speech
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text input to generate speech audio.
- Ingestion points: Text input provided via the
--inputargument,--input-filepath, or batch JSONL files processed byscripts/text_to_speech.py. - Boundary markers: The underlying OpenAI Audio API separates the content to be spoken (
input) from the delivery instructions (instructions), though the model may still be influenced by content within the text. - Capability inventory: The skill performs outbound network requests to the OpenAI API and writes generated audio files to the local file system.
- Sanitization: The
scripts/text_to_speech.pyscript enforces a maximum input length of 4096 characters per request, as required by the API.
Audit Metadata