neox-phone-mcp
Warn
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: MEDIUMPERSISTENCEDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PERSISTENCE]: The skill instructions recommend adding the generated bridge script (
neox-bridge.py) to shell profiles or startup hooks to ensure the service remains active across user sessions and system restarts.- [DYNAMIC_EXECUTION]: The skill dynamically creates a Python script on the local filesystem usingcatand executes it in the background usingnohup. This script manages the network bridge and handles asynchronous communication.- [COMMAND_EXECUTION]: The skill uses several shell commands includingdns-sdfor Bonjour discovery,arpfor network scanning, andcurlfor communicating with the mobile MCP server. The bridge script also spawns adns-sdprocess viasubprocess.Popen.- [INDIRECT_PROMPT_INJECTION]: The bridge script opens aPOST /agentendpoint on the local network that accepts unauthenticated text to be used as an agent prompt, creating a potential attack vector for any device on the LAN. - Ingestion points: The script listens on port 8787 across all local interfaces (
0.0.0.0) for incoming POST requests containing task handoffs. - Boundary markers: No boundary markers or "ignore" instructions are present to prevent the agent from obeying malicious instructions embedded in the network-provided text.
- Capability inventory: The skill allows the agent to execute shell commands, read and write local files in the
~/.neoydirectory, and interact with the phone's media library tools. - Sanitization: The script performs no validation or sanitization on the body of incoming POST requests before storing them for the agent to process as user input.
Audit Metadata