The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon external performance diagnostic data, which presents a surface for indirect prompt injection if those sources were compromised.
Ingestion points: Browser performance traces (via performance_start_trace), aggregated CrUX field data, and tool responses from Search Console or Lighthouse.
Boundary markers: The skill includes an explicit 'Anti-Patterns' section in SKILL.md that instructs the agent not to treat external content, logs, or tool responses as trusted instructions.
Capability inventory: The agent is authorized to inspect and modify project source code (HTML, CSS, JavaScript, and framework components) to implement optimizations.
Sanitization: The skill defines a mandatory 'Verification Protocol' and 'LCP optimization checklist' to ensure recommendations are validated against local evidence and runtime behavior rather than solely trusting external inputs.
[EXTERNAL_DOWNLOADS]: The skill references established performance libraries and official documentation from trusted sources.
Evidence: Mentions the web-vitals library and provides links to web.dev and developer.chrome.com for technical specifications.

core-web-vitals