citation-anchoring
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFE
Full Analysis
- [Indirect Prompt Injection] (SAFE): The skill processes untrusted content from external files. While this presents an ingestion surface, the risk is minimal given the absence of execution capabilities.
- Ingestion points: Reads
output/DRAFT.mdandoutput/citation_anchors.prepolish.jsonl. - Boundary markers: None (absent).
- Capability inventory: Limited to reading files and writing the
output/CITATION_ANCHORING_REPORT.mdreport. - Sanitization: None (absent).
- [Data Exposure & Exfiltration] (SAFE): The skill reads local files in the
output/directory but has no defined network capabilities to exfiltrate information. - [Remote Code Execution] (SAFE): No external dependencies, script downloads, or dynamic execution patterns were detected.
Audit Metadata