research-paper-refiner
Pass
Audited by Gen Agent Trust Hub on Jul 1, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is purely instructional, providing a knowledge base for linguistic improvement. No prompt injection or behavior-override attempts were detected.
- [SAFE]: No data exfiltration or credential exposure patterns were found. The skill does not perform network requests or access sensitive local files.
- [SAFE]: The skill contains no scripts, dependencies, or remote code execution patterns. It operates solely within the context of text transformation.
- [SAFE]: Regarding indirect prompt injection (Category 8): The skill ingests user-provided paper fragments (Ingestion Point: SKILL.md '待润色文本'). It defines specific markdown headers for output (Boundary Markers). The capability inventory is empty, as it performs no subprocess calls, file writes, or network operations. No specific sanitization is implemented, but the lack of dangerous capabilities renders the ingestion surface safe.
Audit Metadata