The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external audio files from URLs and returns the transcription to the agent. If an audio file contains spoken instructions (e.g., instructions to ignore safety guardrails or perform unauthorized actions), the agent could potentially follow these instructions when it processes the resulting text.
Ingestion points: Audio URLs provided to the belt command in SKILL.md examples.
Boundary markers: Not present. There are no instructions to the agent to treat the transcribed output strictly as data or to ignore embedded commands.
Capability inventory: The skill is authorized to use the belt CLI via the Bash tool to perform network-based transcription.
Sanitization: None. The transcription results are passed back to the agent context without filtering or sanitization.
[EXTERNAL_DOWNLOADS]: The skill documentation refers to external resources and secondary skills required for operation.
Mentions installation instructions located at raw.githubusercontent.com/inference-sh/skills.
Recommends the installation of the belt-sh/cli skill to provide the necessary shell environment for executing transcription tasks.

elevenlabs-stt