SKILL.md-Creator-Tester

SKILL.md — skill-creator

Description: Use this skill when you need to create a new SKILL.md from scratch, review and improve an existing one, or auto-fix blank and vague fields — turning rough ideas into a clear, structured, production-ready AI agent skill definition that works with Claude Code, Cursor, and Antigravity IDE.

Task:

1. Read all six fields the user has provided: Name, Description, Task, Input, Output Format, and Never Do.
2. Detect which fields are blank, too short (under 10 characters), or contain vague language such as "help", "assist", "do your best", or "be creative".
3. For any weak or missing field, intelligently generate a detailed, specific, and actionable replacement based on the skill name as the primary context clue.
4. If the skill name itself is missing or too generic (e.g., "skill-1", "my-skill"), generate a descriptive verb-noun slug such as "code-reviewer", "brand-generator", or "test-writer".
5. If an existing SKILL.md is provided for improvement, identify every section that lacks concrete steps, clear constraints, or an explicit output structure, then rewrite those sections to be more specific and scoped.
6. Ensure the Task field uses numbered steps, the Input field uses bullet points, the Output Format field includes a real markdown example (not just a description), and the Never field has at least 3 hard, project-specific constraints.
7. Return the completed SKILL.md in the exact format this app generates — six sections, no extra headers, no frontmatter blocks unless explicitly requested.
8. After generating, provide a short Review Summary noting what was strong, what was improved, and any suggestions for further refinement.

Input:

• Skill Name — a short verb-noun slug describing what the skill does (e.g., "code-reviewer", "logo-generator", "content-writer"); this is the most important context clue for all AI fixes
• Description — one sentence explaining when and why to invoke this skill; should answer "what problem does this solve?"
• Task — a numbered list of concrete actions the AI must take; avoid vague instructions like "help the user" or "do your best"
• Input — bullet list of every piece of information the skill needs from the user to operate correctly, including format hints (e.g., "file path", "hex color code", "free text")
• Output Format — the exact structure and format of the final result, ideally with a markdown example showing section headers, bullet points, or tables as they should appear
• Never Do — at least 3 hard constraints that prevent the AI from producing bad, unsafe, or off-scope output; each item must start with "Never" and be specific to the skill's domain

Output Format: A complete, ready-to-use SKILL.md file structured exactly like this:

SKILL.md — {skill-name}

Description: {One sentence. What does this skill do and when should it be invoked?}

Task:

1. {First concrete, numbered action the AI takes}
2. {Second action — specific and measurable}
3. {Continue until all major steps are covered}

Input:

• {Input item 1} — {format hint or source}
• {Input item 2} — {required or optional}
• {Input item 3} — {example value}

Output Format: {Exact markdown structure the AI must return. Include real section headers, bullet points, tables, or code blocks as they should appear in the final output. Do not just describe the format — show it.}

Never:

• Never {constraint specific to this skill's domain}
• Never {constraint that prevents a common failure mode}
• Never {constraint that protects the user from bad or unsafe output}

Followed immediately by a Review Summary:

✨ Review Summary

✅ Strengths: {what was already well-defined} ⚠️ Changes Made: {what was improved, added, or restructured} 💡 Suggestions: {optional follow-up ideas the user can act on}

Never:

• Never produce a SKILL.md without a Never section — constraints are mandatory and prevent AI misuse
• Never use vague language like "help the user", "do your best", "be creative", or "assist" in any task step
• Never skip the Output Format section or fill it with only a description — always show a real markdown example
• Never name a skill with generic or meaningless terms like skill-1, my-skill, helper, or tool
• Never invent or assume inputs the user did not provide — ask for clarification if the intent is unclear
• Never return a partial SKILL.md — all six sections (Name, Description, Task, Input, Output Format, Never) must be complete
• Never ignore the project's tech stack or context when generating task steps or output format examples