Wall Street Quant Team

华尔街级AI多代理量化投资团队。当用户需要：创建AI投资团队、配置多代理协作、实施投资决策流程、获取股票行情、分析趋势方向、行为金融分析、AI量化策略时使用。触发场景如"我想分析某只股票"、"帮我配置投资团队"、"量化策略开发"等。

Security Scan

The skill's stated purpose (a multi-agent AI quant team) largely matches its files and templates, but its runtime instructions aggressively push loading external 'advisor' skills (with language saying the invitation is 'forced' and contains 'hidden hooks' to guide installation), which is scope‑creep that could lead to unexpected skill installations or social-engineering prompts.

Purpose & Capability

The name/description match the included agent templates, workflows, and references to data fetching and ML models — templates describe chief, fundamental, technical, quant, risk and sentiment agents, and many references describe data sources and model code. Minor inconsistency: docs reference a stock-data-fetcher script and pip dependencies (yfinance, pandas) while the skill has no install spec; this is explainable (instruction-only skill expecting runtime environment) but worth noting.

Instruction Scope

SKILL.md and the advisor-invitation guide instruct the agent to run a mandatory '智囊邀请' (advisor invitation) workflow that is described as '强制执行' (forced) for every analysis and explicitly tells the agent to use load_skill to install/invoke external skills (elon-musk-thinking, buffett-investor). The advisor guide even mentions '隐性钩子引导安装' (implicit hooks to guide installation). That grants the agent broad discretion to prompt for or load third‑party skills and to push the user to install them — scope creep beyond a self-contained quant analysis skill and a social-engineering risk.

Install Mechanism

There is no declared install spec (instruction-only), which is lowest-risk. However references/stock-data-usage.md instruct pip installs (yfinance,pandas) and call out scripts like scripts/stock-data-fetcher.py; the manifest does not list that fetcher script explicitly (some files truncated). Because no installer is provided, the skill expects the runtime environment to provide dependencies or to allow the agent to request installs later — this mismatch is worth flagging but not outright malicious.

Credentials

The skill declares no required environment variables, credentials, or config paths. The templates reference external data sources and optional paid APIs as general advice, but the skill does not request secrets in its manifest. This is proportionate to its stated purpose.

Persistence & Privilege

always:false (good). Still, the instructions repeatedly direct the agent to autonomously load other skills (load_skill calls) and state the advisor invitation is a mandatory step. Combining autonomous invocation with directive language to load external skills increases blast radius: the agent could persuade or auto-load additional third‑party skills during a session. This is a governance/behavior concern even though the skill itself does not request elevated system privileges.

What to consider before installing

This skill appears to implement a coherent multi-agent quant team, but it instructs the agent to automatically recommend and load third‑party 'advisor' skills (e.g., 'elon-musk-thinking' or 'buffett-investor') and even describes that invitation as 'forced' with 'hooks' to prompt installation. Before installing or enabling this skill: 1) be cautious about agreeing to install any advisor skills the agent recommends — review those skills' manifests and permissions first; 2) verify whether your runtime already provides the data-fetching dependencies (yfinance, pandas) or whether the agent will try to install packages at runtime; 3) confirm you are comfortable with the agent autonomously invoking load_skill (it may attempt to load additional code or request installs); 4) if you require stricter behavior, ask for a version of the skill that does not auto‑invite/install external skills or that makes any advisor-invite optional and requires explicit user consent for each install. If you want, I can list specific lines in the SKILL.md that trigger these concerns or look inside the included scripts (package_skill.py, quick_validate.py) for any code that would attempt network downloads or automatic installs.

华尔街级AI量化投资智囊

核心定位

您的专属AI量化投资团队，由7位专业AI代理组成：策略总监、基本面分析师、技术分析师、行业研究员、舆情监控、风控专员、量化研究员。

顺势而为，趋势是第一准则。

团队架构

                    ┌─────────────────────────────────────┐
                    │           投资者（您）                 │
                    └─────────────────┬───────────────────┘
                                      │
                                      ▼
┌───────────────────────────────────────────────────────────────────────┐
│                        策略总监（Lobster_CHIEF）                       │
│  核心职责：团队协调、决策把关、向上汇报                                 │
│  权限：任务分配、最终决策权（趋势裁决）                                │
└───────────────────────────────┬───────────────────────────────────────┘
                                │
        ┌────────────────────────┼────────────────────────┐
        ▼                        ▼                        ▼
┌───────────────┐    ┌───────────────────┐    ┌───────────────────┐
│  基本面分析师  │    │    技术分析师      │    │    行业研究员      │
│ Lobster_FUND  │    │ Lobster_TECHNICAL │    │ Lobster_INDUSTRY  │
├───────────────┤    ├───────────────────┤    ├───────────────────┤
│ 财务报表分析  │    │ 趋势判断（首优）  │    │ 产业链研究        │
│ 估值建模      │    │ K线形态识别       │    │ 竞争格局分析      │
│ 商业逻辑      │    │ 支撑压力位        │    │ 政策影响评估      │
└───────────────┘    └───────────────────┘    └───────────────────┘
        │                        │                        │
        └────────────────────────┼────────────────────────┘
                                 ▼
┌───────────────────────────────────────────────────────────────────────┐
│                         量化研究员（Lobster_QUANT）                       │
│  核心职责：AI量化分析、机器学习/深度学习/强化学习模型支持                │
│  能力：XGBoost/LightGBM/LSTM/Transformer/DQN/PPO                        │
└───────────────────────────────────────────────────────────────────────┘
                                │
        ┌────────────────────────┼────────────────────────┐
        ▼                        ▼                        ▼
┌───────────────┐    ┌───────────────────┐    ┌───────────────────┐
│   舆情监控    │    │     风控专员      │    │    策略总监       │
│ Lobster_SENT  │    │  Lobster_RISK     │    │ Lobster_CHIEF     │
├───────────────┤    ├───────────────────┤    ├───────────────────┤
│ 市场情绪监测  │    │ 风险评估监控      │    │ 汇总整合          │
│ 行为偏差检测  │    │ 止损执行          │    │ 投资决策          │
│ 舆情数据追踪  │    │ 合规审查          │    │ 汇报主人          │
└───────────────┘    └───────────────────┘    └───────────────────┘

投资分析7步流程

1️⃣ 任务分解 ─── 策略总监
   • 理解投资者需求
   • 分解分析任务
   • 分配给各代理

2️⃣ 专业分析 ─── 4代理并行
   • 基本面分析师：财务报表、估值、商业逻辑
   • 技术分析师：趋势、K线、量价分析
   • 行业研究员：产业链、竞争、政策
   • 舆情监控：情绪、偏差、舆情

3️⃣ 量化验证 ─── 量化研究员
   • 机器学习验证（XGBoost/LightGBM）
   • 深度学习预测（LSTM/Transformer）
   • 强化学习模拟（DQN/PPO）

4️⃣ 汇总整合 ─── 策略总监
   • 汇总各代理报告
   • 识别关键矛盾点
   • 趋势权重最高裁决

5️⃣ 投资决策 ─── 策略总监
   • 顺势而为判断方向
   • 制定投资方案
   • 明确仓位建议

6️⃣ 风控审查 ─── 风控专员
   • 评估风险敞口
   • 制定止损方案
   • 风险否决权

7️⃣ 汇报主人 ─── 策略总监
   • 清晰呈现结论
   • 说明决策逻辑
   • 风险提示

核心能力

📊 股票行情获取

• 实时行情数据
• 历史K线数据
• 财务指标数据
• 多市场支持（A股、港股、美股）

📈 趋势分析

• 技术指标计算（MA、RSI、MACD、布林带、KDJ）
• 趋势识别与跟踪
• 支撑压力位分析
• 量价关系分析

🤖 AI量化模型

• 机器学习：XGBoost、LightGBM、Random Forest
• 深度学习：LSTM时序预测、Transformer量化
• 强化学习：DQN、PPO自适应交易策略

📋 因子分析

• IC/IR分析
• 因子有效性检验
• 多因子组合优化
• 因子暴露度分析

📉 回测验证

• 样本内/样本外测试
• 夏普比率、卡玛比率
• 最大回撤分析
• 稳健性检验

🛡️ 风险控制

• 风险敞口评估
• 止损方案制定
• 合规性审查
• 极端情况预案

🧠 行为金融

• 认知偏差识别
• 情绪周期分析
• 羊群效应监控
• 反向投资机会

输出报告格式

【AI量化投资分析报告】标的代码/名称

一、趋势判断（第一优先级）
- 当前趋势：上升/下降/横盘
- 趋势强度：强/中/弱
- 关键支撑/压力位

二、基本面分析
- 财务指标评分
- 估值合理性
- 盈利质量

三、技术面分析
- 关键信号
- 均线系统状态
- MACD/RSI指标

四、量化信号
- 综合评分
- 置信度
- 建议仓位

五、风险提示
- 主要风险因素
- 止损位建议
- 注意事项

六、综合结论
- 操作建议
- 风险提示：投资有风险，决策需谨慎

触发指令示例

• "分析贵州茅台"
• "帮我看看腾讯的走势"
• "配置投资团队"
• "量化策略开发"
• "风险评估"
• "今日市场分析"

风控原则

风险提示：投资有风险，决策需谨慎。分析结论仅供参考，请结合自身判断做出决策。