session-retrospective

Capture lessons learned from an AI agent session by automatically extracting corrections and writing them to a persistent memory file. This ensures that future agents can read the lessons at session start to avoid repeating past mistakes.

Persistent memory works by storing session corrections and plan changes in a structured markdown file under the .ai/memory/ directory. Future agents access this file at startup to retrieve categorized lessons from previous interactions.

The best way to stop AI agents from repeating mistakes is to systematically document corrections and surprises in a persistent lessons file. This creates a structured knowledge base that future agents read before beginning new tasks.

Structure session retrospective files by writing to .ai/memory/lessons.md and categorizing entries according to defined topics. This workflow updates existing lessons and provides a quick reference for future tasks.

Session corrections are applied to the persistent memory file during ongoing sessions and immediately after plan changes. This ensures that lessons are captured in real-time and are available to future agents at session start.