name: creating-skills description: Creates, validates, and packages Agent Skills following the open standard (agentskills.io). Portable across multiple AI agents. Use when building agent-discoverable capabilities with scripts, references, or assets. license: Apache-2.0 metadata: author: coveo version: "1.0"

Creating Skills

Skill Location

Skills following the Agent Skills open standard can be stored in multiple locations:

• .claude/skills/ - Workspace-level (recommended for local development, maximum compatibility)
• .skills/ or skills/** - Alternative workspace conventions
• ~/.claude/skills/ - User-level personal skills

This skill recommends .claude/skills/ for workspace development as it's compatible with multiple AI tools (GitHub Copilot, Claude Code, OpenCode, etc.).

Process

Step 1: Plan the Skill

Before creating, answer:

1. What specific task does this skill accomplish?
2. What triggers should activate this skill?
3. What reusable resources would help (scripts, references, assets)?
4. What is the best gerund-form name? (e.g., generating-tests, migrating-components)

Step 2: Initialize

node <skills-dir>/creating-skills/scripts/init_skill.mjs <skill-name> --path <skills-dir>

Default <skills-dir> is .claude/skills (workspace-level). Other common locations: .skills, skills/, ~/.claude/skills (user-level).

Creates: SKILL.md template, scripts/, references/, assets/ directories.

Step 3: Complete the Skill

1. Set name - Must match folder name, use gerund form
2. Write description - Third person, include what AND when to use
3. Write instructions - Keep under 500 lines, use concrete examples
4. Add resources - Scripts for deterministic tasks, references for detailed docs

For format details, see SPECIFICATION.md. For writing guidance, see best-practices.md.

Step 4: Clean Up

Remove unused placeholder files and directories created during initialization:

# Remove .gitkeep files from unused directories
find <skills-dir>/<skill-name> -name '.gitkeep' -delete

# Remove empty directories
find <skills-dir>/<skill-name> -type d -empty -delete

The skill should only contain directories with actual content. If you didn't add scripts, the scripts/ directory should not exist. Same for assets/ and unused subdirectories.

Step 5: Validate

node <skills-dir>/creating-skills/scripts/quick_validate.mjs <skills-dir>/<skill-name>

Step 6: Package (Optional)

node <skills-dir>/creating-skills/scripts/package_skill.mjs <skills-dir>/<skill-name>

Produces a .skill archive (tar.gz) for distribution.

Converting Prompts to Skills

1. Identify reusable logic and determine gerund-form name
2. Initialize: node <skills-dir>/creating-skills/scripts/init_skill.mjs <name> --path <skills-dir>
3. Move prompt content to SKILL.md, rewrite description in third person
4. Extract code examples to scripts/, detailed docs to references/
5. Validate, then deprecate original prompt

Modifying Existing Skills

Extending a skill:

1. Read existing SKILL.md and references to understand current scope
2. Add new sections/steps without removing existing functionality
3. Update description if scope expanded significantly
4. Run validation, test with representative use cases

Renaming a skill:

1. Create new directory with new name
2. Move all files, update name field in frontmatter
3. Search for references in your workspace (e.g., grep -r "old-skill-name")
4. Update all references, delete old directory

Deprecating a skill:

1. Add to description: "(Deprecated: use replacement-skill instead)"
2. Keep functional until dependents migrate
3. Delete only after confirming no references remain

Reference Documentation

Scripts

ui-kit Notes

Naming pattern: generating-*, migrating-*, validating-*, documenting-*

Integration:

• Skills stored in .claude/skills/ (workspace-level, compatible with all AI tools)
• Skills reference workspace instructions for context
• Agents compose skills for complex workflows
• Prompts should migrate to skills for better portability

Validation Checklist

• name matches folder, uses gerund form
• description explains what AND when (third person)
• SKILL.md under 500 lines
• References linked one level deep
• Scripts tested and executable
• All .gitkeep files removed
• No empty directories remain