The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted audio files which can result in the generation of text containing instructions that might influence a downstream agent or LLM.
Ingestion points: Audio files are processed via the model.transcribe() method in SKILL.md and references/languages.md.
Capability inventory: Transcribed text is written to files and used in Retrieval-Augmented Generation (RAG) pipelines via LangChain.
Boundary markers: The provided examples do not demonstrate the use of delimiters to isolate the transcribed content from agent instructions.
Sanitization: No sanitization of the transcription output is included in the provided code snippets.
[SAFE]: The skill is a standard integration guide for the official OpenAI Whisper toolset.
Dependencies such as openai-whisper, transformers, and torch are standard for speech-to-text applications.
External links lead to the official OpenAI repository and academic research papers.
Installation commands for system tools like ffmpeg are standard across different operating systems.

whisper