SKILL_SLUG

Skill Title

One-paragraph orientation: what this skill installs/does and the shape of the work. Lead with the outcome, not the mechanism.

Step1 → Step2 → [human-in-the-loop gate] → Step3 → Verify → Clean up

1. <First phase>

Concrete, ordered instructions the agent follows. Prefer imperative voice ("Copy X", "Run Y", "Read one generated file and spot-check"). State the why only where a future maintainer would otherwise simplify away something load-bearing.

2. <Second phase>

• •Bullet the decision points (single repo vs. workspace, language coverage…).
• •Reference bundled assets by path: assets/<file>, scripts/<file>.
• •Call out the human-in-the-loop gates explicitly if any.

3. Validate

Don't declare success blind. List the gates: syntax, a real run, an edge-case check, and a measured recall/behavior check. A gate exists because skipping it shipped a bug at least once — say so.

4. Report

Summarize for the user: what was installed/changed, measurements, a gate/pass/evidence table, and what was skipped.

<!-- TEMPLATE NOTES (delete this block when authoring):

• •name MUST be the kebab-case slug and MUST match the repo skill dir name.
• •description is the single most important line — it drives triggering.

Write it last, after the body exists. Front-load trigger phrases.

• •Keep SKILL.md lean; push long material into references/*.md, code into

scripts/ or assets/. The agent reads references on demand.

• •Run the skill-creator skill to lint description quality if available.

-->