Skilllm
Google AntigravityAgent SkillsAI AgentsLLM Development

How to Create Custom Agent Skills for Google Antigravity

Skilllm Team6 min read
Minimalist dark mode illustration with abstract geometric shapes representing custom agent skills for Google Antigravity

The landscape of software development is shifting from "writing code" to "orchestrating intelligence." Google’s new Antigravity IDE is leading this charge, offering an agent-first platform where AI doesn't just autocomplete your syntax—it plans, executes, and validates entire workflows.

But an agent is only as good as its training. While Antigravity comes powerful out of the box, the real magic happens when you extend its capabilities.

In this guide, we’ll walk you through exactly how to add and create custom Agent Skills for Google Antigravity, transforming your AI from a generalist assistant into a specialized expert for your unique codebase.

What Are Agent Skills?

In the context of Google Antigravity, a Skill is a specialized module that teaches the AI agent how to perform a specific task or follow a set of best practices.

Think of a Skill as a "comprehensive prompt package" that is only loaded when needed. Instead of stuffing your system prompt with 50 pages of documentation, Antigravity uses a file-based system to dynamically load skills based on the user's request.

A skill can be anything from:

  • A Protocol: "Here is exactly how we run database migrations in this team."
  • A Tool: "Use this Python script to generate a UUID."
  • A Style Guide: "Ensure all Git commit messages feature emojis and follow conventional commit standards."

The Anatomy of a Skill

Creating a skill is surprisingly simple. It doesn't require compiling code or complex APIs. It is primarily file-based.

A typical skill lives in a folder and contains at least one required file: SKILL.md.

TEXT
my-new-skill/
├── SKILL.md          (Required: The instructions)
├── scripts/          (Optional: Python/Bash scripts)
└── templates/        (Optional: Code boilerplates)

1. The SKILL.md File

This is the brain of your skill. It tells the agent what the skill is and how to execute it. It starts with YAML frontmatter followed by natural language instructions.

Example SKILL.md:

SKILL.md
---
name: git-emoji-commit
description: Generates Git commit messages using emojis and the Conventional Commits specification.
---

# Instructions
When the user asks to "commit changes" or "write a commit message":
1. Analyze the staged changes (git diff --staged).
2. Determine the type of change (feat, fix, chore, docs, style, refactor).
3. Select an appropriate emoji (e.g., 🐛 for fixes, ✨ for features).
4. Generate the message in the format: "<emoji> <type>: <description>"

Step-by-Step: How to Create Your First Skill

Ready to build? You can add skills in two different scopes depending on your needs.

Step 1: Choose Your Scope

  • Workspace Scope (Project Specific):
    Use this for skills that only apply to the current project (e.g., a specific deployment script).
    Path: <workspace-root>/.agent/skills/
  • Global Scope (User Specific):
    Use this for tools you want available in every project you open (e.g., your personal coding style).
    Path: ~/.gemini/antigravity/skills/ (on Mac/Linux)

Step 2: Create the Folder Structure

Navigate to the directory chosen above and create a new folder. The folder name usually acts as the default ID for the skill.

BASH
mkdir -p .agent/skills/security-auditor

Step 3: Write the Definition

Create a SKILL.md file inside that folder. Be sure to write a clear description in the YAML header.

Pro SEO Tip for Agents: The description field is effectively "SEO for the AI." The Antigravity agent uses semantic search on this description to decide if it should use your skill. If your description is vague, the agent will ignore it.

Step 4: Add Executable Scripts (Optional)

If your skill needs to do something complex (like query a database or scrape a web page), you can include scripts.

Create a scripts folder inside your skill directory and add a Python or Bash script. In your SKILL.md, simply instruct the agent to run it:

"To audit the dependencies, run the script located at scripts/audit_deps.py and analyze the output."

Where to Find High-Quality Agent Skills

Building every skill from scratch can be time-consuming. Just as you wouldn't write every library you use, you shouldn't have to write every agent capability.

At Skilllm, we are building the definitive library for this new ecosystem.

If you are looking for ready-to-deploy capabilities, check out our curated collection of Agent Skills. We host a variety of verified skills ranging from automated testing protocols to advanced architectural planning modules compatible with Antigravity and other LLM agent frameworks.

Best Practices for "Skill Engineering"

To ensure your Antigravity agent performs reliably, follow these three golden rules:

  1. Atomic Design: Don't create one massive "Do Everything" skill. Break them down. Create one skill for "React Component Generation" and another for "Unit Testing."
  2. Constraint-Based Prompting: In your SKILL.md, explicitly tell the agent what not to do. (e.g., "Do not remove existing comments.")
  3. Examples are King: LLMs learn best from examples. Include a section in your markdown file labeled ## Examples showing a sample Input and the desired Output.

Conclusion

Google Antigravity is changing the developer experience, but Agent Skills are what make it truly yours. By curating a library of skills, you are effectively programming the programmer, automating the repetitive cognitive loads of your day.

Start building your library today, or browse our marketplace to fast-track your agent's evolution.

Want to supercharge your AI workforce? Discover more at Skilllm.com.