The Agent Skills Directory

[COMMAND_EXECUTION]: The skill launches a local browser instance (Chrome or Chromium) with remote debugging enabled to perform searches. The browser's lifecycle is managed via a bridge server bound to the local loopback address to prevent external network exposure.
[COMMAND_EXECUTION]: The bash scripts search.sh and start-server.sh include a fallback mechanism that generates and executes Node.js code via standard input to perform HTTP requests. This is used to ensure cross-platform functionality when curl or wget are unavailable.
[REMOTE_CODE_EXECUTION]: The browser operations module contains an internal function that uses eval() to execute JavaScript within the page context. This is utilized by the search engines to extract results from the DOM and is not exposed as a general-purpose execution endpoint to the API.
[EXTERNAL_DOWNLOADS]: The start-server.sh script manages dependencies by executing npm install, which downloads required packages like playwright-core and express from the official npm registry.
[DATA_EXFILTRATION]: The CLI search tool supports reading query input from local files using the @ prefix. While intended for non-ASCII character support, this allows for the ingestion of local file content into the search process if a specific file path is provided.
[PROMPT_INJECTION]: The skill processes untrusted data from web search results (snippets and titles). While it lacks explicit boundary markers for this data, the risk of indirect injection is mitigated by the agent's internal safety protocols and the limited context of the extracted snippets.

web-search