implementing-speech-to-text

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.65). The runtime workflow is speech recognition transcript handling: args.transcript from the Web Speech API (user-spoken, thus not authored by the operating user) is inserted into JavaScript variables/DOM and is then commonly POSTed/forwarded (e.g., ProcessVoiceInput(string transcript)), creating an indirect route for outsider free text into any downstream LLM-context-building step.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill includes runtime loading of third‑party JavaScript from the Syncfusion CDN (e.g. https://cdn.syncfusion.com/ej2/dist/ej2.min.js), which is fetched and executed at runtime and is required for the component to function.