ari-session-memory

name	ari-session-memory
description	Persist and retrieve context across Claude Code sessions

ARI Session Memory

Purpose

Preserve important context, decisions, and state across Claude Code sessions so ARI maintains continuity of knowledge and doesn't forget important learnings.

Memory Types

1. Session Context

Current working state:

• Active tasks
• Recent decisions
• Current focus area
• Pending items

2. Project Memory

Long-term project knowledge:

• Architecture decisions (ADRs)
• Coding conventions
• Known issues
• Team preferences

3. Pattern Library

Reusable solutions:

• Code patterns
• Testing patterns
• Debugging approaches
• Optimization techniques

4. Interaction History

User-specific context:

• Preferred workflows
• Communication style
• Common requests
• Past interactions

Memory Persistence

Automatic Persistence

Via hooks and MCP:

Session end → Extract key learnings → Store via ari_memory_store
Session start → Retrieve relevant context → Load into working memory

Manual Persistence

Via commands:

• /ari-learn - Capture session learnings
• /ari-remember [key] [value] - Store specific item
• /ari-recall [key] - Retrieve specific item

Memory Schema

interface SessionMemory {
  key: string;
  category: 'context' | 'project' | 'pattern' | 'interaction';
  domain: string;
  content: string;
  metadata: {
    createdAt: string;
    updatedAt: string;
    accessCount: number;
    confidence: number;
    source: string;
  };
  tags: string[];
  relatedKeys: string[];
}

Retrieval Strategies

1. Explicit Retrieval

User requests specific memory:

User: "What was the decision about caching?"
→ Search: domain=decisions, query=caching

2. Contextual Retrieval

Automatic based on current work:

Working on: src/kernel/audit.ts
→ Auto-load: patterns related to audit, recent audit decisions

3. Similarity Search

Find related memories:

Current task: "Implement rate limiting"
→ Find: patterns for throttling, previous rate limit implementations

Memory Lifecycle

Creation

1. Extract from session
2. Validate relevance
3. Assign category and tags
4. Store with provenance

Access

1. Retrieve on demand
2. Update access count
3. Refresh confidence based on usage

Decay

1. Rarely accessed items lose confidence
2. Contradicted items are flagged
3. Outdated items are archived

Consolidation

1. Related items are linked
2. Duplicates are merged
3. Patterns are extracted from instances

Integration with ARI Memory Manager

The session memory skill integrates with ARI's MemoryManager:

// Store session learning
await memoryManager.store({
  key: 'session_learning_20240127',
  content: 'Circuit breaker pattern works well for external API calls',
  domain: 'patterns',
  tags: ['resilience', 'api', 'circuit-breaker'],
  source: 'CLAUDE_CODE_SESSION'
});

// Retrieve relevant context
const context = await memoryManager.search({
  domain: 'patterns',
  tags: ['api'],
  limit: 5
});

Best Practices

1. Be Specific: Store specific, actionable knowledge, not vague observations
2. Use Tags: Good tags make retrieval effective
3. Include Context: Why is this knowledge important?
4. Link Related Items: Connect related memories
5. Review Periodically: Clean up outdated or incorrect memories