The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites when performing performance audits.
Ingestion points: Data is ingested through evaluate_script, list_network_requests, and performance traces fetched from target URLs.
Boundary markers: None; the skill treats data from the analyzed page as the primary source for diagnostic output.
Capability inventory: The skill utilizes evaluate_script to run JavaScript and navigate_page to interact with websites.
Sanitization: No explicit sanitization or filtering of external DOM content or performance metadata is performed before the agent processes it.
[DYNAMIC_EXECUTION]: The skill uses the evaluate_script tool to execute predefined JavaScript snippets found in references/lcp-snippets.md to extract performance metrics.
Evidence: The snippets use PerformanceObserver and document.querySelectorAll to audit the DOM and performance timeline, which is standard practice for performance monitoring tools.

debug-optimize-lcp