skills/simota/agent-skills/levy/Gen Agent Trust Hub

levy

Pass

Audited by Gen Agent Trust Hub on Apr 10, 2026

Risk Level: SAFENO_CODEEXTERNAL_DOWNLOADS
Full Analysis
  • [SAFE]: The skill provides domain-specific information on Japanese tax laws and contains strict instructions to never write code, never store sensitive personal information such as My Number or bank accounts, and never provide individualized tax advice in accordance with the Tax Accountant Act.\n- [NO_CODE]: This skill consists of markdown instructional files and configuration templates without any associated scripts, executables, or binaries.\n- [SAFE]: A four-tier guardrail system (L1-L4) is implemented to ensure the agent handles requests safely, with explicit rules to refuse assistance for illegal activities such as tax evasion or audit avoidance.\n- [EXTERNAL_DOWNLOADS]: The skill references official and well-known technology and government services, including the National Tax Agency (nta.go.jp) and the Japan Federation of Certified Public Tax Accountants' Associations (nichizeiren.or.jp), for user verification.\n- [SAFE]: Indirect prompt injection surfaces are addressed. Ingestion point: User financial data and questions (SKILL.md). Boundary markers: Mandatory disclaimer templates (disclaimer-templates.md). Capability inventory: None detected (no subprocess, file-write, or network operations). Sanitization: Procedural instructions to prevent the logging of sensitive financial identifiers.\n- [SAFE]: No malicious patterns involving prompt injection, obfuscation, persistence mechanisms, or privilege escalation were detected.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 10, 2026, 04:22 PM