The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection because its core function is to ingest and process data from untrusted external websites provided via the $ARGUMENTS URL.
Ingestion points: The agent uses browser automation tools (Chrome MCP, Playwright, etc.) to navigate to and parse the DOM, text, and metadata of arbitrary URLs.
Boundary markers: The instructions lack specific boundary markers or 'ignore embedded instructions' directives when extracting text content from the target site.
Capability inventory: The agent has the ability to write files to the local filesystem, create directories, execute shell commands (npm run build, npx tsc), and run generated Node.js scripts.
Sanitization: There is no mention of sanitizing or escaping the text content extracted from the target website before it is incorporated into component code or specification files.
[DYNAMIC_EXECUTION]: The skill directs the agent to generate and execute code at runtime.
Script Generation: It specifically instructs the agent to write a Node.js script (scripts/download-assets.mjs) to download all discovered images, videos, and fonts.
Runtime Execution: The agent is then instructed to execute this generated script locally to populate the public/ directory.
Browser Scripting: The skill uses eval-style patterns to execute complex JavaScript snippets within the browser environment to extract computed styles and asset lists from the remote page.
[COMMAND_EXECUTION]: The skill requires the execution of several shell commands to manage the project lifecycle.
Build Commands: It frequently runs npm run build and npx tsc --noEmit to verify the integrity of the cloned components.
Filesystem Operations: It performs multiple mkdir and file-write operations to create the research documentation and component source files.
[EXTERNAL_DOWNLOADS]: The skill is built to download a wide variety of assets from the internet.
Asset Harvesting: It identifies and downloads images, videos, SVGs, and font files from the user-specified target URL to the local environment.

clone-website