The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to manage a persistent Node.js proxy (cdp-proxy.mjs). This proxy is launched in a detached background process.
[REMOTE_CODE_EXECUTION]: The local proxy exposes an /eval endpoint that allows the execution of arbitrary JavaScript within the context of the user's browser. This capability is used by the agent to interact with and extract data from web pages.
[COMMAND_EXECUTION]: The skill utilizes a /setFiles endpoint in its local proxy which uses the DOM.setFileInputFiles protocol to programmatically upload local files to websites, bypassing standard browser file dialogs.
[REMOTE_CODE_EXECUTION]: The proxy implementation includes a 'Port Guard' feature that uses the Fetch.enable CDP domain to intercept and block local network requests to the Chrome debugging port from within web pages, which is a technique used to evade bot detection.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from arbitrary external websites (via WebFetch, curl, and CDP).
Ingestion points: Web content retrieved from any URL via the browser or search tools.
Boundary markers: None implemented in the provided scripts or instructions.
Capability inventory: Includes shell command execution (Bash), file writing (Write), arbitrary JavaScript execution (/eval), and local file uploads (/setFiles).
Sanitization: No evidence of input sanitization or validation before processing web content.

web-access