The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external URLs provided by the user, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Website content is captured into visible-text.txt, asset-descriptions.md, and tokens.json in the capture/ directory (Step 1).
Capability inventory: The agent has the capability to execute shell commands via npx hyperframes, write files for video compositions, and execute generated Javascript in a headless browser via the validate command (Step 7).
Boundary markers: No specific delimiters or warnings to ignore embedded instructions are implemented for the captured website data.
Sanitization: The process relies on the agent to summarize the data, which serves as a manual but incomplete sanitization step.
[DYNAMIC_EXECUTION]: The workflow generates HTML and Javascript code for video compositions which are subsequently executed for validation.
Evidence: Step 6 involves building compositions in the compositions/ directory. Step 7 uses npx hyperframes validate to execute these compositions in a headless Chrome environment to detect runtime JS errors and network failures.
[COMMAND_EXECUTION]: The skill executes multiple shell commands to manage the video production pipeline.
Evidence: The workflow includes execution of npx hyperframes capture, npx hyperframes tts, npx hyperframes transcribe, npx hyperframes lint, and node skills/hyperframes-animation-map/scripts/animation-map.mjs.
[EXTERNAL_DOWNLOADS]: The skill fetches content from the internet based on user input.
Evidence: The npx hyperframes capture <URL> command is used to download screenshots, assets, and metadata from arbitrary websites provided by the user.

website-to-hyperframes