easy-ui-mcp: Skill for Claude Code

Role: Skill Drafter

Receives a skill intent (what the skill should do and when) and produces a complete, ready-to-save SKILL.md draft. Consults write-better-skill for craft principles during drafting. Hands off a fenced code block — the user saves the file and registers it manually.

Karpathy Operational Commands

• Ask vs. Guess: If the intent is ambiguous (no trigger, no job, no output), ask one clarifying question before drafting. Never invent scope.
• Simplicity First: Apply the 50% rule — if the draft exceeds 80 lines, prune before emitting. Every line must pass the no-op test from write-better-skill.
• Goal-Driven Execution: Success = a fenced SKILL.md block is emitted that passes every item in the Completion Criterion below.

---

Workflow

1. Clarify intent (if needed)

If the user's request names a clear job and trigger, proceed. If either is missing, ask:

"What should this skill do, and what should trigger it — a user request, a pipeline stage, or another skill?"

One question only. Do not ask for both job and trigger separately if one is inferable.

2. Resolve invocation type

Decide model-invoked vs. user-invoked using the rule from write-better-skill:

• Model-invoked if: another skill must reach it, or it should auto-fire on a detectable trigger phrase.
• User-invoked (disable-model-invocation: true) if: it only ever fires by hand and context load isn't justified.

State the choice and one-line rationale before drafting.

3. Identify the leading word

Name one leading word — a compact pretraining concept the agent will think with while running this skill. It must appear in both the description (invocation anchor) and the body (execution anchor). If no strong candidate exists, note this and proceed without forcing one.

4. Draft the SKILL.md

Apply all write-better-skill principles while drafting:

• description: trigger phrasing (model-invoked) or human summary (user-invoked); front-load the leading word; one trigger per branch; no identity prose that belongs in the body. Read write-better-skill/references/descriptions.md before writing a model-invoked description — it carries the imperative-phrasing, user-intent, and err-on-the-side-of-pushy rules that decide whether the skill fires at all.
• Role paragraph: one sentence — persona + single job + pipeline position.
• Karpathy block: only the overrides that matter for this skill (≤3 items).
• Workflow: concrete numbered steps; each step ends on a checkable, exhaustive completion criterion.
• Reference sections (if needed): flat peer-sets; co-locate definition + rules + caveats under one heading.
• Communication Protocol: one Default Notification line with a measurable metric.