# Which AI Agent Frameworks Are Compatible with Google Skills? A Complete Integration Guide

> Discover AI agent frameworks compatible with Google Skills. Learn integration with ADK, Gemini Enterprise, Google Agents CLI, and generic harnesses. Your complete guide.

- Repository: [Google/skills](https://github.com/google/skills)
- Tags: integration-guide
- Published: 2026-08-11

---

**Google Skills is framework‑agnostic and officially compatible with four AI agent ecosystems: Agent Development Kit (ADK), Gemini Enterprise Agent Platform, Google Agents CLI, and generic agent harnesses that parse Markdown‑based skill files.**

Google Skills is a repository of framework‑agnostic skill definitions written in Markdown that describe how to work with Google Cloud products. While these skills can theoretically be consumed by any system that reads Markdown, the google/skills repository is explicitly designed for four specific AI agent frameworks that provide official tooling, runtime environments, and discovery mechanisms.

## Agent Development Kit (ADK)

The **Agent Development Kit (ADK)** is the official SDK for building agents on the Gemini Enterprise Agent Platform. ADK agents can directly import a skill's [`SKILL.md`](https://github.com/google/skills/blob/main/SKILL.md) file, call the defined tools, and leverage built‑in memory and runtime capabilities.

In `skills/cloud/google‑cloud‑solution‑build‑deploy‑agents/SKILL.md`, the repository demonstrates how ADK loads skill tool specifications, generates proper API calls, and manages agent sessions. The ADK runtime replaces placeholder tool implementations with actual skill‑defined calls at execution time.

```python
from adk import Agent, tool

@tool
def list_buckets(project: str):
    """List Cloud Storage buckets for the given project."""
    # Implementation provided by skill; ADK replaces body at runtime

    pass

agent = Agent(name="my-storage-agent", tools=[list_buckets])

```

## Gemini Enterprise Agent Platform

The **Gemini Enterprise Agent Platform** (also referred to as Agent Platform) provides the managed runtime that ADK agents depend on. According to `product‑mappings.md` in the `google‑cloud‑solution‑build‑deploy‑agents` skill directory, this platform supplies:

- **Sessions** – conversational state management
- **Memory** – vector‑store for context retention
- **Tool sandbox** – secure execution environment for skill operations

Skills are interpreted as "agent‑aware" prompts that the platform invokes directly, bridging the gap between static Markdown definitions and live agent execution.

## Google Agents CLI

The **Google Agents CLI** (`agents-cli`) offers a command‑line workflow for scaffolding, publishing, and running agents. It integrates with Google Skills through the `npx skills add` command, which pulls skills from the repository and wires them to the selected runtime.

Install a skill using the CLI‑compatible installer:

```bash

# Install the Google Cloud "storage-basics" skill

npx skills add google/skills/cloud/google-cloud-storage-basics

```

Scaffold a new agent with integrated skills:

```bash

# Create agent structure with skill wiring

agents-cli scaffold my-storage-agent

```

The CLI automatically configures the agent to use either ADK or plain LLM runtimes based on your project settings.

## Generic Agent Harnesses

Any **agent harness** that can read Markdown‑based skill files can consume Google Skills without code modifications. The repository enables this through [`.agents/plugins/marketplace.json`](https://github.com/google/skills/blob/main/.agents/plugins/marketplace.json), a plugin manifest that allows auto‑discovery by third‑party tools.

Compatible harnesses include:

- **Claude Code** – Anthropic's agent coding environment
- **Codex** – OpenAI's code generation system
- **Antigravity CLI** – Emerging agent orchestration tools

These harnesses reference skills using a simple JSON configuration:

```json
{
  "plugin": "google/skills",
  "skill_path": "skills/cloud/google-cloud-storage-basics"
}

```

The harness reads [`SKILL.md`](https://github.com/google/skills/blob/main/SKILL.md), extracts `tool` definitions, and presents them as callable functions to the underlying LLM.

## Key Integration Files in google/skills

Understanding the repository structure helps clarify how each framework connects to Google Skills:

| File Path | Framework Purpose |
|-----------|-----------------|
| `skills/cloud/google‑cloud‑solution‑build‑deploy‑agents/SKILL.md` | ADK canonical skill with integration details |
| `skills/cloud/google‑cloud‑solution‑build‑deploy‑agents/references/product‑mappings.md` | Maps skill concepts to Gemini Enterprise Agent Platform components |
| [`.agents/plugins/marketplace.json`](https://github.com/google/skills/blob/main/.agents/plugins/marketplace.json) | Plugin manifest for generic harness discovery |
| `skills/cloud/google‑cloud‑solution‑build‑deploy‑agents/references/related‑guidance.md` | Additional ADK tooling documentation |

All files are available in the main branch of google/skills with stable paths for framework implementers to reference.

## How to Choose the Right Framework

Consider these criteria when selecting a Google Skills integration approach:

1. **ADK** – Choose when building production agents for Gemini Enterprise with full platform runtime support
2. **Gemini Enterprise Agent Platform** – Required as the underlying infrastructure; select tier based on scale needs
3. **Google Agents CLI** – Ideal for rapid prototyping and CI/CD pipelines
4. **Generic harnesses** – Use when operating outside Google Cloud ecosystems or requiring custom toolchains

## Summary

- **Four official integrations** support Google Skills: ADK, Gemini Enterprise Agent Platform, Google Agents CLI, and Markdown‑capable generic harnesses
- **ADK and the Agent Platform** provide the deepest integration with managed runtime, sessions, and memory
- **Google Agents CLI** streamlines skill installation via `npx skills add` and agent scaffolding
- **Generic harnesses** use [`.agents/plugins/marketplace.json`](https://github.com/google/skills/blob/main/.agents/plugins/marketplace.json) for zero‑code skill discovery
- All integrations rely on parsing [`SKILL.md`](https://github.com/google/skills/blob/main/SKILL.md) files that follow the repository's Markdown conventions

## Frequently Asked Questions

### Can I use Google Skills without the Gemini Enterprise Agent Platform?

Yes. While ADK and the Agent Platform provide the richest feature set, generic agent harnesses can consume Google Skills by parsing the Markdown skill files directly. You lose managed runtime features like built‑in memory and tool sandboxing but retain full access to skill definitions.

### What is the `npx skills add` command and where does it come from?

The `npx skills add` command is provided by the Google Agents CLI tool chain. It queries the google/skills repository, downloads the specified skill directory, and registers it with your local agent project. The source is referenced in the root [`README.md`](https://github.com/google/skills/blob/main/README.md) as the standard installation method.

### How do I know if my agent harness supports Google Skills?

Check whether your harness can parse Markdown skill files with tool definitions and whether it implements the discovery protocol defined in [`.agents/plugins/marketplace.json`](https://github.com/google/skills/blob/main/.agents/plugins/marketplace.json). If both are true, your harness should work with Google Skills without modification.

### Are Google Skills tied to specific Google Cloud API versions?

Individual skills specify their tested API versions within each [`SKILL.md`](https://github.com/google/skills/blob/main/SKILL.md) file. The framework‑agnostic design means skills are not locked to particular agent framework versions, but you should verify skill compatibility with your target Google Cloud services as you would with any API integration.