kookr-session-reflectlisted

# Session Reflection Workflow ## When to Use - After a supervision session where the user experienced repeated friction (had to ask agents the same thing multiple times, manually intervened often, or felt the system wasn't proactive enough) - When the user triggers the "Reflect" action from the Kookr UI - When reviewing whether Kookr's anomaly detection is catching the right things - When tuning agent launch templates or supervision policies based on observed patterns ## Core Principle: Friction is Signal Every time a user has to manually intervene, ask a question, or correct an agent, it reveals a gap between what the system does automatically and what the user needs. Session reflection turns these friction events into concrete improvements. ``` Session Events (hooks, anomalies, user input, queue transitions) ↓ Reflection Analysis (pattern detection + LLM summarization) ↓ Findings (friction patterns, root causes, improvement suggestions) ↓ Actions (new anomaly rules, prompt templates, CLAUDE.md updates) ``` ## Friction Pattern Taxonomy ### Category 1: Reactive User (user asks what the system should have surfaced) | Signal | Example | Improvement | |--------|---------|-------------| | User sends question-shaped input | "did you do X?", "are you done?", "what happened?" | Agent should proactively report status at milestones — create/update a skill with milestone reporting patterns | | User checks agent that has no anomaly | Clicks into agent with no finding | Mis