Skills
Entities answer what is this thing; briefs answer in what frame should it be read. A skill answers how should this work be done — a packaged procedure with ordered steps, stop conditions, and a defined output shape, retrieved by an agent when the task at hand matches. A skill proposes; it never applies.
Skills live in Build → Skills — filtered by domain, with + to write a new one; search reaches the procedure text itself, so a remembered step finds its skill. Click one to open its page: Procedure with its output shape as markdown, Meaning for the synonyms the catalog advertises, and Domains for where it’s routed, with Preview, Versions, Activate, and Save in the header. Identity sits behind the settings cog. One Preview / Markdown control moves both boxes — a procedure and its output shape are read together and written together — and Clear empties them, behind a confirmation. A new skill starts as a worked example you type over rather than an empty box. Preview shows the served payload three ways — Formatted (rendered), Markdown (verbatim, what the agent receives), and JSON.
The page lists the skills your org writes. The administration procedures the package ships are not shown or edited here — see below.
Procedure and output shape
The procedure is markdown, delivered to agents exactly as written: the steps in order, the stop conditions (“if the scope exceeds one pass, say so and propose sub-scopes”), and what to treat as data rather than instruction. The output shape is the second slot, kept apart on purpose: free-form analysis is unreviewable, so every skill states the shape its result takes — a review table, a findings list, a report. That is what makes two runs of the same skill comparable.
Runtime and admin surfaces
Every skill you write is a runtime skill, delivered through domains exactly as briefs are. A skill in a domain is listed in that domain’s catalog by name and one line, and Get Skill returns the full text when the task matches. A skill in no domain reaches nobody.
The seven skills the package ships are admin-surface — the admin agent’s own procedures, on the administration topic and MCP server, which the permission model gates. They are not delivered through domains, and Studio neither lists nor edits them: they are the product’s internals, not your org’s configuration. See the admin agent’s skills.
Publishing: activate to serve
Skills version exactly as entities and briefs do: edit the draft, Activate with a comment to publish a numbered version, roll back by activating an older one. Agents always follow the active version, so editing a packaged procedure changes nothing an agent runs until a person activates it. Deactivate takes it out of service immediately. → Versions
Seeded skills arrive with v1 active, so the shipped procedures work out of the box. Copy duplicates a procedure as an inactive, unrouted draft with its lineage stamped — the starting point for a regional or team variant.
Who receives a skill
Add runtime skills on a domain’s Skills tab — search, add, drag to order. The order is the order that domain’s catalog lists them. Domains are the product’s one access model, applied to the third library: they decide who receives what.