openrouter-transcribe
Pass
Audited by Gen Agent Trust Hub on Jun 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [SAFE]: Credentials are managed via the 'OPENROUTER_API_KEY' environment variable, avoiding hardcoded secrets within the script or metadata.
- [SAFE]: The transcription script uses 'jq' with the '--arg' and '--rawfile' flags to safely construct JSON payloads, ensuring that user-supplied prompts and audio data are correctly escaped and cannot lead to injection attacks against the API request structure.
- [COMMAND_EXECUTION]: The skill executes standard system utilities including 'ffmpeg' for audio processing and 'curl' for network requests to 'openrouter.ai'.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) since it processes untrusted audio data through an LLM. 1. Ingestion points: Untrusted audio file content and user-provided transcription prompts in 'scripts/transcribe.sh'. 2. Boundary markers: No specific delimiters are used to wrap the transcript output. 3. Capability inventory: 'ffmpeg', 'curl', 'base64', 'jq', and file write access via the '--out' parameter. 4. Sanitization: While the API request is safely formatted with 'jq', the resulting transcript content is written directly to the output file or stdout without additional sanitization.
Audit Metadata