The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface for external data by instructing the agent to read attachments, metadata, and case files (Section 1.1). This constitutes a standard attack surface for document-processing skills. The instructions mitigate this risk in Section 1.3 by explicitly forbidding the invention of facts or citations and requiring a clear separation between source facts and conclusions.- [SAFE]: The skill's purpose is consistent with its behavior. It handles sensitive legal and financial data (mandates, invoices, KYC data) as part of its primary function, but does not include any commands for external exfiltration, unauthorized file modifications, or privilege escalation.- [SAFE]: No obfuscation, multi-layer encoding, or hidden URLs were found. All instructions and references are in plain German and target internal agent workflows.

juristischer-argumentationskern