0.16.0
BetaSkills: write a procedure once, and every AI client follows it.
Your business already has procedures. How you qualify a lead, how you word a quote, which checks a reconciliation needs. Until now the only way to get one into an AI client was to paste it into a chat and hope it stuck. Skills stores them in your workspace instead, and serves them to every AI client you have connected.
A skill is one short document
A skill is a page of markdown: the procedure, written the way you would explain it to a new hire. Keep each one to a single job. Short skills get followed, long ones get skimmed.
Two fields matter more than they look. The name is what you see. The description is what the model reads to decide whether to open the skill at all, because it becomes the tool description in the tool list. A vague description means a perfect skill nobody ever opens.
Collections are what you attach
Skills live in collections, and the collection is what you attach to a profile, not each skill. That keeps attaching down to a handful of decisions instead of one per document.
Each published skill becomes one tool, named after its collection. A skill called Qualify inside a collection called Marketing is skill_marketing_qualify. The dashboard shows you that name before you create the collection, because the slug is permanent.
Nothing goes live until someone publishes it
A new skill saves as a draft, and a draft reaches nobody. Publishing is a separate, explicit click. That gate is what makes it safe to let more people write than publish: whoever publishes is the person deciding the wording is ready to be followed.
Editing a skill that is already published does go live. Every connected client reads the new text on its next call, and you can attach a note to each edit saying what changed.
Who can write, who can publish
Writing, publishing and managing collections are separate permissions. A marketer can draft the marketing procedures without being able to change what your accountant published. Profiles carry their own per-collection settings on top of that, so one endpoint can read a collection while another is also allowed to write to it.
Skills is in alpha
Everything above works, and your data is handled exactly as it is everywhere else. But this is the first release of a new feature, and the way skills are structured may still change. If that happens we will say so in a release note rather than change it under you. The dashboard marks the section so nobody has to remember, and there is a feedback link in the banner.
Tell us what feels awkward now, while it is still cheap to change.
