Microsoft ships Agent Skills for .NET with approval gates

Microsoft shipped Agent Skills for .NET as a stable, production feature in Agent Framework, dropping the Experimental tag it carried through preview.

None of that runs on its own. A skill can't load, read its own resource files, or execute a script without a human or a policy approving it first, by default.

Three calls make up the API, per Microsoft's post: load_skill, read_skill_resource, and run_skill_script. Skills themselves are reusable packages of instructions, docs, and scripts. Agents pull them in only when a task needs them, instead of loading everything into context up front.

That framing connects to the GitLost story, on skills as independently shippable, combinable capability bundles. This time, per the post, they ship with guardrails built in from the start.

I like that governance shipped with the GA release itself, not as a patch added later. That's usually how I see it handled elsewhere, only after something's already broken.

Here's the bet I'd take the other side of: approval-by-default gets relaxed within a year. Teams get tired of clicking through prompts for every skill call. They'll start pre-approving whole skill sets, which quietly puts most of that risk back. Watch whether Microsoft ships a bulk-approval or trusted-source flag; that's the tell.

This lands as Agent Framework's skills feature moves from experimental to the version .NET teams are expected to run in production.