The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it crawls untrusted external websites and directly uses the extracted content to guide the automated generation of React components.
Ingestion points: Untrusted text, layout data, and attributes are ingested from target URLs via Chrome MCP.
Boundary markers: There are no explicit markers or instructions telling sub-agents to disregard instructions embedded in the target website's text.
Capability inventory: The skill possesses the capability to write files to the local source directory and execute shell commands via npm and npx.
Sanitization: No sanitization of the scraped website content is performed before it is passed to the builder agents.
[COMMAND_EXECUTION]: The skill executes several build and verification commands locally.
Evidence: The instructions require running npm run build and npx tsc --noEmit at multiple stages of the cloning process.
[EXTERNAL_DOWNLOADS]: The skill performs automated downloads of remote assets.
Evidence: It uses a script through Chrome MCP to enumerate and batch-download images, videos, and SVGs from the target URL.

clone-website