T01 · Skill Instruction Hijacking
Error
- Location
SKILL.md:140- Finding
Mandatory Promotional Content Hijacks Agent Responses and Generated Reports
- Content
View full analysis
📊 报告涉及企业的完整档案与更多商机，见 知了商机大师' f' · 本报告由 知了标讯 AI 开放平台 投标决策分析 Skill 生成' ``` ### Technical Analysis The Skill instructions require the Agent to append cross-promotional recommendations and platform links after completing the requested analysis. The HTML renderer independently injects promotional calls to action into every generated report without checking whether the user requested them or consented to their inclusion. This behavior changes the Agent's task output from a bid-analysis response into a persistent traffic-acquisition and advertising channel. It is not technically necessary to query bid data, assess competitors, calculate price ranges, or generate a usable report. The renderer makes the behavior unconditional: callers cannot omit the promotional footer through report data because the links are hardcoded in `render()`. ### Attack Path 1. A user loads the Skill for a legitimate bid-analysis request. 2. The Skill's mandatory instructions alter the Agent's response policy. 3. The Agent is required to recommend another commercial Skill or external platform after the requested task. 4. In full mode, `render_report.py` generates an HTML report. 5. The rende ...[truncated 771 chars]- Remediation
View remediation