AI/ML Skills for Gemini Agent Platform: Complete 2024 Developer Guide

The Gemini Agent Platform provides 15+ specialized AI/ML skills covering model APIs, tuning, deployment, evaluation, and operational management, each defined in standalone SKILL.md files under /skills/cloud/ in the google/skills repository.

The google/skills repository contains a comprehensive skill library for building production-grade AI applications on Google Cloud. Each skill is a self-contained module with structured documentation, executable examples, and integration patterns. This guide catalogs every AI/ML skill available for the Gemini Agent Platform, organized by functional area with source file references and practical code samples.

Core Gemini Model API Skills

Three foundational skills expose the primary interfaces for interacting with Gemini models programmatically.

Gemini API

The Gemini API skill (/skills/cloud/gemini-api/SKILL.md) provides the baseline client patterns for synchronous model calls, multimodal inputs, function calling, and batch prediction workflows. This is the entry point for most Agent Platform integrations.

from google.cloud import aiplatform_v1beta1 as aiplatform

client = aiplatform.PredictionServiceClient()
model_name = "projects/PROJECT_ID/locations/us-central1/publishers/google/models/gemini-pro"
request = aiplatform.GenerateContentRequest(
    model=model_name,
    contents=[{"role": "user", "parts": [{"text": "Explain the difference between Gemini Pro and Gemini Flash"}]}],
)
response = client.generate_content(request=request)
print(response.candidates[0].content.parts[0].text)

Gemini Live API

The Gemini Live API skill (/skills/cloud/gemini-live-api/SKILL.md) implements WebSocket-based streaming for low-latency, bidirectional interactions. Use this for real-time applications requiring continuous audio or text exchange.

gemini_live_api start \
  --model=gemini-1.5-flash \
  --project=$PROJECT_ID \
  --region=us-central1 \
  --session_id=my-session

Gemini Interactions API

The Gemini Interactions API skill (/skills/cloud/gemini-interactions-api/SKILL.md) manages stateful, multi-turn conversations with structured output schemas and persistent context across sessions.

Agent Lifecycle Management Skills

Gemini Agents API

The Gemini Agents API skill (/skills/cloud/gemini-agents-api/SKILL.md) exposes CRUD operations for custom Agent resources: create, list, update, delete, and metadata management.

Agent Platform Skill Registry

The Agent Platform Skill Registry skill (/skills/cloud/agent-platform-skill-registry/SKILL.md) handles skill registration, discovery, and versioning so agents can dynamically invoke capabilities.

{
  "skillId": "my-data-cleaning-skill",
  "description": "Cleans CSV files before feeding them to Gemini",
  "runtime": "python3.11",
  "entrypoint": "clean_data.py"
}

curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  -d @payload.json \
  https://us-central1-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/us-central1/agents/AGENT_ID/skills

Agent Platform Prompt Management

The Agent Platform Prompt Management skill (/skills/cloud/agent-platform-prompt-management/SKILL.md) provides versioning, templating, and runtime injection of prompts for consistent model behavior across environments.

Model Development and Optimization Skills

Agent Platform Tuning

The Agent Platform Tuning skill (/skills/cloud/agent-platform-tuning/SKILL.md) orchestrates hyperparameter tuning for Gemini and compatible open models using the platform's managed training service.

gcloud ai custom-jobs create \
  --region=us-central1 \
  --display-name="gemini-tuning-job" \
  --config=./skills/cloud/agent-platform-tuning/scripts/tune_open_model.py

Agent Platform Model Registry

The Agent Platform Model Registry skill (/skills/cloud/agent-platform-model-registry/SKILL.md) manages the publication and reference of custom or pre-trained models with artifact versioning and lineage tracking.

Agent Platform Inference

The Agent Platform Inference skill (/skills/cloud/agent-platform-inference/SKILL.md) supplies SDK wrappers spanning Vertex AI, OpenAI-compatible endpoints, and GenAI SDKs for unified model invocation.

Evaluation and Quality Assurance Skills

Agent Platform Evaluation (Flywheel)

The Agent Platform Evaluation (Flywheel) skill (/skills/cloud/agent-platform-eval-flywheel/SKILL.md) implements the Flywheel framework for systematic model assessment, including dataset curation, metric computation, and HTML report generation.

Production Operations Skills

Agent Platform Deploy

The Agent Platform Deploy skill (/skills/cloud/agent-platform-deploy/SKILL.md) automates container image builds and deployments to Cloud Run, GKE, and private networking configurations.

Agent Platform Endpoint Management

The Agent Platform Endpoint Management skill (/skills/cloud/agent-platform-endpoint-management/SKILL.md) controls custom serving endpoints with traffic splitting, canary releases, and version routing.

Agent Platform Alert Configuration

The Agent Platform Alert Configuration skill (/skills/cloud/agent-platform-alert-configuration/SKILL.md) configures monitoring alerts for traffic anomalies, latency thresholds, cost budgets, and safety policy violations.

Agent Platform Troubleshooting

The Agent Platform Troubleshooting skill (/skills/cloud/agent-platform-troubleshooting/SKILL.md) provides diagnostic runbooks for gateway errors, identity misconfigurations, IAM policies, Model Armor conflicts, and IAP issues.

Migration and Modernization Skills

Agent Platform Migrate from AI Studio

The Agent Platform Migrate from AI Studio skill (/skills/cloud/agent-platform-migrate-from-ai-studio/SKILL.md) delivers step-by-step guidance for transitioning applications from the legacy Google AI Studio Gemini API to the Agent Platform's enterprise architecture.

Summary

• 15 production-ready AI/ML skills are maintained in google/skills for the Gemini Agent Platform, each as a standalone module with documentation and examples.
• Core API skills (Gemini API, Live API, Interactions API) provide tiered access patterns from synchronous calls to streaming WebSockets.
• Lifecycle skills manage agents, prompts, skills, and models through their full operational lifespan.
• Development skills enable tuning, evaluation, and systematic quality assurance via the Flywheel framework.
• Production skills cover deployment, endpoint management, monitoring, and incident response.
• Migration support is explicitly provided for teams transitioning from AI Studio.

Frequently Asked Questions

What is the difference between the Gemini API and Gemini Live API skills?

The Gemini API skill implements standard HTTP-based request/response patterns suitable for batch processing and synchronous applications. The Gemini Live API skill operates over WebSockets for bidirectional streaming with sub-second latency, designed for real-time conversational interfaces. Both are defined in /skills/cloud/gemini-api/SKILL.md and /skills/cloud/gemini-live-api/SKILL.md respectively.

How do I register a custom skill for agents to use?

Use the Agent Platform Skill Registry skill via the API endpoint https://{region}-aiplatform.googleapis.com/v1/projects/{PROJECT_ID}/locations/{region}/agents/{AGENT_ID}/skills. Submit a JSON payload with skillId, description, runtime, and entrypoint fields as documented in /skills/cloud/agent-platform-skill-registry/SKILL.md.

Where can I find migration guidance from Google AI Studio?

The Agent Platform Migrate from AI Studio skill at /skills/cloud/agent-platform-migrate-from-ai-studio/SKILL.md contains API mapping tables, authentication changes, and code transformation examples for modernizing legacy implementations.

What evaluation capabilities does the platform provide?

The Agent Platform Evaluation (Flywheel) skill implements dataset management, automated metric computation (BLEU, ROUGE, custom), and HTML report generation for systematic model performance assessment, documented in /skills/cloud/agent-platform-eval-flywheel/SKILL.md.

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