reflect

Capturing learnings from conversations involves identifying high, medium, and low confidence patterns to prevent repeating mistakes. You initiate a reflective session to extract these signals and route them to agent history files, decision inboxes, or skill notes.

Categorizing conversation learnings by confidence level involves assigning HIGH, MEDIUM, or LOW signals to extracted patterns. This guides future actions by distinguishing established decisions from uncertain observations, ensuring traceability through stored metadata.

Storing extracted decisions and history requires routing captured learnings to specific paths like `.squad/agents/{agent}/history.md` or creating new decision inbox items. Learnings are documented with metadata for traceability across sessions.

You need to document skill improvements from in-session learnings when conversations reveal successful patterns or operational gaps. Applying signals during a reflective session determines whether to append to history, create decisions, or update skill notes.

Capturing conversation memory works for multi-agent squad workflows by integrating learnings into per-agent history files and team decisions. This propagates successful patterns across sessions and prevents agents from repeating past mistakes.