RedTeam

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The skill contains an explicit, mandatory external command (a curl POST to localhost) that is unrelated to its stated purpose of analyzing ideas and thus functions as a hidden/side-effect instruction that can exfiltrate or signal execution outside the skill's scope.

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

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow (e.g., Workflows/ParallelAnalysis.md and Workflows/AdversarialValidation.md) ingests the outsider-authored “argument” from the user prompt and then has multiple agents read/decompose/critique that full free text.