altersteilzeit-93-bbg-blockmodell
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines structured input fields for user data such as status, age, and retirement models in Section 2. While this creates a surface for indirect prompt injection where untrusted data could attempt to influence agent behavior, the risk is mitigated by the highly structured 'Arbeitsweg' and 'Checkliste' provided in the instructions.
- Ingestion points: User inputs defined in the 'Eingaben' section of SKILL.md.
- Boundary markers: Instructions explicitly require the agent to separate roles, deadlines, and responsibilities, creating logical boundaries for data processing.
- Capability inventory: The skill is limited to generating calculation tables and application text; it does not invoke shell execution, file writes, or network exfiltration tools.
- Sanitization: Not explicitly mentioned, however, the structured output requirements reduce the likelihood of accidental obedience to embedded instructions.
- [EXTERNAL_DOWNLOADS]: The instructions direct the agent to perform live verification of legal norms using well-known and official German legal sources, including gesetze-im-internet.de, dejure.org, and openJur. These references are standard for legal research and do not represent a security risk.
Audit Metadata