scenario-creator

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

• Third-party content exposure detected (high risk: 0.75). Runtime path: Step 3 “URLs provided … fetch them” and “Pasted material … mine it” means the agent ingests outsider-authored free text from arbitrary public web pages or user-pasted third-party transcripts/docs into the LLM context before drafting the ai_instructions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill declares and uses the ToughTongue MCP tool at runtime (ttai mcp URL "https://api.toughtongueai.com/api/public/mcp" in agents/openai.yaml) and explicitly permits server-side generation (ttai:generate_scenario) which returns ai_instructions/user_instructions that would directly control agent prompts, so the external API is a runtime dependency that can control prompts.