---

name: Skill Builder description: Specialized instructions for building high-quality, reusable agent skills following the Vercel/Agentic standards. Use this when creating new skills or documentation for the agent.

Skill Builder

This skill provides the core principles and process for creating new agent skills. Follow these guidelines to ensure skills are concise, effective, and maintainable.

1. Core Principles

1.1 Concise is Key

The context window is a public good. Only add context the agent doesn't already have.

• Challenge every sentence: "Does the agent really need this explanation?"
• Prefer examples: Use concise code snippets over verbose explanations.

1.2 Set Appropriate Degrees of Freedom

Match the level of specificity to the task:

• High freedom: Use text-based instructions when many approaches are valid.
• Medium freedom: Use pseudocode or scripts with parameters for preferred patterns.
• Low freedom: Use specific scripts for fragile, error-prone, or critical sequences.

1.3 Anatomy of a Skill

Copyskill-name/
├── SKILL.md (required)
│   ├── YAML frontmatter (name, description)
│   └── Markdown instructions
└── Bundled Resources (optional)
    ├── scripts/ - Executable code (Python/Bash)
    ├── references/ - Documentation to be loaded as needed
    └── assets/ - Templates, icons, fonts, boilerplate

2. Skill Creation Process

Step 1: Understanding & Examples

Identify the specific domain and gather concrete examples of "Incorrect" vs "Correct" patterns.

Step 2: Planning

Decide on the structure:

• What belongs in SKILL.md (core procedural instructions)?
• What belongs in references/ (detailed schemas, docs)?
• What scripts are needed?

Step 3: Initialization

Create the folder and SKILL.md with accurate YAML frontmatter. The description is critical as it triggers the skill.

Step 4: Iteration

Test the skill and prune unnecessary text. Move detailed reference material to separate files in the references/ folder to keep the main skill lean.

3. Progressive Disclosure

Don't overwhelm the agent. Mention that detailed docs exist in references/ and provide grep patterns or specific file names so the agent can fetch them only when needed.