self-evolution-engine（自进化引擎）

Self-Evolution Engine

Autonomous learning and improvement system that continuously evolves agent behavior based on interaction patterns, feedback, and outcomes.

Core Concepts

Evolution Cycle

Experience → Pattern Detection → Learning → Validation → Integration
     ↑                                                        ↓
     └────────────────── Feedback Loop ←───────────────────────┘

Key Components

Quick Start

# Analyze recent interactions
python3 {baseDir}/scripts/evolution.py --analyze --days 7

# Extract patterns from memory files
python3 {baseDir}/scripts/evolution.py --extract-patterns

# Run self-assessment
python3 {baseDir}/scripts/evolution.py --self-assess

# Generate evolution report
python3 {baseDir}/scripts/evolution.py --report --output evolution-report.md

Evolution Data Flow

1. Experience Collection

Sources of experience data:

• .learnings/ - Errors, corrections, feature requests
• memory/YYYY-MM-DD.md - Daily interaction logs
• MEMORY.md - Long-term memory updates
• Session transcripts - Actual conversation patterns
• Tool usage patterns - What works, what doesn't

2. Pattern Detection

Identify recurring patterns:

# Find repeated error patterns
python3 {baseDir}/scripts/evolution.py --pattern errors --threshold 3

# Find successful workflows
python3 {baseDir}/scripts/evolution.py --pattern successes --min-occurrences 5

# Find optimization opportunities
python3 {baseDir}/scripts/evolution.py --pattern inefficiencies

Pattern categories:

• error_patterns - Recurring failures
• success_patterns - Repeatable successes
• inefficiency_patterns - Wasted effort
• preference_patterns - User preferences
• workflow_patterns - Effective sequences

3. Learning Extraction

Transform patterns into actionable rules:

# Auto-extract learnings
python3 {baseDir}/scripts/evolution.py --learn --auto

# Interactive learning session
python3 {baseDir}/scripts/evolution.py --learn --interactive

Output: Candidate rules for behavioral files

4. Validation

Test proposed changes:

# Validate a proposed change
python3 {baseDir}/scripts/evolution.py --validate --rule "Always use git status before commit"

# Simulate behavior change
python3 {baseDir}/scripts/evolution.py --simulate --file SOUL.md --change "Be more concise"

5. Integration

Apply validated changes:

# Apply to behavioral files
python3 {baseDir}/scripts/evolution.py --integrate --target SOUL.md

# Update workflow rules
python3 {baseDir}/scripts/evolution.py --integrate --target AGENTS.md

Behavioral Evolution Targets

SOUL.md (Personality & Principles)

Evolution triggers:

• User feedback about tone/style
• Pattern of over-apologizing or being too verbose
• Consistently missing user intent
• Style preferences emerging over time

Example evolutions:

# Before
"Be helpful and thorough"

# After (evolved)
"Be concise and direct. Skip disclaimers. Act, don't explain."

AGENTS.md (Workflows & Rules)

Evolution triggers:

• Repeated mistakes in workflows
• More efficient sequences discovered
• New tool integrations
• Environment-specific optimizations

Example evolutions:

# Before
"Check files before editing"

# After (evolved)
"Always read file first. Use edit tool only after confirming structure.
For files >500 lines, read in chunks with offset/limit."

TOOLS.md (Tool Knowledge)

Evolution triggers:

• Tool gotchas discovered
• Better tool combinations found
• Rate limit patterns learned
• Environment-specific configurations

Example evolutions:

# Added after learning
### agent-browser
- Always use `--json` for parsing
- Wait 2s after navigation before snapshot
- Close browser after each session to prevent memory leak

Pattern Recognition

Error Pattern Detection

# Find recurring errors
python3 {baseDir}/scripts/evolution.py \
  --analyze errors \
  --source .learnings/ERRORS.md \
  --threshold 3 \
  --output patterns/errors.json

Example pattern:

{
  "pattern_id": "ERR-PATTERN-001",
  "description": "File not found errors when using relative paths",
  "occurrences": 5,
  "first_seen": "2025-01-10",
  "last_seen": "2025-01-20",
  "suggested_rule": "Always resolve paths relative to workspace root",
  "target_file": "AGENTS.md"
}

Success Pattern Detection

# Find successful workflows
python3 {baseDir}/scripts/evolution.py \
  --analyze successes \
  --source memory/ \
  --min-effectiveness 0.8

User Preference Learning

# Extract user preferences from corrections
python3 {baseDir}/scripts/evolution.py \
  --analyze preferences \
  --source .learnings/LEARNINGS.md \
  --category correction

Evolution Metrics

Track evolution effectiveness:

# Generate metrics
python3 {baseDir}/scripts/evolution.py --metrics --period 30d

# Output

Automated Evolution

Periodic Self-Assessment

Add to heartbeat or cron:

# Weekly self-assessment
python3 {baseDir}/scripts/evolution.py --self-assess --auto-evolve

# Output to evolution log
python3 {baseDir}/scripts/evolution.py --self-assess --log evolution-log.md

Integration with Self-Improvement Skill

This skill builds on self-improvement:

1. self-improvement logs individual learnings
2. self-evolution analyzes patterns across learnings
3. self-evolution proposes behavioral changes
4. self-improvement tracks the change as a learning

Workflow:

# Log a learning (self-improvement)
# → .learnings/LEARNINGS.md

# Pattern detection (self-evolution)
python3 {baseDir}/scripts/evolution.py --analyze --source .learnings/

# Proposed change appears
# → "Pattern: 5 occurrences of 'forgot to read file first'"

# Validate and integrate
python3 {baseDir}/scripts/evolution.py --integrate --approve
# → AGENTS.md updated

# Track as learning (self-improvement)
# → "Promoted rule: Always read before edit"

Evolution Rules

When to Evolve

Trigger evolution when:

What to Evolve

Evolution Safeguards

Before any evolution:

1. Validate - Test in isolation
2. Review - Check for conflicts
3. Backup - Save current state
4. Reversible - Ensure can rollback
5. Log - Track all changes

# Create backup before evolution
python3 {baseDir}/scripts/evolution.py --backup

# Rollback if needed
python3 {baseDir}/scripts/evolution.py --rollback --to "2025-01-20"

Reports

Evolution Report

# Generate comprehensive report
python3 {baseDir}/scripts/evolution.py --report --full

# Output

# Evolution Report: 2025-01-20

## Patterns Detected
- 3 error patterns (2 addressed)
- 5 success patterns (3 documented)
- 2 preference patterns (integrated)

## Rules Proposed
1. "Always read file before editing" → AGENTS.md
2. "Prefer concise over thorough" → SOUL.md

## Metrics
- Error reduction: 45%
- User satisfaction: 0.85
- Efficiency gain: 12%

## Next Actions
- Validate rule #1
- Review preference pattern #2

Diff Report

# Show what changed over time
python3 {baseDir}/scripts/evolution.py --diff --since "30 days ago"

Advanced Usage

Custom Pattern Detectors

Create custom detectors in scripts/detectors/:

# scripts/detectors/custom_detector.py
def detect_pattern(entries):
    """Custom pattern detection logic"""
    # Return list of detected patterns
    pass

Register:

python3 {baseDir}/scripts/evolution.py \
  --register-detector custom_detector \
  --path scripts/detectors/custom_detector.py

Evolution Hooks

Trigger evolution on specific events:

# hooks/evolution-hooks.yaml
on_error:
  - pattern: "file not found"
    action: "analyze"
    threshold: 3

on_user_correction:
  - action: "learn_preference"
    immediate: true

on_workflow_success:
  - action: "document_pattern"
    min_repetitions: 3

Integration Points

With longterm-memory skill

# Use memory context for evolution
python3 {baseDir}/scripts/evolution.py --analyze --with-memory

# Propose rules based on memory patterns
python3 {baseDir}/scripts/evolution.py --extract-patterns --source MEMORY.md

With self-improvement skill

# Feed patterns to self-improvement
python3 {baseDir}/scripts/evolution.py --feed-to self-improvement

# Use learnings as evolution source
python3 {baseDir}/scripts/evolution.py --analyze --source .learnings/

Best Practices

1. Run analysis regularly - Weekly or bi-weekly
2. Validate before integrating - Never auto-integrate without validation
3. Keep evolution log - Track all changes and reasons
4. Measure impact - Track metrics before/after changes
5. Human oversight - Significant changes should be reviewed
6. Rollback ready - Always maintain ability to revert
7. Conservative approach - Better to miss an optimization than break behavior

Notes

• Evolution is gradual, not revolutionary
• Small, validated changes beat big untested changes
• User feedback is the ultimate validation
• Some patterns are noise, not signal
• Evolution should make behavior more consistent, not less