The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external audio files whose content is not controlled by the skill, creating a surface for indirect prompt injection if the resulting transcription contains instructions that an agent might follow.
Ingestion points: The scripts/transcribe.sh script accepts a user-specified path to an audio file for transcription.
Boundary markers: No delimiters or protective framing are used when returning the transcription result to indicate it contains untrusted content.
Capability inventory: The skill has the ability to reach the network via curl and write files to the local filesystem.
Sanitization: There is no sanitization or filtering of the transcribed text before it is returned or stored.
[SAFE]: The skill uses curl to interact with the well-known OpenAI API (api.openai.com), which is appropriate for the skill's stated purpose and aligns with standard security practices for authorized network requests.
[SAFE]: API credentials are handled securely by utilizing environment variables and configuration files rather than hardcoding secrets within the script or metadata.

openai-whisper-api