██╗      ██████╗  █████╗ ██████╗  ██████╗ ██╗   ██╗████████╗
██║     ██╔═══██╗██╔══██╗██╔══██╗██╔═══██╗██║   ██║╚══██╔══╝
██║     ██║   ██║███████║██║  ██║██║   ██║██║   ██║   ██║
██║     ██║   ██║██╔══██║██║  ██║██║   ██║██║   ██║   ██║
███████╗╚██████╔╝██║  ██║██████╔╝╚██████╔╝╚██████╔╝   ██║
╚══════╝ ╚═════╝ ╚═╝  ╚═╝╚═════╝  ╚═════╝  ╚═════╝    ╚═╝

---

This is a personal experiment and not a production-ready solution.

📌 ~/problem

Projects often copy skills from a central skills repository. That creates drift:

• the project copy diverges from the source
• improvements get lost or duplicated
• each client (Codex vs Claude) ends up with a different layout and UX

🧠 ~/solution

Loadout makes skill usage reproducible and agent-friendly:

• manifest + lock (pinned commit SHAs) committed in the project
• per-project source clones in .codex/.loadout/ (gitignored)
• symlink export into .codex/skills/ and .claude/skills/
• explicit trust gate for third-party sources
• JSON by default on stdout (including errors)

Docs:

• docs/PRD_Skill_Manager.md
• docs/Agent_Playbook.md

🚀 ~/quickstart

# in a git project
loadout init --primary-url <skills_repo_url> --primary-ref main

loadout suggest --target codex --query "pdf" --limit 10
loadout add --target codex pdf-processing

# if you select third-party skills, you must trust the source explicitly
loadout source trust third --yes

🧪 ~/dev

cargo test
uvx prek run --all-files

---

🗒️ ~/notes

• v1 targets macOS/Linux (symlink export). Windows fallback export is a roadmap item.

License

MIT