Guides

Most users only need the introduction, quickstart, and the guide for their agent. Open the library-building guides when you are ready to turn real corrections into approved cards.

Getting started

• Introduction: what OME is, why AGENTS.md isn't enough, and how it works.
• Quickstart: install, initialize, and verify one recall explanation.
• First experience card: turn one real correction into an approved card.
• Examples: see a real /goal prompt, the matched card, and the exact injected context.

Agent setup

• Setup: data directory, hooks, skill install, and uninstall details.
• Codex: configure prompt-time recall for Codex.
• Claude: configure prompt-time recall for Claude.

Build the library

• Source scan: build native Codex/Claude user evidence and optionally supplement with Spool.
• Reflect and draft approval: turn real work into approved cards.
• Global and project libraries: choose global storage, repository storage, or both.
• Evaluation: test recall quality without polluting your real library.