core-web-vitals

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.90). The prompt includes an out-of-scope "Companion check" that explicitly instructs the agent to list local skill directories and offer installing packages via shell commands (ls and npx), which is unrelated to Core Web Vitals and attempts to access/modify the local environment—i.e., a hidden/extraneous instruction to perform potentially sensitive actions.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.70). This skill's required workflow instructs the agent to fetch and act on third‑party site data—e.g., the "Diagnose which CWV metric is failing" task (SKILL.md) and scripts in references/lighthouse-ci.md and scripts/check-crux.js call the CrUX API (https://chromeuxreport.googleapis.com) and run Lighthouse against arbitrary public/staging URLs, so external web-origin data is ingested and used to drive diagnostics and actions.