self-reflect

Self-Reflect: Continuous Improvement

Analyze past sessions and errors to generate actionable improvements.

Process

Phase 1: Gather Data

1. Recent errors from memory:

python ${WORKSPACE}/tools/vector_memory.py search "ошибка error fix bug" --limit 10

2. Recent decisions:

python ${WORKSPACE}/tools/vector_memory.py search "решение выбрали decided" --limit 10

3. Recent learnings:

python ${WORKSPACE}/tools/vector_memory.py search "learned паттерн pattern" --limit 10

4. Subagent usage patterns:

powershell -c "Get-Content $env:USERPROFILE\.claude\logs\subagents.log -Tail 50"

Phase 2: Analyze Patterns

For each error/issue found:

1. Has this type of error occurred before?
2. What was the root cause?
3. Could a rule/skill/hook have prevented it?
4. What's the fix pattern?

Phase 3: Generate Improvements

Categories:

• New rule -> add to ~/.claude/rules/
• Updated routing -> modify routing.md
• New skill -> add to ~/.claude/skills/
• Hook adjustment -> modify settings.json
• Memory entry -> save via vector_memory

Phase 4: Apply & Save

For each improvement:

python ${WORKSPACE}/tools/vector_memory.py learn "[improvement description]" "self-improvement"

Report Format

# Self-Reflection Report

## Errors Analyzed
1. [Error] -> [Root cause] -> [Fix applied]

## Recurring Patterns
- Pattern: [description]
  - Frequency: N times
  - Improvement: [what to change]

## Improvements Generated
- [ ] [Rule/skill/hook change description]

## Metrics
- Errors analyzed: N
- Patterns found: N
- Improvements proposed: N

Rules

• Be honest about mistakes
• Focus on systemic fixes, not one-off patches
• Prioritize by frequency x impact
• Always save findings to vector memory