mobile-memory

Mobile Memory Skill

Persistent memory system that maintains mobile development context across sessions.

Purpose

Unlike instincts (which capture patterns), memory retains factual project state:

• What modules exist
• What dependencies are installed
• What the architecture looks like
• What tests cover what code

This survives session breaks and compaction.

Memory Types

Project Structure Memory

Remembers your Android project layout:

{
    "modules": ["app", "core:network", "feature:auth"],
    "buildVariants": ["debug", "release", "staging"],
    "featureModules": ["auth", "home", "profile"]
}

Use when: Starting work on a new feature, need to know project layout

Dependencies Memory

Tracks all Gradle dependencies:

{
    "libraries": [
        {"name": "compose-runtime", "group": "androidx.compose", "version": "1.5.0"}
    ],
    "kgpVersion": "1.9.20",
    "gradleVersion": "8.2"
}

Use when: Adding new dependencies, checking compatibility

Architecture Memory

Documents your architecture patterns:

{
    "pattern": "mvi",
    "uiLayer": {"screens": ["Home", "Profile"]},
    "dataLayer": {"repositories": ["UserRepository"]},
    "di": {"framework": "koin", "modules": ["appModule"]}
}

Use when: Onboarding new developers, explaining codebase

Test Coverage Memory

Tracks test metrics:

{
    "totalCoverage": 78,
    "trend": "improving",
    "failingTests": [
        {"class": "AuthViewModelTest", "method": "testLogin"}
    ]
}

Use when: Planning testing work, tracking quality goals

Compose Screens Memory

Indexes all Composable screens:

{
    "screens": [
        {"name": "HomeScreen", "route": "home", "file": "HomeScreen.kt"}
    ]
}

Use when: Finding screens, understanding navigation

Usage

Load Memory

# At session start - load all memory
/memory-load all

# Load specific type
/memory-load project-structure
/memory-load dependencies

Save Memory

# Save current state
/memory-save project-structure
/memory-save test-coverage

# Save all (usually automatic)
/memory-save all

Query Memory

# Ask questions about project
/memory-query "What modules use Ktor?"
/memory-query "Which screens are not tested?"
/memory-query "What's the test coverage for auth module?"

Forget Memory

# Remove stale memory
/memory-forget recent-changes
/memory-forget --older-than 90days

Summary

# Get overview of all memory
/memory-summary

Memory Refresh Triggers

Memory auto-refreshes on:

• Gradle sync: Dependencies, build variants
• File changes: Recent changes, architecture
• Test runs: Test coverage, failing tests
• Session start: Load all memory
• Session end: Save all memory

Memory vs Instincts

Aspect	Memory	Instincts
Content	Factual state	Patterns
Examples	Module list, deps	"Use collectAsStateWithLifecycle"
Updates	On changes	On observations
Confidence	Binary (exists/doesn't)	0.0-1.0 score
Retention	30-90 days	Persistent

Integration

With Checkpoints

Checkpoints include memory state:

{
    "checkpoint": {
        "memory": {
            "project-structure": {...},
            "dependencies": {...}
        }
    }
}

Restoring a checkpoint restores memory too.

With Compaction

Memory survives compaction:

• Recent memory: Kept as-is
• Old memory: Summarized
• Always available via /memory-query

With Instincts

Memory informs instinct extraction:

• Project structure → Where to look for patterns
• Dependencies → What frameworks are used
• Architecture → What patterns to expect

Best Practices

1. Let it auto-refresh: Memory updates automatically on hooks
2. Query, don't remember: Use /memory-query instead of asking user
3. Validate on load: Check memory matches actual project
4. Update after changes: Run /memory-save after major changes
5. Clean up: Use /memory-forget to remove stale data

Example Session

User: I need to add a new feature for user profiles

Agent: /memory-query project-structure
Response: Found modules: app, core:network, feature:auth
       No profile module exists.

Agent: /memory-query dependencies
Response: Using Compose 1.5.0, Ktor 2.3.0, Koin 3.4.0

Agent: Based on memory, I'll create feature:profile module
       following your existing architecture pattern.

---

Remember: Memory is about what exists, not what should be. That's what instincts are for.