Refactor Agentic Workflow and Skills
Refactor the agentic workflow in this repository. I have AGENTS.md, with CLAUDE.md as a symlink to AGENTS.md. AGENTS.md should have purpose and reason for this repository, which is consise. Break up the current AGENTS.md into several skills which are consise and contained. The skills should be in .agents and create .claude and .codex folders which are symlinks to .agents. Keep the entire hooks so that it does not break entire integration
I’ll inspect the current instructions and hooks, then split the detailed guidance into focused skills and make .agents the shared configuration directory. I’m using the skill-creator skill for this refactor.
The Entire hooks live in .claude/settings.json and .codex/hooks.json; both can keep their existing paths through the symlinks. I also found older plans and agent definitions in .claude that I’ll preserve.
This session marks .agents and .codex read-only, so I can’t install the requested layout here. I’ll prepare and validate the complete migration in /tmp, including an installer you can run outside this restricted session.
Permissions updated to full access
I can now apply the migration directly. I’ll split the guidance into six skills covering design, architecture, data representation, API/CLI contracts, native integration, and development checks, while preserving the existing files and Entire hooks.
The shared layout and six skills are in place. AGENTS.md is now a short purpose statement and skill index. I’m checking that both tools resolve the skills and hooks correctly, and that the move preserved the existing files.
Refactored AGENTS.md to 19 lines with repository purpose and links to six focused skills.
- Skills live in
.agents/skills/. .claudeand.codexnow symlink to.agents.CLAUDE.mdstill links toAGENTS.md.- Existing plans and agent definitions were preserved.
- All 15 Entire agent hooks remain byte-for-byte unchanged; Git hooks remain intact.
All six skills, symlinks, links, and Git visibility checks passed.