Google wires Agent Skills into its Genkit framework for on-demand skill packs
- AI agent manuals keep piling up, and cramming them all into context burns tokens and distracts the model. Google wired Agent Skills into its own framework, Genkit: the manuals stay on disk, and the right one loads only when needed.
- The skills standard started with Anthropic, and Google hooked it into all four language versions. Enabling it takes one line of code.
- Google's official test image is the original painting of "Monkey Jesus" (disfigured in 2012). The model correctly identifies it as an oil painting and summons the matching restoration skill.
Stuffing the manual into the AI's head is a losing game
The Google Developers Blog just published a refreshingly practical tutorial: Genkit, their open-source AI application framework, now officially supports Agent Skills — the "AI skill pack" format started by Anthropic and now open as an industry standard. The short version: you used to have to shove every operating manual into the prompt, but now the manuals stay on disk and only the one you need gets loaded. The tutorial includes two directly runnable Go examples: a recipe bot, and an AI art restorer that picks its own restoration approach.
Every agent builder knows this problem: the more tasks your agent takes on, the more manuals you have to hand it—operating procedures, reference guides, assorted docs. Keep all of it resident in the context window and three problems hit at once: tokens burn fast (every request re-reads the whole pile), attention dilutes (the few lines that matter get buried in hundreds of pages), and the odds of a wrong answer actually climb. It's like making an employee memorize the entire corporate rulebook every morning when they'll only need one page all day.
Agent Skills is the answer: package expertise into individual skills that sit in the background until the moment they're needed. If you've used Skills in Claude Code, this will feel familiar—same format, and Anthropic has opened it up as a standard (agentskills.io). Genkit now plugs into that same spec. We've previously broken down Anthropic's official take on using Skills to build a validation loop:
A skill is just a folder
So what does an "on-demand" skill actually look like? Surprisingly simple: a skill is a folder containing exactly one required file, SKILL.md. That file has two parts. At the top is the frontmatter — a small block of metadata wrapped in --- that says what the skill is called and when to reach for it. Below that is the body: the actual operating instructions. The folder can also hold three types of optional resources: scripts/ (executable code), references/ (documentation), and assets/ (templates or other resources).
