Create Skill

Skill Creation Guide

Reference for creating skills that follow the Anthropic skill standard (github.com/anthropics/skills).

Skill Anatomy

skill-name/
├── SKILL.md (required)
│   ├── YAML frontmatter (name, description required)
│   └── Markdown instructions (<500 lines)
└── Bundled Resources (optional)
    ├── scripts/    - Executable code for deterministic tasks
    ├── references/ - Docs loaded into context as needed
    └── assets/     - Templates, icons, fonts for outputs

Progressive Disclosure

Skills load in three layers:

1. Metadata (name + description) - Always in context (~100 words)
2. SKILL.md body - Loaded when skill triggers (<500 lines)
3. Bundled resources - Loaded as needed (unlimited size)

Keep SKILL.md under 500 lines. If approaching the limit, move content to references/ with clear pointers about when to read each file.

YAML Frontmatter

Two required fields:

---
name: skill-name
description: What the skill does AND when to trigger it. All "when to use" context goes here, not in the body.
---

Description is the Trigger

The description field is the primary mechanism for skill invocation. Claude tends to undertrigger skills, so descriptions should be slightly pushy.

Weak: "Generate LinkedIn posts for professional audiences."

Strong: "Write high-performing LinkedIn posts. Use when user says 'write a LinkedIn post', 'LinkedIn content', 'post for LinkedIn', 'professional audience post', or wants to share insights, lessons, or updates with a professional network - even if they don't explicitly say LinkedIn."

Include:

• What the skill does (1 sentence)
• Explicit trigger phrases users might say
• Adjacent contexts where the skill applies ("even if they don't explicitly say X")

Body Structure

Every skill body should include these components:

1. Overview (optional, 1-2 paragraphs)

What this skill does and the critical distinction that shapes how it works. Name the book or methodology the skill is grounded in.

2. Workflow (required)

Step-by-step process. Use ### Step N: headings. Each step should be actionable and specific.

3. Anti-Patterns (required)

What NOT to do. Format as Bad/Good pairs so the model can calibrate:

## Anti-Patterns

**1. Descriptive name**
Bad: "example of wrong approach"
Good: "example of correct approach"

4. Quality Checklist (required)

Checkbox list the model uses to verify output before delivering:

## Quality Checklist

- [ ] Check 1
- [ ] Check 2

Writing Style

Explain the "Why" Instead of Bare Directives

The model follows instructions better when it understands the reasoning. Heavy-handed MUSTs without context are less effective than explaining why something matters.

Weak: "Never exceed 5 bullets per slide."

Strong: "Keep slides to 3-5 bullets. Beyond 5, the audience reads instead of listening - the slide competes with the speaker instead of supporting them."

Use Imperative Form

Write instructions as commands: "Open with a personal admission" not "You should open with a personal admission."

Generalize, Don't Overfit

Write instructions that work across varied inputs. Avoid narrow tweaks targeting only specific examples - look for underlying patterns.

Examples Are Valuable

Include concrete examples. Format with Input/Output or Bad/Good pairs. Examples calibrate the model's output better than abstract rules.

Ground Every Skill in a Book or Methodology

Skills grounded in a proven framework (book, methodology, model) are more credible, consistent, and unique. Name the source in the Overview section.

Good examples: Radical Candor (feedback), The Pyramid Principle (exec summaries), SPIN Selling (discovery calls), How Google Tests Software (risk-based testing).

Scaffolding a New Skill

Use init_skill.py to create the folder and register in skills.json in one step:

python3 scripts/init_skill.py <skill-name> --role <role> --description "<one-liner>" --tags "<tag1,tag2>"

Valid roles: pm, engineer, qa, designer, content-creator, devops, shared, leadership, program-delivery-manager, customer-success, recruiter, consultant, sales, marketing, founder, data-scientist, data-engineer, security

This creates skills/by-role/<role>/<skill-name>/SKILL.md and adds an entry to skills.json. Always scaffold first, then write the SKILL.md content.

After writing, validate:

python3 scripts/validate_skill.py

Placeholders (for this repo)

When creating skills for the claude-skills public repo, replace all personal/project references:

Creation Checklist

Before finalizing any skill:

• Scaffolded with init_skill.py (creates folder + registers in skills.json)
• YAML frontmatter has name and description
• Description is pushy with multiple trigger phrases
• Body is under 500 lines
• Overview names the book or methodology the skill is grounded in
• Workflow has clear numbered steps with ### Step N: headings
• Anti-patterns section exists with Bad/Good pairs
• Quality checklist exists
• Directives explain "why", not just "what"
• Examples are concrete and specific
• No personal/project-specific content (if for public repo)
• Placeholder brackets used for any context-specific references
• Overflow content moved to references/ with clear pointers
• python3 scripts/validate_skill.py passes