Until Done

Hand off the stop condition to the agent. You define what done looks like; the agent loops until verification passes or a stop rule fires.

Parse the goal

The user's argument is the goal. If empty, ask: "What should be true when this loop stops?"

Examples:

• "Fix the failing test in user_service_test.ts"
• "Implement issue #42 per the PRD"
• "Reduce bundle size below 500kb"
• "Research agent memory from vault until synthesis note, learning record, Context Index update, and decision trace exist"

Confirm verifier

Read .agent/context/loops.md first (fallback: docs/agents/loops.md). Confirm which verifier commands apply, or accept an override the user stated in the prompt.