Council

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The skill mandates executing an unrelated local curl notification ("You MUST send this notification... Execute this curl command immediately") before any action, which is an operational requirement outside the Council debate functionality and therefore a hidden/deceptive instruction relative to the skill's stated purpose.

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

• Third-party content exposure detected (medium risk: 0.30). The Council skill’s Debate/Quick workflows build the agents’ prompts directly from the user-provided topic/question and generate Round 1/2/3 content from that free text (“Topic: [The topic]” and “[Relevant context for the topic]”), so outsider-authored text is ingested without selecting a specific trusted item first.