Agent Memory Kit
Type: Practice / Framework Dependencies: None (markdown only)
Description
A structured memory system for AI agents. Prevents the "forgot how to do things" problem by separating memory into episodic (what happened), semantic (what I know), and procedural (how to do things) layers.
Installation
# Create memory folder structure
mkdir -p memory/procedures
# Copy templates
cp templates/ARCHITECTURE.md memory/
cp templates/feedback.md memory/
cp templates/procedure-template.md memory/procedures/
Usage
- Read
README.mdto understand the system - Add memory loading to your wake routine (AGENTS.md)
- Use templates when logging events, creating procedures, tracking feedback
Files
| File | Purpose |
|---|---|
README.md |
Full documentation |
templates/ARCHITECTURE.md |
Memory system overview (copy to memory/) |
templates/feedback.md |
Success/failure tracking template |
templates/procedure-template.md |
How-to document template |
templates/daily-template.md |
Daily log template |
templates/compaction-survival.md |
NEW: Pre-compaction flush guide |
templates/context-snapshot-template.md |
NEW: Quick context save template |
helpers/check-compaction.sh |
NEW: Token limit checker |
Key Concepts
- Episodic memory: Daily logs of WHAT happened
- Semantic memory: Curated knowledge (MEMORY.md)
- Procedural memory: HOW to do things
- Feedback loops: Learn from success/failure
The Rule
Always capture the HOW, not just the WHAT. Future-you needs the steps.