skill-creator

name

skill-creator

description

Create new skills, modify and improve existing skills, and measure skill performance for the Vue SSR Starter Kit. Use when users want to create a skill from scratch, edit or optimize an existing skill, run evals to test a skill, or optimize a skill's description for better triggering accuracy. ALWAYS load together with the claude-anthropic skill: skill-creator owns the create/eval/iterate workflow, claude-anthropic owns the project rules (naming, placement, frontmatter conventions, anti-triggers, references layout, post-creation audit checklist). Don't use for: project-specific skill conventions, audit of existing skills, or CLAUDE.md/agents/hooks governance (→ claude-anthropic skill).

Skill Creator

Contents: Project context · Creating a skill · Running evals · Improving · Description optimization

A skill for creating new skills and iteratively improving them.

Project context

If the claude-anthropic skill is available in this project, consult it FIRST for project-specific conventions: skill naming, placement under .claude/skills/, frontmatter rules (discriminating description, anti-triggers, domain prefix), references/ layout, and the post-creation audit checklist (scripts/audit.py).

This skill (skill-creator) handles the generic create / evaluate / iterate loop. claude-anthropic handles the what / where / why for this project.

Division of responsibilities — skill-creator ↔ claude-anthropic

Concern	Owner
Generic create/evaluate/iterate workflow	skill-creator
Eval harness, test prompts, assertions	skill-creator
Description-optimisation guidelines	skill-creator
Project description conventions (discriminating, pushy, anti-triggers)	claude-anthropic
Skill naming, placement under .claude/skills/, folder layout	claude-anthropic
SKILL.md anatomy and token budget for this project	claude-anthropic
Project anti-patterns	claude-anthropic
Post-creation audit (scripts/audit.py) + CLAUDE.md index update	claude-anthropic

Handoff convention: ➜ See skill: claude-anthropic — <reason>.

Creating a skill

Capture intent

1. What should this skill enable Claude to do?
2. When should this skill trigger? (what user phrases/contexts)
3. What's the expected output format?
4. Should we set up test cases to verify?

Write the SKILL.md

• name: kebab-case, prefix vue-ssr- for project-scoped skills
• description: Discriminating, pushy, with anti-triggers. Min 80 chars.
• body: Method + index. Knowledge in references/.

Folder layout

.claude/skills/<skill-name>/
├── SKILL.md            (required)
├── references/         (optional, for skills > ~200 lines)
│   └── <topic>.md
├── scripts/            (optional, deterministic tooling)
└── evals/              (optional, test cases)
    └── evals.json

Running evals

Create evals/evals.json with test prompts and assertions:

{
  "skill_name": "my-skill",
  "evals": [
    {
      "id": 1,
      "name": "test-name",
      "prompt": "User prompt to test",
      "assertions": [{ "text": "Expected behavior or content in response" }]
    }
  ]
}

Improving the skill

1. Run evals, note failures
2. Identify patterns in failures
3. Add missing context/rules to SKILL.md or references
4. Re-run evals until assertions pass
5. Run scripts/audit.py to validate structure

Description optimization

• Discriminating: State exact domain, file paths, key types, trigger keywords
• Pushy: "Trigger whenever...", "Always load when..."
• Anti-triggered: "Don't use for: X (→ alternative-skill)"
• Budget: 1,536 characters max (combined description + when_to_use)
• Min length: 80 characters