toulmin-model

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

• Third-party content exposure detected (medium risk: 0.30). The skill’s runtime path (Phases 1–6) ingests the user’s free-text “actual argument already in context” at each gate (e.g., Phase 1 Gate and HALT, Phase 2 Gate and HALT, etc.) and uses it as untrusted data to evaluate; this is outsider-authored input because the user can type/paste it into the interaction.