The Agent Skills Directory

[COMMAND_EXECUTION]: The wrapper script scripts/main.cjs uses child_process.spawn to execute a local JavaScript file located at .claude/tools/chrome-browser/chrome-browser.cjs.
[REMOTE_CODE_EXECUTION]: The skill provides tools for dynamic code execution within the browser context, such as mcp__chrome-devtools__evaluate_script and mcp__claude-in-chrome__javascript_tool. These allow the agent to run arbitrary JavaScript on any website it visits.
[DATA_EXFILTRATION]: The skill is designed to interact with authenticated web applications. Tools like get_network_request, take_screenshot, and get_page_text can be used to capture sensitive personal or corporate data from logged-in sessions.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted external web content.
Ingestion points: Web page content (DOM), accessibility trees, console logs, and network traffic via tools like take_snapshot and read_page.
Boundary markers: None present in the instructions to differentiate between site content and agent instructions.
Capability inventory: Arbitrary JavaScript execution, network inspection, form filling, and interaction with authenticated apps.
Sanitization: No sanitization or validation of the ingested browser data is performed before it is presented to the agent.

chrome-browser