The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of capturing and processing external website data.
Ingestion points: Website content is ingested via npx hyperframes capture <URL> (documented in SKILL.md and references/step-0-capture.md). Additionally, visual snapshots are analyzed by an AI model (Gemini) which can also be influenced by the content of the captured frames.
Boundary markers: There are no explicit instructions or delimiters used when the agent processes the captured site data to ignore potentially malicious embedded instructions.
Capability inventory: The agent has the ability to write files (scripts, HTML, markdown), execute shell commands (npx hyperframes), and make network requests (curl).
Sanitization: The instructions do not specify any sanitization, filtering, or validation of the captured web content before it is used to derive brand voice, messaging, and code.
[EXTERNAL_DOWNLOADS]: The skill performs downloads from external sources and interacts with remote APIs.
It utilizes curl to send data to and download audio from HeyGen and ElevenLabs APIs (references/step-4-vo.md). These are well-known technology services.
The generated video compositions include remote scripts from cdn.jsdelivr.net for the GSAP library (references/step-5-build.md).
[COMMAND_EXECUTION]: The skill frequently executes shell commands via the hyperframes CLI using npx for tasks like site capture, rendering, linting, and snapshotting. While standard for this toolset, it involves executing code from the public npm registry.

website-to-hyperframes