QMT Development

QMT平台开发专用技能。当用户说"QMT"、"实盘"、"QMT策略"、"QMT回测"、"QMT开发"时自动触发。提供QMT API规范、核心函数机制、止损监控方式、实盘注意事项。适用于使用QMT平台的量化策略开发。

Security Scan

This is an instruction-only QMT development guide; its requested resources and instructions are consistent with the stated purpose and it does not ask for unexpected credentials or install code.

Purpose & Capability

The skill is a documentation/instruction-only skill for developing strategies on the QMT platform. It does not request unrelated environment variables, binaries, or installs, and the examples (order calls, stop-loss checks, reading local signal files) match the declared purpose.

Instruction Scope

SKILL.md contains concrete code templates that read local files (e.g., trading_signals/alphagbm_signals_qmt.json) and demonstrate calling trading functions (passorder, get_trade_detail_data). Those are appropriate for a QMT dev guide, but they imply the runtime agent or user environment will have access to local project files and platform APIs; users should be aware the examples assume access to trading account identifiers and platform APIs.

Install Mechanism

No install spec and no code files are included, so nothing will be downloaded or written to disk by the skill itself—this is the lowest-risk install profile.

Credentials

The skill does not request environment variables or credentials. However, its code samples use accID and platform order functions; if the agent environment supplies trading credentials, the agent could interact with live accounts. That is expected for a trading dev guide but worth noting as a security consideration.

Persistence & Privilege

always is false and there is no persistent install; the skill does not request elevated persistence or modify other skills or system-wide settings.

Assessment

This skill is a documentation-only QMT development guide and appears coherent with its purpose. Before enabling it in an agent that can act autonomously, confirm whether the agent runtime has access to any trading account credentials or local signal files — if so, the agent could place live orders if given permission. If you only want help drafting or reviewing code, keep the agent's ability to call external trading APIs or access account credentials disabled; if you intend the agent to execute trades, limit it to a test/sandbox account and review the code templates and waiting_list/anti-duplication logic carefully.

QMT平台开发技能

版本：1.0.0 适用平台：迅投QMT量化交易平台

---

🎯 核心函数机制

1. QMT标准函数

函数	用途	触发时机	说明
init()	初始化	策略启动时	加载信号、设置参数、设置股票池
handlebar()	K线处理	每根K线	主要逻辑，买入/卖出/止损

---

2. ⚠️ 重要：is_last_bar() 机制

场景	is_last_bar() 返回值	止损检查
回测	每根K线都返回True	✅ 每天检查
实盘	只在特定时刻返回True	❌ 盘中可能不检查

关键理解：

• 回测时，每根K线都是"最后一根"，所以is_last_bar()都返回True
• 实盘时，is_last_bar()只在特定时刻返回True（可能是收盘时）

结论：

• ❌ 不要用is_last_bar()控制止损/止盈检查
• ✅ 用时间判断控制检查频率

---

📋 止损监控方式

方式1：时间判断（推荐）

def handlebar(ContextInfo):
    """每根K线都会触发"""
    
    # 止损监控（每5分钟检查一次）
    now = datetime.now()
    if G.last_check_time:
        elapsed = (now - G.last_check_time).total_seconds() / 60
        if elapsed >= 5:  # 每5分钟检查
            check_stop_loss(ContextInfo)
            G.last_check_time = now
    
    # 买入逻辑（可以保留is_last_bar()限制）
    if not ContextInfo.is_last_bar():
        return
    
    # 买入逻辑...

方式2：bar_count判断

def handlebar(ContextInfo):
    """每根K线都会触发"""
    
    # 每5根K线检查一次止损
    if ContextInfo.bar_count % 5 == 0:
        check_stop_loss(ContextInfo)

---

🚀 代码模板

完整策略模板

# 全局状态对象
class GlobalState:
    pass
G = GlobalState()

def init(ContextInfo):
    """初始化"""
    # 全局状态
    G.waiting_list = []
    G.sent_orders = set()
    G.buy_prices = {}  # 记录买入价格
    G.last_check_time = None
    
    # 加载信号
    with open('trading_signals/alphagbm_signals_qmt.json') as f:
        data = json.load(f)
    G.signals = [s['code'] for s in data['signals']]
    
    # 设置股票池
    ContextInfo.set_universe(G.signals)

def handlebar(ContextInfo):
    """每根K线触发"""
    
    # 1. 止损监控（每5分钟）
    now = datetime.now()
    if G.last_check_time:
        elapsed = (now - G.last_check_time).total_seconds() / 60
        if elapsed >= 5:
            check_stop_loss(ContextInfo)
            G.last_check_time = now
    
    # 2. 检查waiting_list
    if check_waiting_orders(ContextInfo):
        return
    
    # 3. 买入逻辑（只在最后一根K线）
    if not ContextInfo.is_last_bar():
        return
    
    # 买入...

---

⚠️ 实盘注意事项

1. passorder的prType参数

# ❌ 错误
passorder(23, 1101, accID, code, -1, ...)  # prType=-1（价格不可控）

# ✅ 正确
passorder(23, 1101, accID, code, 14, ...)  # prType=14（对手价）

---

2. waiting_list防重机制

# 初始化
G.waiting_list = []
G.sent_orders = set()

# 买入时记录
if code not in G.sent_orders:
    passorder(...)
    G.sent_orders.add(code)
    G.waiting_list.append(msg)

# 检查waiting_list
def check_waiting_orders(ContextInfo):
    for msg in G.waiting_list[:]:
        order_result = get_trade_detail_data(...)
        if order_result:
            G.waiting_list.remove(msg)

---

3. 风控设置

QMT层面：

• 金额合规（限制资金上限）
• 反向单交易（禁止买入后立即卖出）
• 对敲单交易（禁止高买低卖）

代码层面：

• 硬编码资金限制
• 止损止盈逻辑
• 异常处理

---

🚫 禁止事项

1. ❌ 使用is_last_bar()控制止损检查
2. ❌ 使用prType=-1
3. ❌ 没有waiting_list防重机制
4. ❌ 没有止损止盈逻辑

---

📚 参考文档

• QMT官方示例（双均线实盘、竞价选股、行业轮动、多因子选股）
• QMT API文档
• QMT_STRATEGY_DEVELOPMENT_STANDARDS.md

---

技能版本：1.0.0