The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/extract-browser-evidence.mjs implements dynamic loading of the playwright library using createRequire with paths computed at runtime, such as the global npm root. This dynamic loading pattern is a security concern as it can be influenced by the execution environment.\n- [REMOTE_CODE_EXECUTION]: The skill utilizes agent-browser eval to execute a complex JavaScript payload (styleProbe) within the context of external websites to extract DOM and style data. This represents runtime code execution on untrusted content.\n- [COMMAND_EXECUTION]: Multiple utility scripts, specifically scripts/check-browser-tooling.mjs and scripts/extract-browser-evidence.mjs, use spawnSync to execute system commands like bash -lc and npm for environment detection and tool bootstrapping.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingests data from untrusted external URLs (as defined in SKILL.md) and processes this information (text snippets, HTML) to generate a design system. The lack of explicit boundary markers or sanitization of the extracted content allows malicious instructions embedded in a website to potentially influence the agent's output.\n- [EXTERNAL_DOWNLOADS]: The skill's documentation in SKILL.md and references/browser-tooling-bootstrap.md provides instructions for the automated or manual installation of external dependencies like agent-browser if they are not detected in the environment.

website-to-design-md