vapi-voice-call

Pass

Audited by Gen Agent Trust Hub on Mar 12, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches the source code for the Vapi CLI tool from the author's GitHub repository at github.com/rustykuntz/rocie-tools.
  • [COMMAND_EXECUTION]: Runs shell commands to clone the repository, install dependencies using npm, and compile the tool. It also executes the generated JavaScript file using the Node.js runtime.
  • [PROMPT_INJECTION]: The skill processes untrusted input that is later used to guide the behavior of an LLM-powered voice assistant, creating a surface for indirect prompt injection.
  • Ingestion points: Accepts user-provided context via CLI arguments, standard input, and local files (e.g., ~/.rocie/context/user.yaml).
  • Boundary markers: The skill documentation does not specify the use of delimiters or instructions to ignore embedded commands within the input.
  • Capability inventory: The tool has the capability to initiate and terminate phone calls via network requests to the Vapi API.
  • Sanitization: No sanitization or validation of the input text is mentioned before it is passed to the Vapi service.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 12, 2026, 04:41 AM
Security Audit — agent-trust-hub — vapi-voice-call