Memory

Memory Skill

Dual-Track System

Track 1: File-Based (.ai_state/)

Purpose: Project-specific state
Persistence: Git-versioned
Contents:
  - requirements/
  - designs/
  - experience/
  - checkpoints/
  - meta/

Track 2: Memory MCP

Purpose: Cross-session memory
Persistence: MCP storage
Contents:
  - User preferences
  - Long-term patterns
  - Project summaries

Session Lifecycle

Session Start

Actions:
  1. Load .ai_state/meta/session.lock
  2. Query Memory MCP for context
  3. Restore last active task
  4. Load relevant experience

Session End

Actions:
  1. Save current state to .ai_state/
  2. Update Memory MCP with key insights
  3. Release session.lock
  4. Trigger continuous-learning if needed

Pre-Compact

Actions:
  1. Save state to pre-compact-state.yaml
  2. Store key context in Memory MCP
  3. Generate session summary

Memory MCP Integration

Store:
  - Key decisions made
  - User preferences
  - Project context summary

Retrieve:
  - Previous session context
  - User interaction patterns
  - Long-term project knowledge

Commands

/vibe-status         # View current state
/vibe-resume         # Restore from memory
/vibe-pause          # Save and pause