What problem does it solve? Autonomous agents running long research or experimentation loops can corrupt their objectives, game their own metrics, lose state after context compaction, or take irreversible actions without approval. This Skill provides the design patterns for building a harness around an agent so it can run safely for extended periods. ## Core Features & Use Cases - Surface Classification: Define locked, editable, append-only, and human-controlled surfaces so agents cannot modify the evaluators that judge them. - Loop Patterns: Implement autoresearch-style optimization loops and research-to-skill pipelines with durable logs, novelty gates, pruning, and rollback. - Governance Boundaries: Separate PR preparation from merge authority and route destructive operations through explicit human approval. - Use Case: You are building a background agent that reads new engineering blog posts and drafts skill updates. Use this Skill to lock the source rubric, require retrieval evidence before citation, log rejected attempts, and have the agent prepare PRs that wait for human merge approval. ## Quick Start Ask the agent to design a harness for an autonomous research loop with locked evaluation rubrics, durable thread logs, and human approval for merges.
