reflect

To capture conversation learnings, you analyze work sessions to extract corrections, praises, and edge cases. The system logs these as HIGH, MED, or LOW confidence signals to prevent repeating errors and preserve successful patterns for future guidance.

Team-memory organizes cross-agent knowledge by persisting conversation learnings into structured history and decisions files. This allows agents to share a unified memory of corrections and best practices to guide future decisions consistently across sessions.

You log corrections and edge cases by applying the capture process during active work sessions. It detects HIGH/MED/LOW signals from conversations and persists results directly to agent-specific history files and shared decisions files for cross-agent learning.

Yes, the tool integrates with existing .squad files. It detects conversation signals and persists extracted learnings directly to the .squad/agents/{agent}/history.md and .squad/decisions.md paths to maintain consistent cross-agent guidance.

The best way to preserve successful patterns is to systematically log high-confidence corrections and medium-to-low confidence edge cases. This creates a persistent team memory that updates agent history and decisions, guiding future actions and preventing repeated errors.

You should not use automated conversation learning when your workflow lacks structured .squad files for persistence, or when conversations do not contain distinct corrections, praises, or edge cases suitable for extracting actionable HIGH/MED/LOW confidence signals.