WorldThreatModelHarness
Pass
Audited by Gen Agent Trust Hub on Mar 16, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The workflows in
TestIdea.md,UpdateModels.md, andViewModels.mdusecurlto send POST requests tohttp://localhost:8888/notify. This is used to provide the user with voice notifications regarding the status of the long-running analysis and research tasks. - [PROMPT_INJECTION]: The skill's
UpdateModels.mdworkflow ingests data from web research and user-provided analysis to generate world model documents. Because these documents are later used as context for theTestIdea.mdworkflow, there is a potential surface for indirect prompt injection. However, the risk is mitigated by the structured nature of theModelTemplate.mdand the adversarial nature of the testing tiers (using RedTeam and FirstPrinciples skills).
Audit Metadata