The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the anti-detect-browser package from the npm registry and fetches a browser kernel (approximately 190-320 MB) from the vendor's infrastructure during the initial setup. These downloads are required for the skill's primary function and originate from the vendor's own distribution channels.
[COMMAND_EXECUTION]: The skill's setup instructions utilize npx to execute the MCP server. It also provides an evaluate tool that enables the AI agent to run JavaScript within the browser context to interact with web pages.
[PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection by design, as it allows the agent to ingest and act upon untrusted data from the internet. It mitigates this risk through a mandatory evidence chain: 1) Ingestion points: get_content, screenshot, and evaluate tools return third-party web content; 2) Boundary markers: explicit instructions are provided to the agent to treat page content as data and not as instructions; 3) Capability inventory: the agent can perform actions such as navigate, click, fill, and evaluate based on its interpretation of page data; 4) Sanitization: the skill relies on the agent's logical filtering as no programmatic sanitization of the DOM content is described.

browser-mcp-agent