The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the transcription of audio files where the resulting text is returned to the agent context. This creates a risk for indirect prompt injection if the audio content contains malicious instructions designed to manipulate the AI agent. \n- Ingestion points: The skill uses fs.createReadStream in examples/core.md to ingest audio data for transcription calls. \n- Capability inventory: The skill has capabilities for reading local files, writing local files (e.g., SRT/VTT subtitles via fs.writeFileSync), and making network requests to the OpenAI API. \n- Boundary markers: No delimiters or protective instructions are provided to wrap the resulting transcript to prevent the agent from accidentally executing commands found within the audio content. \n- Sanitization: There is no evidence of output sanitization for the transcribed text before it is used by the agent.

ai-provider-openai-whisper