The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted audio files to generate text transcriptions which are then returned to the agent's context. This creates a surface for indirect prompt injection if the audio contains malicious instructions designed to influence the agent's subsequent behavior.
Ingestion points: The audio positional argument and the --known-speaker file paths in scripts/transcribe_diarize.py act as entry points for external data.
Boundary markers: The transcription output is returned as plain text or JSON without specific delimiters or instructions to the agent to treat the content as untrusted data.
Capability inventory: The script has the ability to read local files, write to the file system via out_path.write_text, and perform network requests to the OpenAI API.
Sanitization: There is no explicit sanitization of the transcribed text before it is returned to the agent.
[SAFE]: The skill follows established security best practices for handling sensitive information.
Credential Safety: The documentation and script correctly manage the OPENAI_API_KEY by using environment variables and explicitly warning the user never to paste the key into the chat interface.
Secure Dependency Management: The skill recommends using uv or pip to install the official openai Python package, avoiding non-standard registries or untrusted sources.
Transparent Operations: The bundled Python script is readable and its operations (reading audio, encoding for the API, and writing results) are clearly defined and aligned with the skill's stated purpose.

transcribe