The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The scripts/install-deps.sh and scripts/install.sh files use sudo to install system dependencies. Although required for running Chromium on Linux, this grants the skill elevated privileges during the setup phase.\n- [REMOTE_CODE_EXECUTION] (MEDIUM): The scripts/evaluate.js script enables arbitrary JavaScript execution within the page context via eval(). This is an intended feature but represents a code execution surface that could be exploited via malicious web content.\n- [EXTERNAL_DOWNLOADS] (MEDIUM): The installation process fetches packages from NPM and system repositories. While these are standard sources, the extensive dependency chain should be reviewed by users before execution with root privileges.\n- [PROMPT_INJECTION] (LOW): The skill is susceptible to indirect prompt injection (Category 8) due to its processing of untrusted external web content.\n
Ingestion points: Web page data accessed through navigate.js, snapshot.js, and console.js.\n
Boundary markers: Absent. The skill does not implement delimiters or warnings when passing external content to the agent.\n
Capability inventory: Arbitrary JavaScript execution (evaluate.js), local filesystem writes (screenshot.js, network.js), and process execution (ImageMagick via execFileSync).\n
Sanitization: selector.js provides robust validation for XPath and CSS selectors, but the skill lacks sanitization for the JavaScript strings passed to the evaluate.js script.

chrome-devtools