AUZM Capabilities
| Capability | Good fit | Boundary |
|---|---|---|
| App state probes | Activity, fragment, object tree, current route, visible business state. | Probe output is diagnostic until tied to a real action or app response. |
| Internal app actions | Calling owner-approved app APIs, adapters, callbacks, or service methods from the current live context. | Require target-specific evidence before claiming success. |
| Compatibility adaptation | Small patches for app version changes, feature toggles, bridge routing, and API differences. | Prefer narrow hooks over broad global interception. |
| Automation handoff | Generic UI tools prepare state, then AUZM runs the app-internal operation. | AUZM modules should not become UI automation scripts. |
| Observability | Structured logs, request/response traces, latency, package and process proof. | Never log secrets, tokens, credentials, or private payloads unnecessarily. |
| Screenshot-assisted diagnosis | Use screenshot URLs when XML is insufficient. | The screenshot URL is protected by the MCP secret and should not be made public. |
Practical Tool Shape
Prefer one final tool per target behavior:
json
{
"ready": true,
"target_package": "com.example.app",
"action": "send_message",
"proof": {
"callback": "MessageRepository.send returned message_id",
"message_id": "..."
}
}If not ready:
json
{
"ready": false,
"reason": "target fragment not visible",
"observed_activity": "...",
"next_probe": "open the compose screen first"
}