voice-ai-integration

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements a dynamic prompt generation pattern that is susceptible to indirect prompt injection because it incorporates user-controlled inputs into the AI's system instructions.
  • Ingestion points: The startSession tRPC endpoint in SKILL.md accepts inputs such as archetype, difficulty, and practiceMode from the client.
  • Boundary markers: The generateSystemPrompt function in SKILL.md interpolates these variables directly into template strings without using delimiters or instructions to ignore embedded commands.
  • Capability inventory: The AI interface (Hume EVI) has the capability to conduct real-time conversations and the results are stored in a database (practice_sessions table).
  • Sanitization: The provided code snippets do not show any validation or escaping of user inputs before they are included in the generated system prompt.
  • [EXTERNAL_DOWNLOADS]: The skill downloads and installs the official hume and twilio packages from the public NPM registry to enable voice and AI functionality.
  • [COMMAND_EXECUTION]: The documentation includes standard shell commands for installing dependencies using the pnpm package manager.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 07:23 AM
Security Audit — agent-trust-hub — voice-ai-integration