redteam-halluzinationsschutz
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No security issues detected. The skill operates within a defined legal analysis framework and does not utilize risky capabilities such as command execution, network operations, or file system writes.
- [INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing external data (Step 3: original files, metadata, contracts, and screenshots) to perform legal audits.
- Ingestion points: Data ingestion occurs through the "Beweise anfordern" (Request Evidence) step in
SKILL.md. - Boundary markers: Not explicitly defined in the instructions.
- Capability inventory: None. The skill does not contain subprocess calls, network requests, or file writing operations.
- Sanitization: The skill's primary function is to act as a sanitization and validation layer ("Halluzinationsschutz") by explicitly instructing the agent to verify sources and flag unsupported claims, which mitigates typical injection risks.
Audit Metadata