Agent Skills now available in Genkit Go

Genkit Agent Skills add specialized capabilities with progressive disclosure for agents Cut token use, simplify distribution, and build modular workflows in Go and beyond

Google Developers Blog announced support for Agent Skills in Genkit Go, along with TypeScript, Dart, and Python. The update focuses on helping developers add specialized capabilities to agent workflows without loading all reference material into the model context at once. The post explains that Agent Skills use progressive disclosure: Genkit initially exposes only skill metadata, then loads the full instructions only when a request matches a skill description. The article describes the skill file structure, the role of SKILL.md, and how bundled scripts and reference files can be included for more focused execution. The blog also shows how the feature works inside Genkit’s middleware system, with examples for registering skills in a flow and using them in tasks such as recipe generation and image restoration. It highlights practical benefits such as lower token use, simpler distribution, and more modular agent behavior.