Learn how Google DeepMind is using "agent skills" to bridge the knowledge gap between static LLMs and fast-evolving SDKs. See how the new Gemini API developer skill boosted performance from 28.2% to 96.6% by providing models with real-time documentation and updated coding primitives.

GoogleDevs' platform is a central hub for developers interested in Google technologies, APIs, and developer tools. Through articles, tutorials, and documentation, GoogleDevs offers insights into building applications using Google Cloud Platform, Android, Chrome, and other Google services. Developers can learn about cloud computing, machine learning, and mobile app development with Google's developer tools and platforms.

Google Developers

Google DeepMind engineers built a 'Gemini API developer skill' to address the knowledge gap between static LLM training data and fast-evolving SDKs. By providing coding agents with real-time documentation references and updated SDK primitives via a lightweight skill, they boosted task pass rates from a baseline of 28.2% (Gemini 3.1 Pro) up to 96.6%. The skill is installable via Vercel or Context7 and was evaluated against 117 prompts covering agentic coding, chatbots, streaming, and document processing. While results are promising, limitations include the lack of automatic skill updates and evidence that direct AGENTS.md instructions can sometimes outperform skills.

Closing the knowledge gap with agent skills