The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the chromux CLI tool using node:child_process.execFileSync. This implementation is safe as it avoids shell-based command injection by passing arguments as an array rather than a single string. It is used to launch the browser and execute JavaScript for data extraction within the browser context.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it retrieves and processes untrusted text from the public web.
Ingestion points: The vendor/web-search.mjs script fetches search results and full page content (body text and comments) from arbitrary external websites.
Boundary markers: The output formatting provided to the agent does not use specific delimiters or protective instructions to help the model distinguish between tool-provided data and embedded malicious instructions.
Capability inventory: The skill has the ability to run shell commands (via chromux) and perform network operations (via the browser). All subprocess calls are confined to the chromux binary within vendor/web-search.mjs.
Sanitization: The script performs basic data cleanup, such as removing extra whitespace and limiting string length, but it does not filter for potential prompt injection patterns in the scraped content.

google-search