The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates and executes JavaScript files (e.g., scripts/download-assets-<site-key>-<page-key>.mjs) during its operation. These scripts are dynamically constructed using data parsed from untrusted external websites, which presents a risk of executing code influenced by malicious third-party content.
[INDIRECT_PROMPT_INJECTION]: The reconnaissance and extraction phases involve verbatim capture of text and metadata from arbitrary URLs. This data is then inlined into specifications for 'builder' agents, creating an exploitable surface for indirect prompt injection if the source website contains instructions designed to override agent behavior.
Ingestion points: Browser automation processes user-provided URLs to extract DOM content and metadata (Phase 1 and 3).
Boundary markers: The skill lacks explicit instructions for using delimiters or protective markers when passing extracted website content to subsequent agent prompts.
Capability inventory: The skill possesses extensive capabilities, including writing to the local filesystem and executing shell commands via npm and pnpm.
Sanitization: There is no evidence of sanitization or filtering of external content before it is used in the building pipeline.
[COMMAND_EXECUTION]: The skill makes use of powerful system commands such as npm create, pnpm install, pnpm build, and background dev server execution. While necessary for the intended workflow, these provide a platform for persistent or complex modifications to the host environment.
[EXTERNAL_DOWNLOADS]: The skill automatically fetches and stores assets (images, videos, fonts) from any provided URL. This interaction with external, untrusted servers for asset retrieval is a fundamental part of the 'Reconnaissance' phase.

dineway-building-clone