What problem does it solve? AI agent sessions repeatedly make the same mistakes because corrections, reusable procedures, and user preferences are lost when a session ends. This Skill observes tool-using work, identifies durable lessons, and persists them through Prime Agent's continual harness so future sessions benefit. ## Core Features & Use Cases - Continuous observation: Loads at session start and silently watches for user corrections, repeated failures, reusable multi-step procedures, and skill gaps without blocking the user's task. - Single persistence path: Routes evidence-backed lessons through the native refine.run harness with explicit local/global scoping, avoiding duplicate or parallel observation databases. - Approval-gated skill maintenance: Turns strong observations into tested proposal commits on isolated Git branches, with validation, secret scans, and explicit user approval before merging or pushing. - Use Case: A user corrects the agent twice about a deployment preference; the observer schedules a global refinement so the preference persists across all future sessions. ## Quick Start Load the task-observer skill at the start of any tool-using session and let it monitor for reusable lessons automatically.
