Skip to content
Loop Engineering Tools

SKILL.md Templates

Vendor-specific SKILL.md starters with the real install paths and structure. Claude Code and Codex both follow the Agent Skills open standard — the format is shared, the location differs.

What each starter includes

Each page gives you the real location used by that tool, the expected folder structure, a copyable SKILL.md, and a short review checklist. The example is deliberately small: it shows when the skill should trigger, when it should stop, which steps are observable, and what evidence a reviewer should expect.

Treat the file as a maintained procedure, not as a larger prompt. Long reference material, scripts, and examples belong in supporting folders so the agent loads them only when the task needs them.

Choose the version your agent actually reads

Claude Code

Use the Claude Code page when the skill will live under .claude/skills and may be invoked directly with its skill name.

Codex

Use the Codex page for repository, user, or managed skill locations and the optional agent metadata supported by Codex.

Open standard

Start with the vendor-neutral version when you are documenting the method first and will adapt installation details for more than one compatible agent.

Review before you install

• Replace the example goal, commands, and limits with facts from your own project.

• Test one task that should use the skill and one similar task that should not.

• Read every bundled script before allowing an agent to run it.

• Recheck the official source links when install paths or product behavior matter.

• Keep human approval for merge, deploy, deletion, purchase, and external communication.