The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites to generate code and component specifications, creating a potential surface for indirect injection attacks.
Ingestion points: The skill uses Playwright MCP tools (browser_navigate, browser_snapshot, browser_evaluate) to ingest content from arbitrary URLs provided as arguments.
Boundary markers: There are no specific delimiters or instructions provided to the agent to distinguish between the content of the target website and the agent's internal instructions during the extraction and spec-writing phases.
Capability inventory: The skill has the capability to write files to multiple directories (docs/, src/, public/, scripts/), execute local scripts (scripts/download-assets.mjs), and run project build commands (npm run build, npx tsc).
Sanitization: The workflow does not specify sanitization or validation steps for text, CSS, or other DOM data extracted from the live site before it is used to generate component specs or source code.
[COMMAND_EXECUTION]: The skill relies on the execution of local scripts and standard development tools to complete the cloning process.
It explicitly instructs the agent to run scripts/download-assets.mjs for asset retrieval.
It requires running npx tsc --noEmit and npm run build to verify the integrity of the cloned project.
[EXTERNAL_DOWNLOADS]: The skill's primary function involves downloading assets from external, user-defined URLs.
The Asset Discovery Script in references/clone-workflows-reference.md is used to enumerate and extract URLs for images, videos, background images, and favicons from the target website for local storage.

clone-website