The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted HTML markup, network traffic, and web storage data from external websites during its audit process. This creates an attack surface where an attacker-controlled site could include malicious instructions (e.g., in hidden HTML comments or metadata) to influence the AI agent's behavior.
Ingestion points: scripts/lib/pages.mjs and scripts/observe-runtime.mjs fetch and read HTML content and network data from user-provided URLs.
Boundary markers: The skill does not implement explicit delimiters or safety warnings for the agent when processing external content.
Capability inventory: The skill utilizes Node.js built-ins for file system access (node:fs) and network operations (fetch, Playwright) across its scripts.
Sanitization: No sanitization or filtering of external content is performed beyond basic HTML entity decoding.
[DYNAMIC_EXECUTION]: The helper script scripts/lib/browser.mjs dynamically resolves and imports the playwright or playwright-core packages from the local project environment using createRequire and import(). While this is a functional requirement for browser-based performance auditing, it involves the dynamic loading of executable code from computed paths at runtime.

performance-readiness