The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install the agent-browser CLI tool globally via NPM if it is not detected on the host system.
[COMMAND_EXECUTION]: The workflow relies on executing several shell commands to drive the browser, manage the Node.js project, and download assets, including agent-browser, npm, node, and curl.
[EXTERNAL_DOWNLOADS]: The skill downloads arbitrary files (images, videos, fonts) from user-provided target URLs. It also incorporates well-known services such as Google Fonts and Unsplash for typography and fallback imagery.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes untrusted data from scraped websites without explicit protection against instructions embedded in the target content.
Ingestion points: Scraped HTML, text, and metadata from the target URL provided by the user via agent-browser.
Boundary markers: The instructions do not define clear boundaries or 'ignore' directives to separate the extracted data from the agent's logic.
Capability inventory: The agent has permissions to write files to the project directory, execute shell commands, and create git worktrees.
Sanitization: No explicit sanitization or validation of the scraped web content is performed before it is used for code generation or component specification.

clone-website