Give agents reusable skills

Teach an agent a reusable capability once — a "skill" with instructions, tools, and reference files — and hand it to any agent that needs it, instead of bloating every persona.

What you'll use

The Skills framework in AI Studio. A skill bundles a set of instructions, an allowed toolset, and reference material; agents load it on demand (progressive disclosure) so they only carry the context they need for the task at hand.

Steps

1. Go to AI Studio → Build → Skills and create a skill — give it a name, a description, the instructions the agent should follow, and the tools it's allowed to use while the skill is active. You can attach reference files too.
2. Already have a skill written as a SKILL.md? Import it directly.
3. Assign the skill to an agent in the agent wizard/editor (the Skills picker). An agent can hold several skills.
4. At work time the agent calls load_skill to pull a skill's instructions and activate its toolset only when relevant — keeping prompts lean and behavior focused.
5. Prefer not to build from scratch? Install a prebuilt skill from the catalog — e.g. deep research, diagram, infographics, or office documents.

Result

Capabilities become modular and shareable: define "how to write a competitive brief" or "how to produce a slide deck" once, and every agent can use it consistently. Skills support multiple tiers (native, JavaScript, Python, and SiloLink-backed) so the same mechanism scales from a simple instruction set to a registered Python script.

Python skills run only admin-approved, registered scripts — agents don't write or execute arbitrary code, so the power stays governed.