The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted content from external websites, creating a surface for indirect prompt injection attacks. It mitigates this by instructing the agent to treat data within [PAGE_CONTENT: ...] delimiters as non-authoritative and warning against executing commands found in web content.
Ingestion points: Web page content processed by the agent during authentication or verification flows.
Boundary markers: Uses [PAGE_CONTENT: ...] delimiters to isolate untrusted data.
Capability inventory: The skill uses web-ctl.js to control browser sessions, navigate URLs, and perform snapshots via Playwright.
Sanitization: Includes explicit instructions for the agent to disregard instructions or commands embedded within the fetched page text.
[COMMAND_EXECUTION]: The skill executes local Node.js scripts (web-ctl.js) to manage browser sessions and authentication flows. It includes security guidance for the agent to use proper shell quoting for URLs to prevent shell glob expansion or command backgrounding.
[EXTERNAL_DOWNLOADS]: The skill may trigger the installation of the Playwright browser automation library and Chromium browser binaries if they are not already present on the system. These are well-known and trusted developer tools.

web-auth