reflectorlisted

# Reflector A system for getting better at what you do, systematically, over time. ## The Problem Most agents accumulate experience but don't learn from it. They make the same mistakes, miss the same patterns, and never refine their judgment. Memory files capture what happened. Reflector captures what it means. ## The Solution A feedback loop: ``` ACT -> OBSERVE -> EXTRACT -> REFINE -> MEASURE -> ACT ``` 1. **Act** - Do work normally 2. **Observe** - Daily review classifies outcomes from all channels 3. **Extract** - Identify patterns across accumulated observations 4. **Refine** - Update decision-making principles based on evidence 5. **Measure** - Track whether changes actually improve outcomes Improvement happens through better principles and sharper self-knowledge, not model retraining. ## Setup ### Step 1: Initialize Run from your workspace root: ```bash node /path/to/reflector/scripts/init-reflector.js ``` This creates: - `PRINCIPLES.md` - Your decision-making framework (if none exists) - `memory/reflector/outcomes.jsonl` - Structured outcome log - `memory/reflector/principles-history.jsonl` - Principle change tracking - `memory/reflector/weekly-summaries/` - Weekly review outputs Run `--help` for options, `--dry-run` to preview. ### Step 2: Create cron jobs Use the OpenClaw cron tool to create two jobs. The prompt text is in the `prompts/` directory - read the file and use its contents as the agentTurn message. **Daily review** (reads `prompts/daily-re