office-research

Pass

Audited by Gen Agent Trust Hub on Jul 6, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: No malicious patterns, obfuscation, or unauthorized commands were detected in the skill instructions.
  • [PROMPT_INJECTION]: The skill processes untrusted external information from web searches and user-provided materials, which constitutes a potential surface for indirect prompt injection. Ingestion points: External information from web search results and user documents (SKILL.md). Boundary markers: No explicit prompt boundary markers or ignore-previous-instruction warnings are defined in the instructions. Capability inventory: The skill uses web retrieval tools to gather data. Sanitization: The instructions include defensive guidelines requiring the agent to cite sources, identify unknown items, and verify reliability of the findings.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 6, 2026, 01:34 AM