The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to manage the video generation and verification workflow.
Evidence: SKILL.md directs the agent to generate random seeds using /dev/urandom and to execute video processing tasks using npx commands for the @helios-project/cli tool.
[EXTERNAL_DOWNLOADS]: The skill fetches external tools and resources at runtime to perform its primary functions.
Evidence: SKILL.md uses npx to download and run the latest version of @helios-project/cli. Reference HTML files in the references/genres/ directory (e.g., era-timeline.html, kinetic-type.html) fetch fonts from Google Fonts (fonts.googleapis.com).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data which is interpolated into video briefs and subsequently handled by the agent, creating a surface for potential instruction injection.
Ingestion points: references/brief.md defines an INPUTS section where users provide names, facts, and exact on-screen copy.
Boundary markers: The brief utilizes a structured template and "LOCKED COPY" directives to define context, though it relies on the agent's adherence to instructions rather than technical sanitization.
Capability inventory: The skill has the capability to write files (video HTML) and execute shell commands (npx render/verify) across all identified scripts.
Sanitization: The brief includes a "BANNED" list to restrict specific visual styles and requires all copy to be sourced from a facts file, reducing the risk of arbitrary content generation.

guided-video-recipes