agent-platform-inference

What problem does it solve?

This Skill eliminates the hassle of manually configuring connections to Google Cloud Agent Platform GenAI models, handling authentication across multiple SDKs, setting up correct endpoints, and diagnosing common inference errors like 429 Resource Exhausted, 400 User Validation, and 404 Not Found, which typically require sifting through scattered documentation.

Core Features & Use Cases

• Multi-Model Inference Support: Generate code for calling both first-party Gemini models and third-party OpenMaaS models (Llama, DeepSeek, Qwen, etc.) on Google Cloud Agent Platform.
• Cross-SDK Compatibility: Provides guidance and example code for using the GenAI SDK, OpenAI SDK, and legacy Vertex AI SDK for model calls.
• Error Troubleshooting: Includes step-by-step fixes for common inference errors, including DSQ quota management for 429 errors and region validation for 404 errors.
• Use Case: A developer needs to call a DeepSeek model via the OpenAI SDK for a production workload, encounters a 429 Resource Exhausted error, and uses this Skill to implement exponential backoff and verify the correct global endpoint configuration.

Quick Start

Use the agent-platform-inference skill to generate Python code for calling the gemini-2.5-pro model via the GenAI SDK with proper Google Cloud authentication and endpoint setup.