The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites to guide the reconstruction of code, styles, and layouts.
Ingestion points: Data enters the agent's context through browser automation tools such as browser_navigate and the page.evaluate scripts provided in SKILL.md, which extract DOM structure and computed styles from external URLs.
Boundary markers: The skill includes a 'Lane' system (Faithful vs. Reference) and specific instructions to 'Decline' cloning of sensitive pages (e.g., bank logins, payment portals), which serves as a conceptual boundary against processing malicious or deceptive content.
Capability inventory: The skill has access to file system operations (fs.writeFileSync, fs.mkdirSync), network operations via the browser, and the ability to install and execute Node.js packages.
Sanitization: The instructions require the agent to manually verify the right to copy and explicitly omit tracking IDs, secrets, or third-party widgets found in the source code.
[COMMAND_EXECUTION]: The workflow relies on the execution of several CLI commands to set up the environment and perform visual regression testing.
Evidence: The instructions include npm install --save-dev playwright pixelmatch pngjs, npx playwright install chromium, and the execution of Node.js scripts for site capture (node capture/shots.mjs) and image comparison (node capture/diff.mjs).
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading web assets and browser binaries.
Evidence: Use of playwright to fetch remote website assets (images, CSS, scripts) and npx playwright install chromium to download browser binaries from well-known sources.

clone-website