The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes audio content from external URLs for transcription. This establishes an attack surface where a maliciously crafted audio file, once converted to text, could attempt to influence the agent's behavior through instructions embedded in the speech.
Ingestion points: Audio source URLs provided in the audio input parameter within SKILL.md examples.
Boundary markers: The instructions do not specify explicit delimiters or guardrails to isolate transcribed text from agent instructions.
Capability inventory: The skill is restricted to the belt CLI via Bash, limiting the potential impact of an injection.
Sanitization: Content sanitization is dependent on the downstream processing of the transcription output by the agent platform.

elevenlabs-stt