The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references installation instructions hosted on the author's official GitHub repository (https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md).
[COMMAND_EXECUTION]: The skill instructions utilize the belt CLI tool (via Bash tool permissions) to execute remote AI model tasks, which is the intended purpose of the skill.
[DATA_EXFILTRATION]: The skill requires passing external URLs for images and audio to remote processing endpoints through the belt tool. This is a standard functional requirement for cloud-based media generation.
[PROMPT_INJECTION]: As an AI production skill, it possesses an indirect prompt injection surface because it processes untrusted user-provided content (like voice scripts or external media descriptions) and feeds them into AI model prompts.
Ingestion points: voice_script, voice_prompt, and video_prompt fields within the belt app run commands (SKILL.md).
Boundary markers: The skill uses structured JSON objects to separate instructions from data, which provides a basic level of segmentation.
Capability inventory: The skill uses Bash(belt *) to interact with remote services.
Sanitization: There is no explicit sanitization or filtering of the script content demonstrated in the skill instructions.

talking-head-production