: Writes approval decisions to .moai/harness/proposals/, executes approved changes via moai harness apply --execute, and restores snapshots with moai harness rollback <date>. - Safety-Gated Lifecycle: Every auto-update passes a 5-Layer Safety Pipeline (Frozen Guard, Canary Check, Contradiction Detector, Rate Limiter, Human Oversight), with frozen paths like .claude/skills/moai-*/** never modified. - Use Case: After a week of harness observations, run moai harness status to see three pending proposals, fetch the next one, let the orchestrator ask the user to approve or reject it, and apply the approved change to a user-area hns-* skill description. ## Quick Start Ask the assistant to check harness learning status and surface any pending auto-update proposals for approval.