Skip to main content

Using AI Skills

[!NOTE] This content is a work in progress and is updated regularly as AI steering methodologies evolve.

In the Shape & Ship paradigm, the primary mechanism for steering an agentic AI developer is through AI Skills.

An AI Skill is a machine-readable document (usually SKILL.md or packaged inside a .zip file along with static assets and schemas) that provides an AI agent with context-specific instructions, constraints, and resources for a project or brand.

Why Use Skills?​

Standard LLMs are trained on public web data and lack context about your specific project's architectural decisions, color palettes, folder structures, or prohibited libraries.

If you do not specify these constraints, agents will guess—often defaulting to outdated or generic code. Feeding a skill solves this issue:

[Agent Context] + [No Skill]     --> Outputs generic/standard code (e.g., Tailwind, generic colors)
[Agent Context] + [Brand Skill]  --> Outputs styled code conforming to brand guidelines (#EA5148, Sora font)

How Agents Consume Skills​

Different agentic setups consume skills in different ways:

1. File Reference (System Prompts)​

You can directly reference a skill file in your prompt or append it to the system context.

  • Instruction: "Follow the design guidelines and use the tokens defined in docs/SKILL.md to modify our styles."
  • Outcome: The agent reads the file, parses the color tokens, and applies them exactly as instructed.

2. Auto-Discovery (llms.txt or brand.json)​

Modern agents scan project roots for standard files:

  • brand.json: Contains color tokens, logo paths, and voice constraints in machine-readable JSON format.
  • llms.txt: A markdown-like directory that links to documentation files, detailing how agents should interact with your API or code.
  • SKILL.md: Direct agent instructions containing rules, styling guides, and code templates.

Real-World Example: Incrementic Brand Skill​

On brand.incrementic.com, developers and agents can download the Brand Skill Bundle (incrementic-brand-skill.zip).

When unpacked, this skill contains:

  1. SKILL.md: Markdown containing CSS color tokens, rules on how to format company text, logo alignment policies, and constraints (e.g. "never rotate or stretch the logo").
  2. brand.json: Machine-readable configuration detailing the primary color #EA5148, fonts, and allowed variations.
  3. Logo Assets: High-resolution SVGs that the agent can insert directly into codebases.

Feeding a Brand Skill to a Coding Agent​

When starting a task (like branding a website), the agent follows these sequential steps:

  1. Locate and read brand.json or SKILL.md.
  2. Import tokens into the project's styling sheet (e.g., mapping --ifm-color-primary to the primary brand color).
  3. Retrieve and place downloaded SVG assets into the /static/ or /assets/ folders.
  4. Compile & Verify that the layout doesn't violate forbidden guidelines (such as scaling the logo incorrectly).

Next Steps​

Now that you know how agents use skills to maintain code consistency, learn how to build your own:

  • Creating AI Skills: Guide on structure, schema, and authoring guidelines for custom project skills.
🤖Share with your AI Agent

Copy the full text content of this guide formatted for ingestion by LLMs and programming assistants (like Claude, Cursor, or ChatGPT).