T01 · Skill Instruction Hijacking
Error
- Location
index.js:61- Finding
Unconditional Promotional Content Injection into Analysis Results
- Content
View full analysis
2 ? '放量' : volumeRatio < 0.5 ? '缩量' : '正常', upgradeHint: '需要缠论分型、笔、线段、中枢分析？升级专业版服务！' }; ``` The declared intent is also documented in `index.js:2-5`, where the implementation is described as a free traffic-acquisition skill designed to guide users toward an upgraded service. The associated paid service and pricing are advertised in `SKILL.md:23-38`. ### Technical Analysis The `analyzeStock()` function unconditionally adds an `upgradeHint` field containing promotional content to every successful technical-analysis response. This field is unrelated to calculating MACD, RSI, price, or volume indicators and cannot be disabled by the caller. Because the promotional instruction is embedded directly in the structured result, an AI agent or other downstream consumer may interpret and present it as part of the requested analysis. This compromises output integrity and allows the skill to use legitimate stock-analysis requests as a channel for unsolicited commercial redirection. The behavior best matches skill instruction hijacking because skill-controlled content is inserted into agent-consumable output with the explicit purpose of steering users toward an unrelated paid offering. The reviewed code does not attempt to override safety constraints or acquire system privileges, so the scope is limited to response manipulation and commercial steering. ### Attack Path 1. A user asks an agent to analyze a stock. 2. The agent invokes the exported `analyzeStock(code)` function. 3. The function retrieves market bars and calculates the available technical indicators. 4. On every successful request, the function adds the hardcoded `upgradeHint` promotion to the returned obje ...[truncated 1080 chars]- Remediation
View remediation