Memory Hygiene

Knowledge persistence decision framework for AI agents. Stop dumping everything into memory.

What This Is

An Agent Skills compatible skill that teaches AI agents WHERE to persist knowledge — routing reusable experience to rules, skills, and docs instead of memory files.

The Problem

Every AI agent platform has "memory" — and every AI agent abuses it. LLMs default to memory because it's the path of least resistance. But memory is:

• Not shareable (local to one machine/instance)
• Not trackable (not in git)
• Not in any dependency chain
• Effectively dead when you switch context

The Solution

A three-question test before persisting anything:

1. Can others see it? — Route to git-tracked location
2. Does it create impact? — Route to publishable location
3. Is it in a dependency chain? — Route to where dependents can find it

All NO? Memory is fine (for local state snapshots only).

Install

Claude Code

git clone https://github.com/motiful/memory-hygiene ~/motifpool/memory-hygiene
ln -s ~/motifpool/memory-hygiene/skill ~/.claude/skills/memory-hygiene

Hard Enforcement (Optional)

Copy the rule template to your platform's rules directory for always-on constraint enforcement. See skill/references/rules-template.md for templates.

Other Platforms

Clone the repo and symlink/copy skill/ to your agent's skills directory.

Part of the Rules-as-Skills Pattern

This skill follows the Rules-as-Skills methodology — encoding constraints as dynamically-loaded skills for context efficiency and cross-platform portability.

License

MIT