# AI/ML Skills on Google Cloud: Complete Guide to the Google Cloud Skills Repository

> Explore AI/ML skills on Google Cloud with the google skills repository. Automate deployments for GKE, Vertex AI, and Gemini APIs. Get your complete guide today.

- Repository: [Google/skills](https://github.com/google/skills)
- Tags: getting-started
- Published: 2026-09-04

---

**The Google Cloud Skills repository contains modular, self-contained SKILL definition files that automate deployment, troubleshooting, and code generation for AI/ML workloads across GKE, Vertex AI, and Gemini APIs.**

The [google/skills](https://github.com/google/skills) repository hosts a comprehensive collection of automation capabilities organized as [`SKILL.md`](https://github.com/google/skills/blob/main/SKILL.md) files and supporting reference documentation. These **AI/ML skills on Google Cloud** cover the entire machine learning lifecycle—from GPU cluster upgrades and TPU troubleshooting to generative AI application development using GenKit and the Vertex AI Agent Platform.

## Model Inference and Serving on GKE

The **`gke-inference`** skill, defined in [`/skills/cloud/gke-inference/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/gke-inference/SKILL.md), automates the deployment and optimization of AI/ML inference workloads on Google Kubernetes Engine (GKE). This skill supports GPU, TPU, and CPU-only node configurations, generating deployment manifests and returning service endpoints upon completion.

Reference material for CLI commands is available in [`references/cli-reference.md`](https://github.com/google/skills/blob/main/references/cli-reference.md) within the skill directory.

## Cluster Management for Large-Scale AI/ML

For organizations running heavy GPU or TPU workloads, the **`gke-upgrades`** skill at [`/skills/cloud/gke-upgrades/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/gke-upgrades/SKILL.md) provides upgrade-time best practices specifically tailored for large-scale AI/ML clusters. This ensures minimal disruption to training jobs and inference services during cluster maintenance windows.

## AI/ML Training Troubleshooting

The repository includes four specialized skills for diagnosing and recovering from training job failures:

- **`gke-ai-troubleshooting-jobset-interruption`** ([`/skills/cloud/gke-ai-troubleshooting-jobset-interruption/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/gke-ai-troubleshooting-jobset-interruption/SKILL.md)): Handles diagnosis, restart, and pre-emption of distributed training jobs using JobSet.

- **`gke-ai-troubleshooting-tpu-metrics-monitoring`** ([`/skills/cloud/gke-ai-troubleshooting-tpu-metrics-monitoring/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/gke-ai-troubleshooting-tpu-metrics-monitoring/SKILL.md)): Monitors TPU health metrics to detect hardware degradation or performance bottlenecks.

- **`gke-ai-troubleshooting-tpu-dynamic-slices-monitoring`** ([`/skills/cloud/gke-ai-troubleshooting-tpu-dynamic-slices-monitoring/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/gke-ai-troubleshooting-tpu-dynamic-slices-monitoring/SKILL.md)): Tracks TPU slice allocation and utilization for dynamic workload scheduling.

- **`gke-ai-troubleshooting-tpu-vbar-oom`** ([`/skills/cloud/gke-ai-troubleshooting-tpu-vbar-oom/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/gke-ai-troubleshooting-tpu-vbar-oom/SKILL.md)): Automates recovery from TPU vBar out-of-memory events, a critical failure mode in large model training.

## Generative AI Frameworks (GenKit)

**GenKit** skills provide end-to-end libraries for building, prompting, and evaluating generative AI applications across multiple programming languages:

- **GenKit Python**: [`/skills/cloud/genkit-python/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/genkit-python/SKILL.md) (references: [`references/setup.md`](https://github.com/google/skills/blob/main/references/setup.md), [`references/generation.md`](https://github.com/google/skills/blob/main/references/generation.md))
- **GenKit JavaScript/TypeScript**: [`/skills/cloud/genkit-js/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/genkit-js/SKILL.md)
- **GenKit Go**: [`/skills/cloud/genkit-go/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/genkit-go/SKILL.md)
- **GenKit Dart**: [`/skills/cloud/genkit-dart/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/genkit-dart/SKILL.md)

These skills abstract interactions with Gemini models, Vertex AI, and evaluation frameworks, enabling consistent implementation regardless of language choice.

## Gemini Generative AI APIs

For direct model interaction, three distinct API skills are available:

- **`gemini-api`** ([`/skills/cloud/gemini-api/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/gemini-api/SKILL.md)): Legacy interface for standard Gemini model calls.
- **`gemini-interactions-api`** ([`/skills/cloud/gemini-interactions-api/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/gemini-interactions-api/SKILL.md)): Enhanced interface supporting chat and function calling patterns.
- **`gemini-live-api`** ([`/skills/cloud/gemini-live-api/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/gemini-live-api/SKILL.md)): Streaming interface for real-time response generation and low-latency applications.

## Vertex AI Agent Platform

The Vertex AI Agent Platform skills manage the complete lifecycle of Retrieval-Augmented Generation (RAG) agents:

- **`agent-platform-inference`** ([`/skills/cloud/agent-platform-inference/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/agent-platform-inference/SKILL.md)): Handles model serving and endpoint management (reference: [`references/agent-identity.md`](https://github.com/google/skills/blob/main/references/agent-identity.md)).
- **`agent-platform-tuning`** ([`/skills/cloud/agent-platform-tuning/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/agent-platform-tuning/SKILL.md)): Manages fine-tuning workflows for domain-specific model adaptation.
- **`agent-platform-rag-engine-management`** ([`/skills/cloud/agent-platform-rag-engine-management/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/agent-platform-rag-engine-management/SKILL.md)): Provisions and configures RAG engines with vector store integration (reference: [`references/create.md`](https://github.com/google/skills/blob/main/references/create.md)).
- **`agent-platform-prompt-management`** ([`/skills/cloud/agent-platform-prompt-management/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/agent-platform-prompt-management/SKILL.md)): Version control and deployment of prompt templates.
- **`agent-platform-deploy`** ([`/skills/cloud/agent-platform-deploy/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/agent-platform-deploy/SKILL.md)): End-to-end deployment and endpoint management.
- **`agent-platform-alert-configuration`** ([`/skills/cloud/agent-platform-alert-configuration/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/agent-platform-alert-configuration/SKILL.md)): Monitoring and alerting setup for production agents.

## Storage Optimized for AI/ML Workloads

High-performance storage configuration is critical for model checkpoints and dataset access:

- **`google-cloud-storage-bucket-architect`** ([`/skills/cloud/google-cloud-storage-bucket-architect/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/google-cloud-storage-bucket-architect/SKILL.md)): Configures buckets for AI/ML checkpointing with hierarchical namespaces and rapid cache settings (reference: [`references/phase_draft_plan.md`](https://github.com/google/skills/blob/main/references/phase_draft_plan.md)).
- **`google-cloud-storage-basics`** ([`/skills/cloud/google-cloud-storage-basics/SKILL.md`](https://github.com/google/skills/blob/main//skills/cloud/google-cloud-storage-basics/SKILL.md)): General high-performance storage guidance for AI/ML data pipelines.

## Developer Knowledge and Documentation Access

The **`retrieving-developer-knowledge`** skill ([`/skills/developers/retrieving-developer-knowledge/SKILL.md`](https://github.com/google/skills/blob/main//skills/developers/retrieving-developer-knowledge/SKILL.md)) provides programmatic access to official AI/ML documentation from ai.google.dev, ADK, and TensorFlow via MCP server or REST fallback (reference: [`references/supported-domains.md`](https://github.com/google/skills/blob/main/references/supported-domains.md)).

## Practical Implementation Examples

Below are executable snippets demonstrating how to invoke these AI/ML skills within the Google Cloud Developer Assistant environment.

### Deploying AI/ML Inference on GKE

```json
{
  "skill": "gke-inference",
  "args": {
    "project_id": "my-gcp-project",
    "region": "us-central1",
    "model_name": "gemini-1.5-flash-001",
    "hardware": "GPU"
  }
}

```

This payload creates a GKE cluster if needed, builds the deployment manifest, and returns the service URL for model serving.

### Running GenKit Python Generation

```python
import genkit as gk
import genkit.plugins.vertexai as vertex

vertex.configure(project="my-gcp-project", location="us-central1")
prompt = "Summarize the latest research on transformer scaling laws."
response = gk.run(vertex.generate(prompt=prompt, model="gemini-1.5-flash-001"))
print(response.text)

```

The **GenKit Python** skill streams the response from the Vertex AI Gemini endpoint and returns structured output.

### Creating a RAG Engine with Vertex AI

```python
import vertexai
from vertexai.preview import rag

vertexai.init(project="my-gcp-project", location="us-central1")
rag_engine = rag.RagEngine(
    name="my-rag-engine",
    vector_store=rag.VectorStore(name="my-vector-store"),
    retrieval_config=rag.RetrievalConfig(top_k=5)
)
rag_engine.create()
print(f"Engine created: {rag_engine.resource_name}")

```

This invokes the **Agent-Platform RAG Engine Management** skill to provision a retrieval engine linked to the specified vector store.

## Summary

- The **google/skills** repository organizes AI/ML capabilities into discrete, documented [`SKILL.md`](https://github.com/google/skills/blob/main/SKILL.md) files covering GKE inference, training troubleshooting, GenKit frameworks, and Vertex AI management.
- **Infrastructure skills** target GPU/TPU cluster upgrades (`gke-upgrades`) and high-performance storage configuration (`google-cloud-storage-bucket-architect`).
- **Operational skills** provide automated recovery for TPU OOM events and JobSet interruptions.
- **Development skills** include multi-language GenKit bindings and comprehensive Vertex AI Agent Platform lifecycle management.
- Each skill references specific implementation files (e.g., [`references/generation.md`](https://github.com/google/skills/blob/main/references/generation.md), [`references/create.md`](https://github.com/google/skills/blob/main/references/create.md)) for detailed command syntax and configuration options.

## Frequently Asked Questions

### How do I deploy a model inference service using the GKE skills?

Invoke the `gke-inference` skill by sending a JSON payload specifying your project ID, region, model name, and hardware type (GPU, TPU, or CPU) to the Google Cloud Developer Assistant. The skill automatically provisions GKE resources and returns the service endpoint URL.

### What troubleshooting capabilities are available for TPU training failures?

The repository provides four dedicated TPU troubleshooting skills: metrics monitoring for hardware health, dynamic slice monitoring for resource allocation, JobSet interruption handling for distributed training recovery, and vBar OOM management for automatic recovery from memory exhaustion events.

### How do the Vertex AI Agent Platform skills differ from the Gemini API skills?

**Vertex AI Agent Platform** skills (`agent-platform-*`) manage full RAG agent lifecycles including vector stores, prompt versioning, and deployment infrastructure, while **Gemini API** skills (`gemini-api`, `gemini-interactions-api`, `gemini-live-api`) provide low-level model interfaces for direct text generation, chat, and streaming without the RAG orchestration layer.

### Where can I find setup instructions for the GenKit Python skill?

Configuration details for GenKit Python are located in [`references/setup.md`](https://github.com/google/skills/blob/main/references/setup.md) within the `/skills/cloud/genkit-python/` directory, covering Vertex AI project initialization, plugin installation, and environment configuration requirements.