agentische-datenbank-recherche
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from both user uploads ("Erfindungsmaterial") and external patent databases (Espacenet, Google Patents, DPMA, etc.). This creates a vulnerability surface where malicious instructions embedded in the processed data could influence the agent's behavior.
- Ingestion points: Processes user-supplied descriptions, claims, and sketches, as well as live search results and full-text documents from multiple external patent registries as described in SKILL.md.
- Boundary markers: The instructions lack specific markers or delimiters to isolate untrusted data from instructions, increasing the risk of the agent obeying embedded commands.
- Capability inventory: The agent is instructed to perform complex browser-based tasks including URL navigation, data collection from table structures, and invoking subsequent skills based on analysis results.
- Sanitization: There is no mention of sanitizing, escaping, or validating the content retrieved from external databases or user files before processing.
Audit Metadata