Capabilities & Skills
Two related but distinct things live on an agent's page under this heading: its mesh capabilities — what it can do in conversation with other agents — and its skills and files — what it has available to draw on while it works.
Mesh capabilities
Hovering an agent's tile in the roster, or right-clicking it and choosing Mesh Capabilities…, opens a card with four parts:
| Part | Shows |
|---|---|
| Header | Avatar, name, handle, and whether it's currently a live mesh peer. |
| Skills | Every skill available to this agent, each with its own on/off toggle for whether other agents can see it. |
| Protocol Capabilities | The Streaming and Push notifications toggles (below). |
| Extensions | A read-only list of the protocol extensions this agent supports. |
Per agent, you can turn on:
| Capability | What it does |
|---|---|
| Streaming | Incremental output while the agent works, rather than only a final result. |
| Push notifications | Unsolicited updates pushed to a subscriber. |
A baseline reply capability is always on and can't be turned off — every agent can respond, even with nothing else turned on. Each skill's own toggle in this card controls whether that skill is exposed to other agents, independent of Streaming and Push notifications.
The orchestration allow-list is separate
Turning on mesh capabilities for an agent is not the same as letting it participate in Cross-Agent Orchestration — chains of agents triggering agents. That allow-list is a single global setting in App Settings, not a per-agent toggle here. It's off by default, gated to plans that include it, and inert until you fill it in — an agent having every mesh capability turned on still can't start a governed chain unless it's also named in that allow-list.
Skills and files
| List | What it holds | How items get there |
|---|---|---|
| Skills | Skills discovered automatically from your project and from the agent's own home folder. | Discovered, not attached — each has its own on/off toggle. |
| Files | Files you attach explicitly. | You choose them via Attach File…, which accepts more than one at once; NEXUS copies each into the agent's home. Dragging a bit page onto an agent attaches it the same way — see Workspaces. |
The distinction that matters: skills are discovered, and you only decide whether a discovered one is switched on or off for this agent; files are attached, and you choose exactly which ones and when. A project-discovered skill and one discovered from the agent's own home folder look the same in the list — nothing distinguishes where a skill came from, except a "not found" label that appears if a previously linked skill can no longer be located.
Once a file is attached, Reveal and Remove act on it individually — removing it takes effect immediately, with no confirmation step, and doesn't affect the original file wherever you attached it from.
Where to go next
- Creating Agents — the roster gallery this section's capability card is reached from.
- Settings reference — the Cross-Agent Orchestration allow-list and its limits.
- Agent Mesh — what agents actually do with mesh capabilities once they're turned on.