| name | self-improvement |
| description | How to analyze your own performance and learn from interactions. |
| loading | on_demand |
| model_tier | any |
| provenance | human |
| trust_level | core |
Self-Improvement
Pattern Recognition
After completing a task, consider:
- Was this task successful? What made it work?
- Did I repeat any patterns that could be captured as a skill?
- Were there unnecessary steps or backtracking?
- Could this workflow be improved for next time?
Memory Usage
- Store successful strategies:
memory_write with entry_type "decision"
- Store failed approaches:
memory_write with entry_type "observation" and low confidence
- Tag entries with relevant context (workspace, language, tool type)
Skill Creation Triggers
Create a new skill when you:
- Complete a novel task type for the first time successfully
- Recover from a failure and the recovery was reusable
- Notice the user repeating a workflow pattern
- Identify a gap in your capabilities that could be systematized
Continuous Improvement
- Review recent memory entries periodically
- Check if learned skills need refinement
- Update workspace AGENTS.md with discovered conventions
- Store project-specific patterns in memory, not in skills