openai-whisper-api
Pass
Audited by Gen Agent Trust Hub on Mar 6, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill performs its documented function using legitimate API interactions and does not contain any malicious patterns or safety guideline bypasses.
- [COMMAND_EXECUTION]: The bash script uses the
curlbinary to perform HTTP requests to the transcription endpoint. - [EXTERNAL_DOWNLOADS]: Communicates with the official OpenAI API at
api.openai.comto send audio data and receive transcriptions. - [DATA_EXFILTRATION]: Transmits user-specified audio files to OpenAI's servers. This is considered safe as it is the primary intended function of the skill and targets a well-known, trusted service provider.
- [CREDENTIALS_UNSAFE]: Utilizes an
OPENAI_API_KEYfor authentication. The skill correctly instructs the user to provide this via environment variables or a configuration file rather than hardcoding credentials. - [PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by processing external audio data and accepting user-provided prompts. Evidence chain: 1. Ingestion points: audio file and
--promptargument inscripts/transcribe.sh; 2. Boundary markers: None; 3. Capability inventory: curl network request and local file writing; 4. Sanitization: None. This is documented as a baseline surface risk for transcription-related skills.
Audit Metadata