israeli-insurance-duplication-checker
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: No malicious injection patterns or attempts to bypass safety filters were detected. The instructions include safety guidelines for the AI (e.g., "never let them cancel before replacement cover is confirmed active") to protect the user from irreversible financial loss. A static detection hint regarding concealment was found to be a false positive, as the text actually encourages transparency regarding legal risks.
- [DATA_EXFILTRATION]: No hardcoded credentials or unauthorized network operations were found. The skill directs users to legitimate, official government and clearing house websites (e.g., harb.cma.gov.il, swiftness.co.il) to retrieve their own insurance data.
- [EXTERNAL_DOWNLOADS]: The skill recommends a model context protocol (MCP) server hosted at agentskills.co.il. This domain is associated with the skill's author (skills-il) and represents standard vendor infrastructure for extending agent capabilities.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided insurance data. While this creates an attack surface for indirect prompt injection, the instructions include significant legal warnings, verification requirements, and structured output formats that mitigate the risk of accidental obedience to embedded instructions.
- [DYNAMIC_EXECUTION]: The skill generates a Hebrew cancellation letter based on a static template in references/cancellation-letter-he.md. This is a benign use of template-based generation from local resources.
Audit Metadata