The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill captures raw text and assets from user-provided URLs using npx hyperframes capture (referenced in references/step-0-capture.md). This untrusted data is then ingested into the agent's context and used to generate the video's script and visual compositions.
Ingestion points: capture/extracted/visible-text.txt and capture/extracted/asset-descriptions.md (Step 2 and 3).
Boundary markers: The instructions mention a "brand-truth" framing but do not specify robust sanitization or delimiters to prevent embedded instructions in the captured site from influencing the agent's creative or technical decisions.
Capability inventory: The skill can execute shell commands via npx, perform network requests via curl, and write/read local files throughout the 7-step process.
Sanitization: No explicit sanitization or filtering of external content is documented before it is interpolated into prompts or script files.
[DYNAMIC_EXECUTION]: The skill generates .html files in the compositions/ directory that include JavaScript (GSAP animations). These files are executed at runtime by the hyperframes engine during the snapshot, validate, and render phases (Step 5 and 6).
Implementation: The agent (or sub-agents) assembles executable code based on templates and user/site data, which is then loaded into a headless Chrome environment for rendering.
[EXTERNAL_DOWNLOADS]: The skill downloads brand assets, fonts, and scripts from external URLs.
References: Interacts with api.elevenlabs.io and api.heygen.com for media generation and google.dev for API keys.
Registry: Downloads registry blocks from heygen-com/hyperframes using the npx hyperframes add command.

website-to-video