Auto-activation
How Claude decides which installed persona to use for a given request.
Personas are built from Claude skills: each capability ships with a description of what it does and when it applies. Claude reads your request, compares it against every installed skill's description, and activates the one that matches, without you naming it. This is one of two ways to run a capability, the other is typing its namespaced slash command directly (see Using personas); auto-activation is what happens the rest of the time, whenever you just describe what you want.
When you have one persona installed
Matching is straightforward: if your request looks like the kind of work that persona covers, it activates. If it does not, Claude answers normally without pulling in the persona at all.
When you have more than one installed
Claude uses the specifics of what you are asking for to pick the right one. A request to "draft a PRD" routes to the Product Manager persona; a request to "map this process" routes to the Business Analyst persona, even with both installed. If a request is ambiguous, or could reasonably belong to more than one persona, be explicit: name the persona, or describe the deliverable precisely enough that only one of them fits (see Tips for good results).
No manual switching required
You never need to "activate" or "deactivate" a persona yourself. Activation happens per request, based on what you ask for, not as a mode you turn on for a whole session.