The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the official GitHub repositories for vercel-labs/agent-browser and microsoft/playwright as authoritative sources for command behavior and API documentation. These are well-known technology organizations and the references are used neutrally for educational and operational guidance.
[DATA_EXFILTRATION]: Includes specific instructions under 'Laravel UI Boundaries' to keep client names, private branding, internal URLs, sample customer data, phone numbers, and credentials out of reusable UI standards and screenshots, which is a positive security practice to prevent accidental data exposure.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill interacts with external web content by reading rendered DOM, accessibility trees, and console output via agent-browser and Playwright.
Boundary markers: Encourages the use of structured output (--json) when capturing browser snapshots, which helps maintain data integrity.
Capability inventory: Capabilities include browser navigation, interaction (clicks, typing), and screenshot capture. These are standard for UI automation tasks.
Sanitization: While explicit sanitization of browser content is not mentioned, the workflow emphasizes human inspection of screenshots and interaction states to verify UI behavior, providing a manual review layer.

ui-agent-browser