| name | deep-improve |
| description | Self-improving skill that analyzes pipeline history, feedback, and recurring issues to propose concrete improvements to skills and agents. Runs periodically.
|
Deep Improve Skill
Analyze pipeline history and propose concrete improvements.
Invocation
/deep-improve # Analyze recent pipeline runs and suggest improvements
Step 1: Gather Data
Read from:
docs/plans/*/insights.md — Improvement suggestions from pipeline runs
docs/feedback/FEEDBACK_INDEX.md — All user corrections
docs/backlog/out-of-scope-issues.md — Recurring issue categories
- Current skill and agent files
Step 2: Analyze Patterns
Identify:
- Frequently repeated mistakes
- User corrections that keep happening
- Out-of-scope issues that should be in-scope
- Skill instructions that are consistently misinterpreted
Step 3: Propose Improvements
For each pattern, propose:
- Which skill/agent to modify
- What specific instruction to add/change/remove
- Why it will help
- Risk of the change
Discussion checkpoint: Present proposals. Only apply approved changes.
Step 4: Apply
Amend the approved skill/agent files with [IMPROVED] markers.