The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the belt CLI and other related skills from external sources including npx skills add belt-sh/cli and various inference-sh/skills. It also references documentation and installation scripts hosted on inference.sh and github.com/inference-sh/skills.
[INDIRECT_PROMPT_INJECTION]: The skill ingests audio and video data to generate text transcripts which may then be used as input for subsequent commands (e.g., video captioning). This creates a vulnerability surface where malicious instructions embedded in audio could be executed if the agent processes the resulting transcript without proper delimitation.
Ingestion points: Transcription output (e.g., transcript.json) derived from external audio/video URLs in SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the example workflows.
Capability inventory: The skill uses the belt CLI tool via Bash to perform network requests and process data.
Sanitization: No sanitization or validation of the transcribed text is demonstrated before it is passed to downstream tools.

speech-to-text