create

name: create description: Create new Claude Code skills following established conventions argument-hint: [skill name or description] allowed-tools: Bash, Read, Write, Glob, Grep, Edit

Create

Create or update skills with strict frontmatter, compact instructions, and progressive-disclosure structure.

How to Use

• /create - interactive creation (name + purpose + style)
• /create <name> - create named skill and infer remaining metadata
• /create <name> <description> - create with explicit purpose line

Arguments

Optional: $ARGUMENTS

• first token: skill name (lowercase-hyphen)
• remaining tokens: one-line description
• omit arguments to enter interactive requirement gathering

Workflow

1. Gather concrete trigger/use examples from user request.
2. Determine style:
   • procedural (workflow-first)
   • reasoning (principle/reference-first)
3. Inspect nearest existing skills for convention drift.
4. Draft frontmatter with required fields.
5. Author compact SKILL body with strict sections.
6. Create or update agents/skills/<name>/SKILL.md.
7. Validate structure, references, and trigger clarity.

Frontmatter Contract

Required fields:

• name: lowercase-hyphen, matches directory
• description: concise trigger summary
• argument-hint: bracketed hint for invocation
• allowed-tools: only tools genuinely needed

Optional fields are allowed only when required by runtime behavior.

SKILL Body Contract

Procedural skill baseline

1. H1 + one-line purpose
2. ## How to Use (minimum 3 concrete invocations)
3. ## Arguments (explicit parsing rules)
4. ## Workflow (numbered deterministic steps)
5. domain-specific constraints
6. ## Quality Standards

Reasoning skill baseline

1. H1 + one-line purpose
2. ## How to Use
3. ## When to Apply
4. framing section (Role, Philosophy, or equivalent)
5. principle/rule sections
6. review/audit contract
7. reference index

Writing Rules

• MUST be directive, dense, and operational.
• MUST avoid filler, narrative preamble, and redundant explanation.
• MUST keep one core action per bullet where possible.
• MUST use MUST/SHOULD/NEVER for enforceable constraints.
• MUST include defaults and exception conditions when relevant.
• SHOULD keep SKILL files compact by moving heavy material to references/.
• SHOULD keep references one level deep and linked directly from SKILL.
• NEVER create auxiliary docs (README, CHANGELOG, process notes) inside skill folders.

Progressive Disclosure Rules

• Keep SKILL body focused on workflow and constraints.
• Place large examples, matrices, and long domain docs in references/.
• Add scripts/ only for deterministic repeated automation.
• Add assets/ only for output resources, not instruction content.
• Avoid duplicating the same knowledge between SKILL and references.

Quality Standards

• Valid YAML frontmatter and matching directory/name.
• Three invocation examples in How to Use.
• Correct section structure for selected style.
• No broken reference links.
• No emojis, no extraneous files, no dead guidance.
• Skill is self-sufficient and executable by another agent without external explanation.