rechnung
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as case files, correspondence, and timesheet narratives which could potentially harbor hidden instructions.
- Ingestion points: Data is sourced from external files including the Zeit- und Narrative-Ledger, Mandatsvereinbarung, Schriftsätze (pleadings), and Verwaltungsakte (administrative records) as described in the 'Datenquellen' and 'Arbeitsweg' sections.
- Boundary markers: The instructions do not define explicit delimiters or 'ignore' instructions to isolate untrusted data from the system prompt during processing.
- Capability inventory: The skill has capabilities to read local template files (e.g., assets/templates/rechnungsdatenblatt.md) and generate structured output (PDF, Markdown, XRechnung). No network exfiltration tools or arbitrary command execution were identified.
- Sanitization: The skill includes a dedicated 'Narrative-Übernahme' section that instructs the agent to review all positions for suitability, explicitly removing internal secrets and ensuring data is client-ready.
Audit Metadata