Skills it learns once — and reuses, still gated

Each time the agent solves a new kind of problem, the solution is captured as a skill in a library scoped to your tenant. Skills are learned from work you supervised and approved — not installed from a public marketplace — so the library’s provenance is your own operating history, end to end.

Routine work stops being improvised: the weekly report, the pipeline cleanup, the monthly reconciliation become known quantities — predictable in steps, cost, and output.

• Learned from your supervised work, not marketplace downloads
• Scoped to your tenant — your plays stay yours
• Routine work becomes dependable instead of improvised

Skills compose: the agent chains them into larger jobs the way an experienced operator combines known plays, and routines can put them on a schedule so recurring work runs without being asked. Reuse is where delegation pays — deliberation the first time, dependability every time after.

And because skills run against your live memory and current policies, a reused skill reflects what you know and allow today, not a snapshot of the day it was learned.

Whatever a skill has done before, its outward actions are gated now: every mutation-capable call a skill makes passes the same proposal gate and the same risk threshold as ad-hoc work. Trusting a procedure and authorising an action stay two separate decisions — the first never quietly implies the second.

That separation is what makes a growing skill library safe. Capability compounds; authority only changes when you change it.