Self Reflection (1)

**Judgement:** A well-designed meta-skill for adding structured self-critique, mistake logging, and pattern-based learning. Particularly valuable for agents doing technical, multi-step, or high-stakes work where repeated errors are costly. **What it does well** - Enforces a short, explicit pre-delivery QA loop using 7 clear evaluation dimensions. - Treats user corrections as training data: logs misses, analyzes root causes, and creates prevention rules. - Tracks patterns across sessions (categories, frequencies, streaks) and proactively surfaces lessons before similar tasks. - Uses transparent, file-based memory under `~/reflection/` so behavior and history are auditable. **Key risks / limitations** - Adds cognitive and latency overhead; not ideal for fast, low-stakes chat or real-time interactions. - Effectiveness depends on the agent actually following the reflection and logging protocol consistently. - Pattern detection quality is only as good as the granularity and honesty of reflections and categories. - Local logs may accumulate sensitive contextual details; users should treat `~/reflection/` as potentially sensitive data. **Recommended scenarios** - Coding, architecture, or data-pipeline agents where regressions and repeated mistakes are common. - Analytical, research, or strategy assistants where user corrections carry important, reusable lessons. - Any "production"-oriented agent that should behave more like a careful reviewer than a one-shot responder.