Continuous Knowledge Transfer Between Claude and Codex

For the last 8 months I've developed strictly using Claude Code, setting up context layers, hooks, skills, etc. But relying on one model has been limiting, so here is how I setup context knowledge transfer between Claude and Codex.

The key idea is that just like Claude Code (.claude/skills/ + CLAUDEmd), you can generate matching Codex CLI docs (AGENTSmd + .agents/skills/). Then, the only things is to keep documentation current for both. Aspens can generate both doc sets once and an optional git post-commit hook can auto-update them on commits. You can work with both models or just one. It works either way.

Claude Code: .claude/ skills/ auth/skill md settings json # permissions, hooks hooks/ # optional project scripts used by hooks agents/ # subagent definitions commands/ # custom slash commands CLAUDE md # root instructions Codex: .agents/ skills/ billing/SKILL md auth/SKILL md .codex/ config toml # optional local config AGENTS md # instructions src/billing/AGENTS md # optional scoped instructions src/auth/AGENTS md # optional scoped instructions 

I would love to see if others have found better ways for this ?