Skills
Written procedures the built-in assistant follows, so you stop re-explaining them.
Contents
What is a skill?
A skill is a short document your workspace writes to tell the built-in assistant how to do something. It might say how your team names things, what a quarter means at your company, the steps for the report you rebuild every month, or which source to trust when two disagree. You write it once, and the assistant follows it whenever it applies.
The skills list is under Settings → Assistant → Skills. Turn on Show in sidebar there to add a Skills row to your sidebar in that workspace.
A skill has a name, a description and a body, and the editor saves as you type. See Writing a skill.
How does the assistant use a skill?
You can call a skill yourself. Type / in any composer and pick one, and its body goes in ahead of
your message.
/Explain why Brazil's GDP per person moved after 2014
The assistant also reaches for skills on its own. It is given the names and descriptions of up to 100 skills you can read, and picks the right one when your question matches. A skill nobody calls by name is found by its description alone.
Your agent can find and read skills too, with list_skills (a query searches them) and
get_skill. See the skill tool reference.
Who can read and edit a skill?
Every full member of the workspace can read and edit a skill by default. To narrow that, open Settings → Sharing on the skill, then name people or change the workspace baseline.
A guest, invited by email to specific objects, sees a skill only with an explicit grant and can never edit it. See Members.
Turn on public access and anyone with the skill's address can read it, as with a dashboard. The public view shows the document. See How sharing works.
What happens when an agent writes a skill?
Your agent can write a skill with save_skill, and the change saves at once with no confirmation. The
built-in assistant asks first on a confirmation card, because a skill steers every conversation it
applies to.
A skill written or changed by your agent or the built-in assistant is used only in its author's own chats at first. It reaches everyone else's chats once a person edits it in the skill editor. A brand new skill works the same way until someone types in it. The hold means neither your agent nor the built-in assistant can quietly install instructions that steer the whole workspace.