What problem does it solve?
Managing self-improving agent harnesses requires a controlled way to review, approve, and roll back automatically generated configuration updates without letting the CLI prompt users directly or modify protected files.
Core Features & Use Cases
- Proposal Payload Generation: Runs
moai harness apply to fetch Tier 4 auto-update proposals as structured JSON payloads for the orchestrator to surface via AskUserQuestion.
- Apply/Rollback Orchestration: Coordinates approval decisions, executes approved updates through a 5-Layer Safety Pipeline, and restores snapshots via
moai harness rollback.
- Learning Lifecycle Control: Inspects tier distribution and rate limits with
moai harness status, and disables learning via moai harness disable while preserving YAML formatting.
- Use Case: When the harness observes repeated usage patterns and proposes updating a user-area skill's triggers, this skill fetches the proposal, hands it to the orchestrator for user approval, and applies or discards it safely.
Quick Start
Ask the agent to check harness learning status and review any pending auto-update proposals for approval or rollback.