Persistent Memory Manager

Persistent Memory Manager

Context

Use this skill to ensure the AI agent retains context across crashes, pauses, and long-running sessions. The memory system relies on JARVIS_MEMORY.md to track active state, and .antigravityignore to keep token counts optimized.

Instructions

1. Initialize: Check if JARVIS_MEMORY.md and .antigravityignore exist in the workspace root. If missing, create them using standard scaffolding.
2. Restore: At the beginning of a session, read JARVIS_MEMORY.md. Acknowledge the last task and crash marker.
3. Log Progress: After completing a major task, update the [ ] checkboxes to [x] in JARVIS_MEMORY.md.
4. Update Timestamp: Update the Last Timestamp field to the current time.
5. Extract Permanent Knowledge: If a task reveals a new structural rule or codebase pattern, append it to NOTES_TECHNIQUES.md and remove it from JARVIS_MEMORY.md to avoid bloat.

Constraints

• NEVER store raw logs or large code snippets in JARVIS_MEMORY.md.
• NEVER index directories listed in .antigravityignore unless explicitly overridden by the user.
• NEVER let JARVIS_MEMORY.md exceed 100 lines. Refactor old tasks out if it grows too large.

Examples

Ideal JARVIS_MEMORY.md Update

# Active Session
- Last Timestamp: 2026-05-05T21:30:00+08:00
- Crash Marker: Clean exit

# Current Focus
- [x] Implement login component
- [ ] Connect login to auth API