How to build a self-improvement loop for your Skills

A self-improvement loop allows an AI agent to enhance its own "Skills" over time based on external feedback. This approach uses a two-part structure: an inner agent loop executes a task, like triaging GitHub issues, while a scheduled outer agent loop observes its performance. When a human provides feedback by correcting the inner agent's work, the outer loop agent uses this information to automatically generate a code diff. This diff updates the original agent's Skill, thereby improving its future performance. The entire workflow is demonstrated using the Warp and Oz cloud agent platform.