datenbank-und-softwaredaten
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external documents such as contracts, legal briefs, administrative acts, and expert opinions, which serves as an ingestion point for untrusted data that could contain malicious instructions designed to influence the agent's behavior.
- Ingestion points: The 'Arbeitsweg' and 'Workflow' sections of SKILL.md specify the ingestion of 'Verwaltungsakte, Vertragsurkunden, Schriftsätze, Bescheide, Protokolle, Sachverständigengutachten'.
- Boundary markers: The skill does not define clear delimiters or specific instructions for the agent to ignore hidden directives within these documents.
- Capability inventory: The skill primarily involves textual analysis; no scripts, system commands, file writing, or network exfiltration capabilities were detected.
- Sanitization: There are no instructions for sanitizing or filtering content from the ingested documents before processing.
Audit Metadata