🛠️ @genai-sdk for learning how to use the latest Google Gen AI SDK.
🚀 @vertex-deploy for deploying Open Models from Model Garden directly to endpoints.
🎯 @vertex-tuning for fine-tuning models with your own data.
⚡ @vertex-inference for executing Generative AI inference.

⚙️ Tool Configuration

These skills follow the universal SKILL.md format and work with any AI coding assistant that supports agentic skills.

Tool	Type	Invocation Example	Path
Gemini CLI	CLI	`(User Prompt) Use skill-name...`	`.gemini/skills/`
Antigravity	IDE	`(Agent Mode) Use skill...`	- Global:`~/.gemini/antigravity/skills/` - Workspace: `.agent/skills/`
Claude Code	CLI	`>> /skill-name help me...`	`.claude/skills/`
Cursor	IDE	`@skill-name (in Chat)`	`.cursor/skills/`

Architecture & Routing

These skills are designed hierarchically to guide the LLM agent from a broad user intent down to the specific technical implementation steps.

Primary Router: `vertex-ai`

The entry point for any general Vertex AI task. The vertex-ai/SKILL.md file acts as a traffic controller. When an agent receives a generic request (e.g., "I want to use Vertex AI"), it reads this file to determine the next step based on a Decision Tree:

Deploying a model → Routes to vertex-deploy
Using Gemini API & Gen AI SDK → Routes to genai-sdk
Generating text, chat, or embeddings → Routes to vertex-inference
Fine-tuning a model → Routes to vertex-tuning

🧰 Sub-Skills Directory

Category / Skill	Purpose	Capabilities
`genai-sdk`	Guides the usage of Gemini API with the Gen AI SDK across multiple languages.	Core inference, Live API, function calling, structured output, caching, and batch prediction.
`vertex-deploy`	Instructions and bash scripts for deploying Open Models or custom weights to a dedicated Endpoint.	`gcloud ai model-garden models deploy`, checking status, cost estimates, undeploying.
`vertex-inference`	Code samples and instructions for authenticating and executing Generative AI inference.	First-Party (Gemini) via `google-genai` and Third-Party OpenMaaS via OpenAI SDK.
`vertex-tuning`	A secondary router specifically for model fine-tuning.	Directs the agent to specific tuning procedures for Open Models or First-Party Gemini models.

How It Works Under The Hood

Agent skills are essentially specialized prompt contexts. When building agents (like those using Semantic Router, LangChain, or other LLM orchestration frameworks), the agent is instructed to read the relevant SKILL.md files.

Trigger: The agent recognizes a keyword (like "Vertex AI") and reads skills/vertex-ai/SKILL.md.
Determine Intent: The LLM processes the user prompt against the conditions in the Router's Decision Tree.
Navigate: The LLM actively uses file-reading tools to navigate to the correct underlying SKILL.md (e.g., skills/vertex-inference/SKILL.md).
Execute: The agent reads the highly-detailed instructions, caveats, and code snippets in the sub-skill to fulfill the user's request.

Installation & Usage

To use these skills with your AI agent:

Clone the repository (without checking out files yet):

git clone --no-checkout https://github.com/GoogleCloudPlatform/vertex-ai-samples.git

Enable sparse checkout in cone mode:

cd vertex-ai-samples
git sparse-checkout init --cone

Specify the skills folder:
```
git sparse-checkout set skills
```
Check out the branch:
```
git checkout main
```
Expose to Agent: Ensure the skills/ directory is accessible to your AI assistant. This typically involves mounting or copying the folder into your agent's specific custom routing directory. See the Tool Configuration section above for the correct path for your tool.
Trigger the Skills: Prompt your agent with a Vertex AI related task. The agent should automatically scan for skills, find the primary vertex-ai/SKILL.md router, and follow the documented workflow.

Example Prompts

Here are some example prompts you can use to trigger the routing logic and test the skills:

Testing Deployment:

"I want to deploy a Llama 3.3 model from Model Garden to a Vertex AI endpoint. Can you help me write the script?"

Testing Inference (Gen AI SDK):

"Can you show me how to connect to Vertex AI and get text embeddings using the new Gen AI SDK?"

Testing Tuning:

"I need to fine-tune a Gemini 2.5 Pro model using Vertex AI. Where should I start?"

How to Add a New Skill

If Vertex AI introduces a new capability (e.g., Vertex AI Search or Vertex AI Pipelines) and you want to teach the agent to handle it:

Create the Skill Directory:

mkdir vertex-search
touch vertex-search/SKILL.md

Write the Instructions (SKILL.md):
- Add frontmatter (name and description).
- Write clear, step-by-step instructions.
- Include code snippets and strict rules using markdown alerts (e.g., > [!WARNING]).
Update the Router:
- Open vertex-ai/SKILL.md.
- Add a new branch to the "Workflow Decision Tree" that tells the LLM when to route to ../vertex-search/SKILL.md.

Best Practices for Authoring Skills

Be Directive: Use imperative verbs ("Do this", "You MUST check that"). AI agents perform best when instructions are explicitly strict.
Use Code Samples: Scripts prevent the agent from hallucinating SDK syntax. Always include fully functional snippets.
Handle Errors: Include a Troubleshooting section in your skills. If a quota error is common, tell the LLM exactly how to interpret a 429 error and explain it to the user.

README.md

🤖 Agent Skills for Google Cloud Vertex AI

Table of Contents

🌟 Top Starter Skills