Meta-Skill: Add a Skill (Any Repo / Any Purpose)

Use this skill to create a new agent skill for any repository or use case. You gather requirements by asking the right questions, then produce a valid SKILL.md (and optional supporting files) that follows the Agent Skills specification.

When to use this skill

• User says: "Create a skill for…", "Add a skill that…", "Write a SKILL.md for…"
• User wants to add a skill to a repo (this one or any other)
• User asks how to structure or author an agent skill

Not for: Adding skills only to top-coder-agent-skills with design-pattern/architecture focus → use top-coder-meta-skill instead.

Phase 1: Discovery — Ask These Questions

Before writing the skill, gather (or infer from context):

1. Purpose and scope
   "What specific task or workflow should this skill help with?"
   One clear purpose per skill.

2. Target location
   "Where should the skill live?"

   • Personal: ~/.cursor/skills/<name>/ (or equivalent for other agents)
   • Project: .cursor/skills/<name>/ in a specific repo
   • Another repo: e.g. skills/<name>/ in that repo

3. Trigger scenarios
   "When should the agent automatically apply this skill?"
   Get concrete phrases or situations (e.g. "when reviewing PRs", "when user says 'deploy my app'").

4. Key domain knowledge
   "What must the agent know that it might not already know?"
   Tools, conventions, formats, domain terms.

5. Output format preferences
   "Any required templates, formats, or styles?"
   e.g. commit message format, report structure, file naming.

6. Existing patterns
   "Any existing skills or docs to match?"
   Repo conventions, similar skills to align with.

How to ask: If the user hasn’t provided enough detail, ask these questions conversationally (or use structured prompts). If context already answers some, infer and confirm briefly. Skip only when the answer is obvious.

Phase 2: Design

1. Name: Kebab-case, 1–64 chars, lowercase letters and hyphens. No leading/trailing or consecutive hyphens. Must match the skill directory name.

   • ✅ code-review, deploy-vercel, pdf-extract
   • ❌ helper, MySkill, do-stuff

2. Description: Third person. Include WHAT (what the skill does) and WHEN (when to activate). Trigger-heavy.

   • ✅ "Reviews code for security and style per team standards. Use when the user asks for a code review or says 'review this PR'."
   • ❌ "Helps with code" or "You can use this to review."

3. Outline: Main sections and any supporting files (references/, scripts/, assets/).

Phase 3: Authoring Rules

When writing the generated skill:

• Concise: No filler. Assume a capable agent; only add context it needs.
• SKILL.md under 500 lines. Long content → references/ or examples.md, linked one level deep from SKILL.md.
• Paths: Forward slashes only (e.g. scripts/helper.py).
• One default approach; avoid "you can use A, or B, or C" unless an escape hatch is needed.
• Consistent terminology (one term per concept).
• No time-sensitive rules (e.g. "before date X"); use a "Deprecated / legacy" section if needed.

Required frontmatter

---
name: <kebab-case-name>
description: <Third-person WHAT. Use when [WHEN/triggers].>
---

Optional: license, compatibility, metadata, allowed-tools (per spec).

Phase 4: Implementation

1. Create the skill directory (e.g. skills/<name>/ or the path agreed in Phase 1).
2. Write SKILL.md with frontmatter and body (instructions, steps, examples as needed).
3. Add references/, scripts/, or assets/ only if needed; link from SKILL.md.
4. Ensure valid Markdown and YAML; name and description within spec limits.

Phase 5: Verify

• Description is third person, specific, and includes trigger scenarios.
• Name is kebab-case and matches directory.
• SKILL.md is under 500 lines; references are one level deep.
• No Windows-style paths; terminology consistent.
• Output is ready to save and (if applicable) to install via npx skills add or manual copy.

General Meta-Skill — for adding any agent skill, any repo