---

name: cfn-automatic-memory-persistence description: "Automatic, structured persistence of agent outputs to SQLite database. Use when tracking agent outputs across CFN Loop workflows, persisting confidence scores, or querying agent execution history." version: 1.0.0 tags: [memory, persistence, sqlite, auto, agent-tracking]

Automatic Memory Persistence Skill

Overview

This skill provides automatic, structured persistence of agent outputs to a SQLite database, ensuring consistent memory tracking across CFN Loop workflows.

Key Components

• persist-agent-output.sh: Primary script for saving agent outputs
• query-agent-history.sh: Query and retrieve past agent outputs
• test-memory-persistence.sh: Validation test suite

Memory Schema

Copy{
  "agent/[agent-id]/output/[task-id]": {
    "output": "Raw agent output text",
    "confidence": 0.85,
    "iteration": 1,
    "timestamp": "2025-10-20T15:30:00Z",
    "parsed_data": null
  }
}

Integration Points

• Loop 3 output processing
• Loop 2 output processing
• Product Owner decision parsing

Usage Examples

Persist Agent Output

Copy./persist-agent-output.sh \
  "task_authentication_v1" \  # Task ID
  "backend-dev" \             # Agent ID
  "Implemented JWT auth" \    # Output
  0.85 \                      # Confidence
  1                           # Iteration

Query Agent History

Copy# Get last 5 outputs for a specific agent
./query-agent-history.sh "backend-dev" "" 5

# Get outputs for a specific task
./query-agent-history.sh "backend-dev" "task_authentication_v1"

Security & Performance

• ACL Level 1 (Read-only access)
• Indexed by tags for fast retrieval
• Minimal performance overhead
• Escaped and normalized inputs

Testing

Run comprehensive tests:

Copy./test-memory-persistence.sh

Sprint 7 Insights

• Confidence: 0.95
• Priority: 9
• Provides robust, zero-configuration memory tracking
• Eliminates manual output saving
• Supports multi-iteration workflows

Best Practices

• Always use script for output persistence
• Do not modify SQLite database directly
• Use query script for retrieval
• Add structured parsing for complex outputs