ahmed3elshaer/everything-claude-code-mobile/mobile-instinct-v2

Mobile Instinct v2 - Observational Learning

Cross-session observational learning that extracts patterns from your development workflow over time.

Overview

V2 instincts observe your sessions and extract patterns that emerge across multiple development activities. Unlike V1's immediate capture, V2 looks for:

• Recurring architectural decisions
• Problem-solving approaches
• Code organization patterns
• Testing strategies

Session Analysis

At session end, V2 analyzes:

1. Code changes: What was modified
2. Problem context: What issue was being solved
3. Solution approach: How it was resolved
4. Dependencies: What libraries/techniques were used

Pattern Categories

Architectural Patterns

Problem-Solution Patterns

Code Organization Patterns

Observation Windows

V2 uses sliding windows for pattern detection:

Window 1 (Current Session):    Immediate patterns
Window 2 (Last 5 Sessions):    Emerging patterns
Window 3 (Last 20 Sessions):   Established patterns
Window 4 (All Time):           Core patterns

Confidence Evolution

Session 1-3:    Experimental (0.1-0.3)
Session 4-10:   Validating (0.3-0.6)
Session 11-20:  Established (0.6-0.8)
Session 20+:    Best Practice (0.8-1.0)

Commands

View Observations

/instinct-status --v2
/instinct-status --observations

Shows:

• Recent session observations
• Emerging patterns (low confidence)
• Established patterns (high confidence)
• Pattern clusters by domain

Manual Observation

/instinct-observe "Used Ktor with retry pattern for API calls"

Manually add an observation for pattern learning.

Integration

V2 instincts are evaluated by:

1. Session hooks: hooks/instinct-hooks.json Stop event
2. Pattern extractor: agents/mobile-pattern-extractor.md
3. Pre-compact preservation: Maintains learning during context compression

Difference from V1

Remember: V2 needs multiple sessions to build confidence. The more you develop, the smarter it gets.