扫描报告
5 /100
recognize_intent
识别自然语言意图类别并解析语义指标与维度(零售BI问数系统)
This is a legitimate BI intent-recognition skill for a retail data query system. All functionality (Gemini LLM calls, Milvus vector search, MySQL metric config loading) is fully documented, necessary, and standard for its stated purpose.
可以安装
No action required. The skill is safe to use.
安全发现 1 项
| 严重性 | 安全发现 | 位置 |
|---|---|---|
| 低危 | Hardcoded IP address as default Gemini API backend 文档欺骗 | recognize_intent.py:969 |
| 资源类型 | 声明权限 | 推断权限 | 状态 | 证据 |
|---|---|---|---|---|
| 文件系统 | READ | READ | ✓ 一致 | SKILL.md declares READ from .workflow/, code reads rewrite_output.json |
| 网络访问 | READ | READ | ✓ 一致 | SKILL.md declares external service injection (Milvus, MySQL, Gemini); all outbou… |
| 命令执行 | NONE | NONE | — | No shell execution found in any file |
| 数据库 | NONE | READ | ✓ 一致 | MySQL reads for metric config are injected via Protocol interfaces; not hardcode… |
1 高危 3 项发现
高危 IP 地址 硬编码 IP 地址
47.77.199.56 recognize_intent.py:969 中危 外部 URL 外部 URL
http://47.77.199.56/api/v1beta recognize_intent.py:969 中危 外部 URL 外部 URL
https://dashscope.aliyuncs.com/compatible-mode/v1 recognize_intent.py:1016 目录结构
3 文件 · 93.0 KB · 2119 行 Python 1f · 1930L
JavaScript 1f · 135L
Markdown 1f · 54L
├─
index.js
JavaScript
├─
recognize_intent.py
Python
└─
SKILL.md
Markdown
依赖分析 5 项
| 包名 | 版本 | 来源 | 已知漏洞 | 备注 |
|---|---|---|---|---|
httpx | * | pip | 否 | HTTP client for Gemini REST calls; not malicious |
pymilvus | * | pip | 否 | Vector database client; only for metric search |
pymysql | * | pip | 否 | MySQL client for metric config; read-only queries |
openai | * | pip | 否 | DashScope embedding client; standard embedding API |
python-dotenv | * | pip | 否 | Optional .env loader; only for dev convenience |
安全亮点
✓ No shell execution, subprocess, or eval patterns found
✓ No credential harvesting or environment variable iteration for sensitive keys
✓ No access to ~/.ssh, ~/.aws, .env, or other sensitive paths
✓ No base64-encoded payloads or obfuscation techniques
✓ No hidden functionality — all external service calls are declared in SKILL.md
✓ External service injection uses proper Protocol interfaces (not hardcoded connections)
✓ JWT token (BI-*) is a hardcoded default fallback, not dynamically stolen
✓ Dependency packages (httpx, pymilvus, pymysql, openai) are standard and appropriate
✓ No cron jobs, startup hooks, or persistence mechanisms
✓ No prompt injection or jailbreak instructions
✓ Graceful degradation when optional services fail