continuous-learning-v2

name	continuous-learning-v2
description	Instinct-based learning system that observes sessions, creates atomic instincts with confidence scoring, and evolves them into skills/commands.

Continuous Learning v2

An advanced learning system based on atomic "instincts" with confidence scoring that evolve into skills and commands over time.

Philosophy

v2 improves on the original continuous learning with:

• Atomic instincts instead of full skills
• Confidence scoring (0.3-0.9 weighted)
• Observation hooks for reliable pattern detection
• Evolution path from instincts to skills/commands

Key Concepts

Instincts

Small, atomic learned behaviors with confidence scores:

{
  "id": "inst_abc123",
  "trigger": "typescript type error with 'any'",
  "behavior": "Suggest explicit typing instead of 'any' type",
  "confidence": 0.7,
  "domain": "code-style",
  "occurrences": 15,
  "last_used": "2025-01-15"
}

Confidence Scoring

• 0.3-0.4: New instinct, few observations
• 0.5-0.6: Moderate confidence, useful pattern
• 0.7-0.8: High confidence, consistently helpful
• 0.9: Ready for promotion to skill/command

Confidence increases with:

• Successful application
• User positive feedback
• Repeated occurrence

Confidence decreases with:

• User corrections
• Contradicting observations
• Disuse over time (decay)

Domain Tags

Instincts are tagged by domain:

• code-style - Formatting, naming, structure
• testing - Test patterns, coverage
• git - Version control workflows
• debugging - Error resolution
• performance - Optimization patterns
• security - Security practices

Evolution Path

Observation → Instinct → Cluster → Skill/Command

1. Observe: Hook captures pattern
2. Create: New instinct with low confidence
3. Reinforce: Confidence increases with use
4. Cluster: Related instincts group together
5. Promote: High-confidence clusters become skills/commands

Workflow

Creating Instincts

When a useful pattern is observed:

1. Check if similar instinct exists
2. If yes: increase confidence and update
3. If no: create new instinct with 0.4 confidence

Using Instincts

When working on a task:

1. Match active instincts to current context
2. Apply high-confidence instincts (>0.6)
3. Track success/failure of application

Promoting to Skills

When instinct cluster reaches threshold:

1. Group related instincts
2. Generate skill definition
3. Request user approval
4. Save as skill file

Best Practices

1. Start conservative - Low initial confidence prevents false positives
2. Let confidence build - Don't force promotion
3. Review periodically - Prune low-confidence instincts
4. Domain separation - Keep instincts focused
5. Export/share - High-confidence instincts benefit teams