mrbeast-perspective

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

• Third-party content exposure detected (high risk: 1.00). The skill's required Agentic Protocol Step 2 explicitly mandates using WebSearch and other tools to fetch real-world data ("⚠️ 必须使用工具（WebSearch等）获取真实信息，不可跳过") and includes runtime scripts like scripts/fetch_youtube_subtitles.sh that download YouTube content, so the agent will ingest untrusted public web/user-generated content (webpages, social media/YouTube) and base decisions/actions on it.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill mandates runtime web research and includes a local script (fetch_youtube_subtitles.sh) that downloads arbitrary YouTube URLs (e.g., https://www.youtube.com/watch?v=) and uses the fetched subtitle/content as required input to drive the agent's prompts/decisions, so runtime-fetched external content can directly control the agent.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)