leeif

Use when executing an implementation plan with a contract — runs a RALPH loop per task with evaluator feedback, retries, and model escalation

Use when verifying implementation quality against a contract — dispatched as read-only evaluator in the easyharness-develop RALPH loop, or standalone for post-implementation audits

Use when starting a new feature or project — explores requirements, designs architecture, decomposes into tasks, and generates verifiable contracts for each subtask