voice_command_to_skill

Pass

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill facilitates an attack surface for indirect prompt injection by taking raw ASR transcripts and inserting them into prompt templates for subsequent actions.
  • Ingestion points: Raw speech-to-text transcripts are processed in the voice_command_to_skill workflow to determine intent and fill parameters.
  • Boundary markers: Prompt templates such as 'The user said: {transcript}' lack explicit delimiters or instructions to disregard potential malicious commands within the transcript.
  • Capability inventory: The router triggers downstream skills with significant capabilities, such as file-system writes (export_experiment_data_to_excel), video analysis, and data extraction.
  • Sanitization: No input validation or filtering is documented to protect against malformed or malicious transcripts before they are utilized in instructions for subsequent skill calls.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 15, 2026, 04:33 AM