agentphone

Pass

Audited by Gen Agent Trust Hub on Jun 26, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADS
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill configures an MCP server to be downloaded and executed at runtime via npx -y agentphone-mcp@0.2.0. This package is the official telephony adapter maintained by the vendor.
  • [PROMPT_INJECTION]: The skill provides tools for reading external data, which introduces an indirect prompt injection surface.
  • Ingestion points: Incoming SMS messages and call transcripts are accessible via get_messages, get_call, and conversation tools.
  • Boundary markers: No specific delimiters or "ignore" instructions are defined for processing external message content.
  • Capability inventory: The agent can respond to inputs by sending messages, making calls, and modifying account settings.
  • Sanitization: The skill does not implement content filtering or sanitization for incoming communications.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 26, 2026, 11:40 PM
Security Audit — agent-trust-hub — agentphone