Skill: Behavior Memory Integration

Purpose & Scope

This skill teaches AI agents to leverage @lanonasis/mem-intel-sdk for behavioral pattern learning - recording successful workflows and recalling them in future sessions.

Key Principle: Local-first processing with cached embeddings. API calls are fallback, not primary.

API Reference

Base URL: https://api.lanonasis.com/api/v1

Operation	Endpoint	Method	Description
Store pattern	`/memories`	POST	Create with `type: workflow`
Recall patterns	`/memories/search`	POST	Semantic search with `type: workflow`
List patterns	`/memories?type=workflow`	GET	Filter by workflow type
Update pattern	`/memories/{id}`	PUT	Update use_count in metadata
Delete pattern	`/memories/{id}`	DELETE	Remove stale patterns
Analyze patterns	`/intelligence/analyze-patterns`	POST	mem-intel-sdk
Find related	`/intelligence/find-related`	POST	mem-intel-sdk
Extract insights	`/intelligence/extract-insights`	POST	mem-intel-sdk

Authentication (3 methods)

code

OAuth2 PKCE (Primary): Authorization header with token
API Key (Fallback):    X-API-Key: lano_*
JWT (Legacy):          Authorization: Bearer <jwt>

MemoryType Enum

typescript

type MemoryType = 'context' | 'project' | 'knowledge' | 'reference' | 'personal' | 'workflow';
//                                                                                  ^^^^^^^^
//                                                                      Use this for behavior patterns

API Payload Examples

Record Pattern (POST /memories)

json

{
  "title": "Workflow: Fix authentication bug in TypeScript API",
  "content": "{\"trigger\":\"fix auth bug\",\"actions\":[{\"tool\":\"Read\",\"outcome\":\"success\"},{\"tool\":\"Edit\",\"outcome\":\"success\"},{\"tool\":\"Bash\",\"outcome\":\"success\"}],\"final_outcome\":\"success\",\"duration_ms\":45000}",
  "type": "workflow",
  "tags": ["auth", "bugfix", "typescript-api"],
  "metadata": {
    "confidence": 0.85,
    "use_count": 1,
    "context": {
      "directory": "/home/user/onasis-gateway",
      "project_type": "typescript-api",
      "branch": "main"
    }
  }
}

Recall Patterns (POST /memories/search)

json

{
  "query": "fix authentication bug in typescript",
  "type": "workflow",
  "threshold": 0.7,
  "limit": 5
}

Architecture Understanding

code

+------------------------------------------------------------------+
|                 BEHAVIOR MEMORY FLOW                              |
+------------------------------------------------------------------+
|                                                                   |
|  SESSION START                                                    |
|  +-------------------+                                            |
|  | 1. Recall         |---> Query cached patterns for context      |
|  |    Patterns       |---> Semantic match: current task -> history|
|  +-------------------+                                            |
|           |                                                       |
|           v                                                       |
|  +-------------------+                                            |
|  | 2. Inject         |---> Enrich agent context with patterns     |
|  |    Context        |---> "User typically does X when Y"         |
|  +-------------------+                                            |
|           |                                                       |
|           v                                                       |
|  SESSION EXECUTION                                                |
|  +-------------------+                                            |
|  | 3. Agent          |---> Execute with pattern-informed behavior |
|  |    Actions        |---> Track tools used, outcomes             |
|  +-------------------+                                            |
|           |                                                       |
|           v                                                       |
|  SESSION END                                                      |
|  +-------------------+                                            |
|  | 4. Record         |---> Store: trigger -> actions -> outcome   |
|  |    Pattern        |---> Generate embedding for future recall   |
|  +-------------------+                                            |
|                                                                   |
+------------------------------------------------------------------+

Memory Types for Behavior Patterns

Memory Type	Use For
`workflow`	Multi-step action sequences
`context`	Session context (directory, project type)
`project`	Project-specific patterns
`knowledge`	Learned preferences and rules
`personal`	User-specific behavior preferences

MCP Tools Available

Analysis Tools (Read-Only)

typescript

// Analyze user's behavior patterns over time
memory_analyze_patterns({
  user_id: "uuid",
  time_range_days: 30,
  response_format: "markdown"
})

// Find related memories using vector similarity
memory_find_related({
  memory_id: "uuid",
  user_id: "uuid",
  limit: 10,
  similarity_threshold: 0.7
})

// Extract insights from memories
memory_extract_insights({
  user_id: "uuid",
  topic: "deployment",
  memory_type: "workflow",
  max_memories: 20
})

Critical Rules - NEVER Do

Data Privacy

•NEVER store PII in behavior patterns
•NEVER record authentication tokens or secrets
•NEVER log sensitive file contents in patterns
•NEVER share patterns across user boundaries

Pattern Quality

•NEVER record failed sessions as successful patterns
•NEVER store patterns with confidence < 0.5
•NEVER record incomplete workflows
•NEVER overwrite high-confidence patterns with lower ones

Performance

•NEVER call API when cache is fresh and relevant
•NEVER skip processingMode: 'offline-fallback' for recall operations
•NEVER generate embeddings client-side when API can do it
•NEVER store duplicate patterns (use detectDuplicates first)

Required Patterns - MUST Follow

Memory Storage

javascript

// MUST include these fields for workflow memories
{
  type: 'workflow',           // Required for behavior patterns
  tags: [...],                // Required for filtering
  metadata: {
    confidence: 0.0-1.0,      // Required for ranking
    directory: string,        // Required for context matching
    project_type: string      // Required for project filtering
  }
}

Pattern Recall

javascript

// MUST use offline-fallback for recall operations
const client = new MemoryIntelligenceClient({
  processingMode: 'offline-fallback',
  enableCache: true,
  cacheTTL: 300000  // 5 minutes
});

Confidence Scoring

javascript

// MUST calculate confidence based on outcome
function calculateConfidence(outcome) {
  let confidence = 0.5;  // Base

  if (outcome.userExplicitApproval) confidence += 0.3;
  if (outcome.noErrors) confidence += 0.1;
  if (outcome.completedAllSteps) confidence += 0.1;

  return Math.min(confidence, 1.0);
}

Integration with Other Skills

Skill	Integration Point
`skill-base-client.md`	Record API call patterns
`skill-compliance-manager.md`	Filter sensitive data before storage
`skill-metrics-collector.md`	Track pattern usage metrics
`skill-version-manager.md`	Version-tag workflow patterns