How to Make Skills Compatible Across Multiple AI Assistants Using Antigravity Awesome Skills

Antigravity Awesome Skills enables cross-compatibility by using a single SKILL.md file with standardized YAML front-matter that every supported assistant reads identically, eliminating the need for assistant-specific code paths.

The sickn33/antigravity-awesome-skills repository defines a portable skill format that works across Claude Code, Gemini CLI, Codex CLI, Kiro IDE, Cursor, OpenCode, and AdaL CLI without modification. By adhering to a strict schema and using a universal installer, you can write a skill once and deploy it everywhere.

The SKILL.md Standard: A Universal Format

Cross-compatibility starts with the standardized skill schema. Every skill resides in skills/<skill-name>/SKILL.md and must contain required front-matter keys: name, description, risk, source, and date_added.

The parser in tools/lib/skill-utils.js extracts these values through the parseFrontmatter → readSkill function chain (lines 11-18). Because the parser normalizes tags and validates the YAML mapping once, the resulting skill object behaves identically regardless of which assistant consumes it.

Three-Layer Architecture for Cross-Assistant Compatibility

The repository enforces compatibility through three distinct layers:

Standardized Skill Schema

The file tools/lib/skill-utils.js validates that each SKILL.md contains the required front-matter keys and rejects malformed YAML. This ensures that skills like skills/brainstorming/SKILL.md follow the exact same structure whether used in Claude Code or Gemini CLI.

Universal Installer

The CLI installer in tools/bin/install.js detects target assistants via flags (--claude, --gemini, --codex, --open-code, etc.) and copies the entire skills/ tree into assistant-specific directories (lines 173-219). The installer creates:

• ~/.claude/skills/ for Claude Code
• ~/.gemini/skills/ for Gemini CLI
• ~/.codex/skills/ for Codex CLI
• ~/.agents/skills/ for OpenCode
• ~/.adal/skills/ for AdaL CLI

The installer also generates the Claude Code marketplace entry at .claude-plugin/marketplace.json.

Assistant-Agnostic Usage

According to docs/users/usage.md, users invoke skills using the generic pattern @skill-name. Because the skill name derives from the front-matter name field, the same invocation syntax works across all environments without per-assistant customization.

Creating Your First Cross-Compatible Skill

To make skills compatible across multiple AI assistants, follow this workflow:

1. Create the markdown file at skills/<my-skill>/SKILL.md with the required front-matter.
2. Write pure markdown content—the security tests in tools/scripts/tests/docs_security_content.test.js explicitly reject shell pipelines like | bash or | iex to ensure safe cross-platform execution.
3. Run the installer to deploy to all supported assistants.

Here is a minimal example that works everywhere:

---
name: hello-world
description: "Print a friendly greeting. No external tools, pure markdown."
risk: low
source: community
date_added: "2026-03-01"
---

# Hello-World Skill

You are a helpful assistant. When invoked, simply respond with:

Hello, world! 🚀

Use this skill when you need a quick sanity-check or a friendly greeting.

Installing Skills for Multiple Assistants

Run the universal installer to deploy skills to every supported assistant simultaneously:

# Installs the whole repository to all detected assistant directories

npx antigravity-awesome-skills

This single command executes the logic in tools/bin/install.js, populating all assistant-specific paths with the same SKILL.md files. The installer handles OS detection and directory creation automatically.

Invoking Skills Across Different Assistants

Because the skills are assistant-agnostic, invocation differs only in the surface syntax, not the underlying behavior:

• Claude Code: Type @hello-world
• Gemini CLI: Type Use @hello-world
• OpenCode: Run opencode run @hello-world

All three commands cause the assistant to read the identical SKILL.md from their respective skill directories and execute the same instructions.

Summary

• Single file format: The SKILL.md standard with YAML front-matter provides the foundation for cross-compatibility.
• Universal validation: tools/lib/skill-utils.js parses and normalizes skills once, ensuring consistent behavior.
• Automated distribution: tools/bin/install.js deploys to .claude/skills/, .gemini/skills/, and other directories without manual copying.
• Generic invocation: The @skill-name pattern works across Claude Code, Gemini CLI, Codex CLI, Cursor, OpenCode, and AdaL CLI.
• Security enforcement: Automated tests reject embedded shell commands, keeping skills safe across platforms.

Frequently Asked Questions

Do I need to write separate versions of a skill for each AI assistant?

No. Write a single SKILL.md following the repository's front-matter schema. The tools/lib/skill-utils.js parser extracts the skill definition once, and the tools/bin/install.js installer handles assistant-specific placement. The same markdown file works in Claude Code, Gemini CLI, Codex CLI, and others without modification.

What are the required front-matter keys for cross-compatibility?

Every SKILL.md must include five keys in the YAML header: name (the skill identifier), description (usage context), risk (security level), source (provenance), and date_added (ISO 8601 date). The parser in tools/lib/skill-utils.js validates these keys before the installer distributes the file.

Is the installer open source and extensible?

Yes. The installer source resides in tools/bin/install.js and supports adding new assistants by extending the target directory mappings (lines 173-219). The repository uses standard Node.js without proprietary dependencies, allowing community contributions for emerging AI assistants.

How does the repository prevent security issues in cross-platform skills?

The test suite in tools/scripts/tests/docs_security_content.test.js scans all SKILL.md files for dangerous patterns like | bash or | iex (PowerShell). Skills containing shell pipelines fail validation, ensuring that skills remain pure markdown instructions rather than executable scripts that could behave differently across operating systems.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}