brand-mention-monitor

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

• Third-party content exposure detected (high risk: 0.95). Outsider free text is ingested into the LLM context via the runtime web/social mention content returned by Nimble searches/extractions (e.g., Reddit/X/news post bodies and excerpts authored by non-user parties), which the skill then embeds into the HTML/markdown feed and cards.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill's required HTML template loads and executes third-party JavaScript at render time (e.g., https://cdnjs.cloudflare.com/ajax/libs/d3/7.8.5/d3.min.js and https://cdnjs.cloudflare.com/ajax/libs/topojson/3.0.2/topojson.min.js) and fetches remote geometry data (e.g., https://cdn.jsdelivr.net/npm/world-atlas@2.0.2/countries-110m.json), so these external URLs are used during runtime, execute remote code in the interactive widget, and are required dependencies for the skill's inline triage console.