Verified Memory Vault
A self-checking Obsidian vault for AI coding agents (Claude Code, Codex, Gemini CLI, …) — persistent memory that proves it works.
Why "verified"? Plain vault templates assume your agent's memory still works.
This one checks itself and protects itself:
- Health score:
python3 tools/memory_check.pyreports protocol violations, dead links, bloat, inbox pressure — honest, with exit codes. - Self-protection: an optional git pre-commit hook refuses the commit that deletes your agent's brain — the documented way agent memories die.
- Zero dependencies: two ~200-line Python stdlib scripts. No Node, no TypeScript, no plugins, no database.
- Bilingual: EN + DE templates; the agent mirrors your language.
- License: vault content CC BY 4.0 (commercial use fine), tools MIT.
The three rules
- Capture, don't sort — everything lands in
00_Inbox/first. - Every session leaves a daily note — so the next session resumes cold.
- Promote durable facts to
MEMORY.md— append-only, dated, one line, verifiable only.