dragon-knowledge-search
Pass
Audited by Gen Agent Trust Hub on May 14, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill uses instructional language to guide the agent's autonomy in identifying internal company context (e.g., 'Use this skill aggressively', '不需要用户显式说'). This is intended behavior for a specialized search skill and does not attempt to bypass core safety guardrails or ignore user instructions.
- [DATA_EXFILTRATION]: No evidence of unauthorized data exfiltration. The skill identifies internal keywords and domains (e.g., didainternal.com, system names like LionKing, Beaver) to trigger searches within the company's own knowledge base via a provided MCP tool. All retrieved data remains within the conversation context.
- [EXTERNAL_DOWNLOADS]: The README.md references an internal documentation link on Feishu (didatravel.feishu.cn), which is a well-known enterprise collaboration platform. This is a legitimate dependency for the skill's setup and not a hidden or suspicious download.
- [SAFE]: The skill focuses on enhancing the agent's utility for internal company development tasks at Dida Travel. It demonstrates security best practices by requiring the agent to notify users before performing searches ('我先查一下 Dragon 知识库...') and providing clear attribution for the information retrieved.
Audit Metadata