xuefeng-agent-gaokao-advisor
Pass
Audited by Gen Agent Trust Hub on Jun 21, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references its official source code on GitHub (
github.com/ziqihe10-droid/xuefeng-agent.git) and requires standard Python dependencies (flask,openai,tavily-python). These resources are consistent with the skill's primary functionality. - [PROMPT_INJECTION]: Identified an indirect prompt injection surface. The skill ingests data from external web searches and interpolates it into LLM prompts to provide school recommendations.
- Ingestion points: Results from
web_search(Tavily/Baidu) are used in context synthesis withinserver.pyand Example 2. - Boundary markers: Absent in the provided prompt templates for the synthesis step.
- Capability inventory: Local database reads (
sqlite3), network requests (requests,TavilyClient), and structured JSON extraction. - Sanitization: SQL queries are properly parameterized to prevent SQL injection, though web content is not explicitly sanitized before prompt interpolation.
- [COMMAND_EXECUTION]: The skill documentation includes instructions for launching the application via a Windows batch file (
启动.bat) or the Python interpreter. These are standard deployment methods and do not exhibit malicious intent.
Audit Metadata