Coverage added after cases 1 and 2: the exact detail of what the assistant really fills in on these two settings screens, and what it only points to. ## 7.1 AI Agents — create or edit an agent | It can fill in and save | It only points to (you do it) | | --- | --- | | Agent name, Tone (closed list), instructions (system prompt), fallback message when it doesn't know what to answer, and turning the agent on/off | AI model (the catalog depends on what your plan has enabled — manuals 31/40), escalation rules, the agent's visible resources, and which other agent conversations are handed off to — these are lists that repeat row by row or depend on your own data | Saving the hand-off setting has its own button, separate from the rest of the form — the assistant treats it as a separate action, with its own confirmation. > **Note — "Delete an agent" doesn't exist** > > If you ask for it, the assistant will tell you honestly: the platform doesn't have that feature today — it isn't even in your own menu, so it can't be in what the assistant does either. ## 7.2 Channels — set up a new channel It can fill in the channel type, the name, and press Save (manual 24). > **Notice — An already-connected channel: guide-only, always** > > Once the channel exists, the assistant **never** fills in or taps anything inside its record — not the assigned agent, not the people in charge, not the widget language, not connecting/disconnecting, not even that channel's own "Copilot" section. Honest reason: you can have several channels of the same type at once (several WhatsApp numbers, for example — manual 24), and today there's no safe way for the assistant to know which one you mean without risking touching the wrong one. It's a deliberate limit, not an oversight — same as connecting WhatsApp in case 2 above.