The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/perfection/lighthouse-audit.py contains a recommendation for the user to download and execute a remote script via curl -LsSf https://astral.sh/uv/install.sh | sh in its documentation comments.
[COMMAND_EXECUTION]: The Python script scripts/perfection/lighthouse-audit.py automatically performs system-wide global installations of npm packages (npm install -g lighthouse chrome-launcher) using subprocess.run if dependencies are not detected.
[REMOTE_CODE_EXECUTION]: The skill mandates the execution of several external tools using npx (e.g., grab, react-scan, react-doctor), which involves downloading and executing unverified code from the npm registry at runtime.
[EXTERNAL_DOWNLOADS]: The skill fetches content and source code from multiple third-party domains, including beui.dev, reactbits.dev, and lazyweb.com, as well as several non-whitelisted GitHub repositories (e.g., github.com/changeroa/StyleGallery).
[DYNAMIC_EXECUTION]: The lighthouse-audit.py script dynamically generates a Node.js script during execution, writes it to a temporary file, and runs it via a subprocess call to node.
[DATA_EXFILTRATION]: The lazyweb.md reference provides a mechanism for the agent to send queries and authentication tokens to the lazyweb.com domain, creating a potential vector for exfiltrating sensitive project data via prompt injection.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from numerous untrusted external websites, code registries, and search APIs, creating a significant attack surface for indirect prompt injection.

frontend