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

Pattern	Detected By	Example
layer-separation	Consistent data/ui/domain separation	Repository + ViewModel + Composable
dependency-injection	Koin module patterns	factoryOf, viewModel
navigation-pattern	Compose Navigation usage	NavHost with routes
state-management	MVI/MVVM consistency	StateFlow + sealed classes

Problem-Solution Patterns

Pattern	Detected By	Example
error-boundary	Try-catch with UI feedback	Error state in Composable
loading-state	isLoading + Content pattern	Box with progress
pagination	LazyColumn with Pager	Paging 3 integration
caching-strategy	Repository layer caching	Cached repository pattern

Code Organization Patterns

Pattern	Detected By	Example
feature-module	Self-contained feature folders	feature/auth/ structure
shared-UI	Reusable Composables	ui/components/
test-mirroring	Test structure matching src	Parallel test folders
naming-convention	Consistent naming patterns	XxxViewModel, XxxScreen

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

Aspect	V1	V2
Trigger	Code write	Session observation
Scope	Single file	Cross-file patterns
Timing	Immediate	End of session
Focus	Code patterns	Architectural patterns

---

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