Memory

Memory - Persistent Knowledge System

Purpose: Maintain persistent knowledge about projects, environments, and user preferences so you never have to repeat yourself.

Core Principle

If you tell me something once, I should remember it forever.

This skill captures and organizes:

• Project-specific facts (hosting, domains, tech stack, integrations)
• Environment details (API keys locations, deployment workflows)
• User preferences (coding style, communication preferences)
• Decisions and their rationale
• Lessons learned from debugging sessions

Memory Structure

All memories stored in ${PAI_DIR}/memory/:

memory/
├── projects/           # Project-specific knowledge
│   ├── carbene-website.md
│   ├── carbene-homelab.md
│   ├── pai.md
│   └── cyberdefense-tactics.md
├── user/               # User preferences and personal info
│   └── preferences.md
└── global/             # Cross-project knowledge
    └── integrations.md

Workflows

1. Capture Learning (Proactive)

When to use: Automatically after learning something new about a project or user.

TRIGGER: Any of these situations:
- User corrects a mistake ("It's Vercel, not Cloudflare")
- User provides new information ("The domain is carbene.ai")
- Discovering project structure or integrations
- User states a preference
- Resolving a confusing issue

ACTION:
1. Identify the memory category (project/user/global)
2. Read existing memory file (if exists)
3. Append new fact with timestamp
4. Confirm capture to user (briefly)

2. Recall Context (On Session Start)

When to use: At the beginning of each session, based on working directory.

ACTION:
1. Detect current working directory
2. Map to project memory file
3. Load relevant memories into context
4. Silently apply (no need to announce)

3. Query Memory

When to use: When user asks "What do you know about X?"

ACTION:
1. Search memory files for relevant content
2. Return organized summary
3. Offer to update if anything is outdated