openai-whisper-api

Pass

Audited by Gen Agent Trust Hub on Mar 12, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The script interacts with api.openai.com, which is a well-known and trusted service for AI tasks.\n- [DATA_EXFILTRATION]: Audio files are uploaded to the OpenAI API for processing. This aligns with the skill's documented purpose of transcription.\n- [COMMAND_EXECUTION]: The script invokes curl and mkdir using shell variables derived from user input.\n- [CREDENTIALS_UNSAFE]: The skill relies on the OPENAI_API_KEY environment variable, which is a standard but sensitive method for managing API access.\n- [PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface through the lack of input sanitization in transcribe.sh.\n
  • Ingestion points: Audio file paths and the --prompt hint provided as CLI arguments.\n
  • Boundary markers: No delimiters or ignore-instructions warnings are applied to user-supplied parameters.\n
  • Capability inventory: Network access via curl and file creation/writing capabilities.\n
  • Sanitization: Arguments are passed directly to curl -F. Because curl treats values starting with @ as file paths, an attacker who can control these arguments could potentially read and exfiltrate arbitrary local files.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 12, 2026, 02:37 AM