| name | skill-creator |
| description | Create a new domain-specific skill for the agentic skill library |
| activation | {"keywords":["create skill","new skill","add skill","custom skill","skill template","make a skill"],"file_patterns":[".claude/skills/**"]} |
Skill Creator
Design Principle
Skills inject domain-specific knowledge Claude doesn't have reliably on its own.
They are NOT methodology templates. Before creating a skill, ask:
"Does this inject facts, patterns, or standards that Claude lacks — or does it just
describe a process Claude already knows?"
If Claude would do fine without the skill, don't create it. Good candidates:
- Specific tool syntax (Dockerfile best practices, Terraform patterns, Mermaid syntax)
- Specific standards with enumerable criteria (WCAG levels, OWASP Top 10)
- Specific domain checklists (SQL indexing patterns, cache invalidation strategies)
- Specific output tasks Claude doesn't naturally do (de-ai-ify text, extract PDF content)
Poor candidates (delete these, don't create more):
- How to debug, refactor, review code, write tests — Claude knows these
- Generic "best practices" without domain-specific content
- Methodology templates with phases and steps
Skill File Format
Create the file at .claude/skills/<skill-name>.md:
---
name: <skill-name>
description: <one-line description of what domain knowledge this injects>
activation:
keywords: ["keyword1", "keyword2"]
file_patterns: ["**/*.ext", "specific-file"]
---
# Skill Title
## Purpose
[What specific domain knowledge does this inject? Be concrete.]
## [Domain Knowledge Sections]
[The actual expert content — patterns, standards, checklists, syntax examples]
[This is the value. Make it specific and actionable.]
## Failure Modes
[Anti-patterns, edge cases, and things that look right but go wrong in this domain.
Only include observed failures — not inferred ones. Format: "Avoid X because Y."
Leave this section empty or omit it if no real failures have been observed yet.
Do NOT populate this with generic caution — it must be specific and earned.]
## Output Format (optional)
[Only if the skill produces a specific structured output]
Also Update skill-rules.json
Add an entry to .claude/skills/skill-rules.json:
"<skill-name>": {
"keywords": ["keyword1", "keyword2"],
"filePatterns": ["**/*.ext"],
"toolTriggers": ["Read", "Write"]
}
toolTriggers options: Read, Write, Edit, Bash, Glob, Grep
Also Update CLAUDE.md
Add the new skill to the appropriate category row in the Skills Library table in CLAUDE.md.