Concepts
Three Agent Classes
WebMCP, DOM-driver, and screenshot agents have different detection guarantees.
Three Agent Classes
Agentronics models agents by how they interact with the page:
| Class | Interaction path | Detection confidence |
|---|---|---|
webmcp | Structured tools through navigator.modelContext | Stable, exact signal when present |
dom | Browser automation over DOM/click/type APIs | Beta, heuristic confidence score |
screenshot | Pixel-level mouse and keyboard control | Not reliably detectable client-side |
const identity = await client.detect()
if (identity?.class === 'webmcp') {
// Structured tool client.
}
Screenshot agents should declare themselves with presentIdentity() or authenticate() if they need more trust than anonymous browsing.