The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external URLs via the npx hyperframes capture tool. This data is used to generate narratives and visual designs, creating a potential surface for indirect prompt injection if the target website contains malicious instructions.
Ingestion points: SKILL.md Step 1 invokes npx hyperframes capture "<URL>" to crawl external websites.
Boundary markers: The skill instructions do not explicitly define boundary markers or safety instructions when interpolating captured text into prompts for storyboarding and script writing in Step 3.
Capability inventory: The skill possesses extensive capabilities including local file system access, network operations via platform tools, and shell command execution for media processing (ffmpeg/ffprobe).
Sanitization: While scripts/assemble-index.mjs provides attribute whitelisting for hoisted video elements, the high-level script generation relies primarily on the underlying model's inherent safety filters to handle potentially malicious content in crawled text.
[DYNAMIC_EXECUTION]: Several scripts in the skill generate or execute code dynamically at runtime to assemble the final video project.
Evidence: scripts/audio.mjs executes a Node.js engine script using a path that can be dynamically set via the HF_MEDIA_ENGINE environment variable.
Evidence: scripts/transitions.mjs and scripts/captions.mjs generate and inject GSAP animation code into HTML compositions using string templates and project-specific metadata.
[EXTERNAL_DOWNLOADS]: The skill facilitates remote updates and project initialization from external sources.
Evidence: SKILL.md instructs the user to run npx hyperframes skills update product-launch-video, which fetches updates from GitHub. As GitHub is a well-known and trusted service for these platform tools, this is documented as neutral behavior.

product-launch-video