jingni-trader

name	jingni-trader
version	1.0.0
description	A股量化交易全流程主调度器。负责解析用户意图，管理投研阶段状态机，维护跨 Skill 的上下文对象，按流程依次调度七个子 Skill 完成从数据采集到绩效报告的全链路工作。本身不执行任何量化计算，只做编排。
author	quant-team
license	MIT
tags	["quant-trading","A股","master-skill","workflow","量化","调度器"]
dependencies	["importlib (Python 标准库)","logging (Python 标准库)","json (Python 标准库)"]
environment_variables	[{"name":"TUSHARE_TOKEN","description":"Tushare Pro API Token（启用 tushare 数据源时需要；tushare 是 opt-in 源，默认不参与降级链）","required":false},{"name":"GM_TOKEN","description":"掘金量化API Token，用于实盘交易","required":false},{"name":"IFIND_USERNAME","description":"同花顺 iFinD 登录账号（启用 ifind 数据源时需要）","required":false},{"name":"IFIND_PASSWORD","description":"同花顺 iFinD 登录密码（启用 ifind 数据源时需要）","required":false},{"name":"JINGNI_URL","description":"惊泥因子库服务地址（启用 jingni-datafeed 因子库时需要）","required":false},{"name":"JINGNI_TOKEN","description":"惊泥因子库 API Token（启用 jingni-datafeed 因子库时需要）","required":false},{"name":"DATA_BACKENDS","description":"数据源优先级链，逗号分隔（如 \"tushare,baostock,akshare,websearch\"）。默认 \"baostock,akshare,websearch\"（仅真正免费源）。用户对话指定时优先级高于此变量","required":false,"default":"baostock,akshare,websearch"},{"name":"QUANT_WORK_DIR","description":"数据和工作目录","required":false,"default":"./workspace"},{"name":"QUANT_FORCE_REFRESH","description":"强制刷新所有阶段，忽略缓存产物（设为 \"1\" 启用）","required":false,"default":"0"},{"name":"FACTOR_BACKEND","description":"因子计算后端（pandas_ta / talib），默认 pandas_ta（纯 Python，无需安装 C 依赖）","required":false,"default":"pandas_ta"},{"name":"BACKTEST_BACKEND","description":"回测引擎后端（native / rqalpha / backtrader / gm），默认 native","required":false,"default":"native"},{"name":"ALLOW_SYNTHETIC_FALLBACK","description":"全部数据源失败时是否生成模拟数据兜底（默认 true）","required":false,"default":"true"},{"name":"AUTO_INSTALL_BACKENDS","description":"数据源依赖缺失时自动 pip install 后重试（默认 true）","required":false,"default":"true"},{"name":"LOG_LEVEL","description":"日志级别","required":false,"default":"INFO"}]
language	python
python_version	3.9+
entry_point	engine.py
allowed_sub_skills	["data-engine","factor-engine","strategy-model-engine","backtest-engine","portfolio-risk-engine","execution-monitor-engine","reports-engine"]
included_skills	["skills/data-engine","skills/factor-engine","skills/strategy-model-engine","skills/backtest-engine","skills/portfolio-risk-engine","skills/execution-monitor-engine","skills/reports-engine","skills/jingni-datafeed"]
trigger_keywords	["量化","回测","选股","因子","实盘","组合优化","A股","策略开发","分析","技术面","基本面","诊股"]

jingni-trader

概述

jingni-trader 是量化交易 Skill 套件的主协调中枢，负责：

1. 解析用户自然语言意图，判断当前投研阶段
2. 管理任务状态机，按流程调度子 Skill
3. 维护会话状态和任务上下文
4. 输出结构化的量化研究报告

意图解析与路由

系统采用单一工作流模型：根据用户是否明确需要构建可回测/可交易策略，选择执行深度。因子计算/IC 分析本身是两条路径共用的前置步骤，不构成"策略构建"意图。

项目用户意图 — 阶段/场景/报告/引擎流程/触发规则对照表

agent 运行本 skill 时，根据下表匹配用户意图 → 调度对应引擎 → 生成对应报告：

投研场景	用户意图表述	意图标识	阶段路径	触发的引擎流程（按序）	触发规则（关键词/条件）	最终报告产物
量化投研开发	因子分析 / IC分析	factor（隐式附加）	DATA → FACTOR → REPORT	data-engine → factor-engine → reports-engine（factor_analysis_report）	FACTOR 产物存在时伴随触发；不进门户、不阻断主报告	因子分析报告（附加产物）
量化投研开发	策略构建 / 回测验证 / 选股	strategy_required=True（隐式）	DATA → FACTOR → MODEL → BACKTEST → PORTFOLIO → EXECUTION → REPORT	data-engine → factor-engine → strategy-model-engine → backtest-engine → portfolio-risk-engine → execution-monitor-engine → reports-engine（backtest_report）	含"回测/策略/模型/选股/实盘/下单"等动作关键词	策略回测报告（净值曲线/夏普/回撤/行业归因）
量化投研开发	组合优化	portfolio	同上（策略管线完整跑，REPORT 阶段命中 portfolio_report）	… → portfolio-risk-engine → reports-engine（portfolio_report）	report_intent=portfolio	组合优化报告
量化投研开发	执行监控	execution	同上（策略管线完整跑，REPORT 阶段命中 execution_report）	… → execution-monitor-engine → reports-engine（execution_report）	report_intent=execution	执行监控报告（账户概览/持仓/成交/委托）
主观投研分析	个股技术面分析	technical	DATA → FACTOR → REPORT	data-engine → factor-engine → reports-engine（technical_report）	report_intent=technical 或 或含"技术面/K线/形态/指标"等关键词