The Agent Skills Directory

[COMMAND_EXECUTION]: The skill identifies and executes local browser binaries (Chrome, Brave, Edge, Chromium) to perform web searches and page rendering. It uses child_process.spawn to launch these processes with specific flags like --remote-debugging-port and --no-sandbox. This is a core part of its functionality.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch search results from Google and DuckDuckGo, and downloads arbitrary web page content for processing. It also manages its own dependencies via bun install during setup.
[PROMPT_INJECTION]: The skill processes untrusted content from the public internet and converts it into Markdown for the AI agent. This presents an indirect prompt injection surface where malicious web pages could attempt to influence the agent's behavior through their content.
Ingestion points: Web content is ingested via lib/fetch.js and lib/search.js.
Boundary markers: The skill does not include explicit boundary markers or warnings in its output to help the agent distinguish between tool metadata and fetched content.
Capability inventory: The agent can invoke the web-search.js CLI which has network access and the ability to spawn local processes.
Sanitization: The skill uses @mozilla/readability and turndown to sanitize HTML into Markdown, which removes scripts and styling but preserves text content.

web-search