Shuchang dev

.gitignore

@@ -22,4 +22,5 @@ build/*
 **/cache/**
 *.pdf
 *.csv
-site/*
+site/*
+tool_cache/*

extract_ths_tool_cache/compress.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+# 定义目标文件夹和输出文件名前缀
+TARGET_DIR="tool_cache"
+OUTPUT_PREFIX="extract_ths_tool_cache/tool_cache.tar.gz.part_"
+SIZE_LIMIT="50m"
+
+echo "正在压缩并切分 $TARGET_DIR ..."
+
+# tar -c: 创建
+# -z: gzip压缩
+# -f -: 将结果输出到标准输出流
+# split -b: 按大小切分
+# -: 从标准输入流读取
+tar -czf - "$TARGET_DIR" --exclude='.DS_Store' | split -b $SIZE_LIMIT - "$OUTPUT_PREFIX"
+
+echo "压缩完成！生成的切分文件如下："
+ls -lh ${OUTPUT_PREFIX}*

extract_ths_tool_cache/extract.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+
+# 定义切分文件的前缀
+TARGET_DIR="tool_cache"
+INPUT_PREFIX="extract_ths_tool_cache/tool_cache.tar.gz.part_"
+
+# 检查是否有分卷文件
+if ! ls ${INPUT_PREFIX}* 1> /dev/null 2>&1; then
+    echo "错误：未发现分卷文件 ${INPUT_PREFIX}*"
+    exit 1
+fi
+
+echo "正在合并并解压文件..."
+
+# cat: 合并所有分卷
+# tar -x: 解压
+cat ${INPUT_PREFIX}* | tar -xzf -
+
+echo "解压完成！文件夹 '$TARGET_DIR' 已还原。"

extract_ths_tool_cache/merge.py

@@ -0,0 +1,103 @@
+import os
+import json
+import shutil
+
+# 配置路径（Python 字符串不需要反斜杠转义空格）
+SRC_ROOT_BASE = "/Users/tsc/研究工作/金融ASIO/代码/finance-mcp_3"
+DST_ROOT_BASE = "/Users/tsc/研究工作/金融ASIO/代码/finance-mcp"
+
+# 需要合并的根文件夹
+TARGET_FOLDERS = ["tool_cache", "cache"]
+
+def load_json_data(file_path):
+    """加载 JSON 数据，支持列表或单个对象"""
+    try:
+        with open(file_path, 'r', encoding='utf-8') as f:
+            data = json.load(f)
+            return data if isinstance(data, list) else [data]
+    except Exception as e:
+        print(f"  [跳过] 无法解析 JSON: {file_path}. 错误: {e}")
+        return None
+
+def save_json_data(file_path, data):
+    """保存 JSON 数据"""
+    with open(file_path, 'w', encoding='utf-8') as f:
+        json.dump(data, f, ensure_ascii=False, indent=2)
+
+def get_code(sample):
+    """根据你提供的结构提取 code"""
+    try:
+        # 路径: sample -> tool_args -> code
+        return sample.get("tool_args", {}).get("code")
+    except:
+        return None
+
+def merge_json_files(src_file, dst_file):
+    """核心逻辑：合并两个文件中的 samples，按 code 去重"""
+    src_data = load_json_data(src_file)
+    dst_data = load_json_data(dst_file)
+
+    if src_data is None or dst_data is None:
+        return False # 读取出错，不执行合并
+
+    # 获取目标文件中已有的所有 code
+    existing_codes = {get_code(s) for s in dst_data if get_code(s) is not None}
+
+    initial_count = len(dst_data)
+    for sample in src_data:
+        code = get_code(sample)
+        if code and code not in existing_codes:
+            dst_data.append(sample)
+            existing_codes.add(code)
+
+    new_added = len(dst_data) - initial_count
+    if new_added > 0:
+        save_json_data(dst_file, dst_data)
+        print(f"  [合并完成] {os.path.basename(dst_file)}: 新增了 {new_added} 条 code 样本")
+    else:
+        print(f"  [无需合并] {os.path.basename(dst_file)}: 未发现新 code")
+    return True
+
+def start_recursive_merge():
+    for target in TARGET_FOLDERS:
+        src_folder = os.path.join(SRC_ROOT_BASE, target)
+        dst_folder = os.path.join(DST_ROOT_BASE, target)
+
+        if not os.path.exists(src_folder):
+            print(f"源文件夹不存在，跳过: {src_folder}")
+            continue
+
+        print(f"\n>>> 正在扫描目录: {target}")
+
+        # 使用 os.walk 递归遍历所有子文件夹
+        for root, dirs, files in os.walk(src_folder):
+            # 计算当前子目录相对于源根目录的路径
+            rel_path = os.path.relpath(root, src_folder)
+            # 对应的目标子目录路径
+            target_dst_dir = os.path.join(dst_folder, rel_path)
+
+            # 1. 如果目标子目录不存在，直接创建
+            if not os.path.exists(target_dst_dir):
+                os.makedirs(target_dst_dir)
+                print(f"创建新目录: {target_dst_dir}")
+
+            # 2. 处理当前目录下的所有文件
+            for filename in files:
+                # 隐藏文件跳过 (如 .DS_Store)
+                if filename.startswith('.'): continue
+
+                src_file_path = os.path.join(root, filename)
+                dst_file_path = os.path.join(target_dst_dir, filename)
+
+                if not os.path.exists(dst_file_path):
+                    # 如果目标位置没有这个文件，直接整体复制
+                    shutil.copy2(src_file_path, dst_file_path)
+                    print(f"  [新文件] 已复制: {rel_path}/{filename}")
+                else:
+                    # 如果目标位置有重名文件，执行深度合并
+                    merge_json_files(src_file_path, dst_file_path)
+
+if __name__ == "__main__":
+    print("开始深度合并任务...")
+    start_recursive_merge()
+    print("\n所有任务已结束。")

finance_mcp/config/ths_local.yaml

@@ -0,0 +1,156 @@
+flow:
+  crawl_ths_company:
+    flow_content: |
+      ReadLocalThsOp(tag="company")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取公司资料信息，例如：详细情况，高管介绍，发行相关，参控股公司，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_holder:
+    flow_content: |
+      ReadLocalThsOp(tag="holder")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取股东研究信息，例如：股东人数、十大流通股东、十大股东、十大债券持有人、控股层级关系，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_operate:
+    flow_content: |
+      ReadLocalThsOp(tag="operate")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取经营分析信息，例如：主营介绍、运营业务数据、主营构成分析、主要客户及供应商、董事会经营评述、产品价格，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_equity:
+    flow_content: |
+      ReadLocalThsOp(tag="equity")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取股本结构信息，例如：解禁时间表、总股本构成、A股结构图、历次股本变动，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_capital:
+    flow_content: |
+      ReadLocalThsOp(tag="capital")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取资本运作信息，例如：募集资金来源、项目投资、收购兼并、股权投资、参股IPO、股权转让、关联交易、质押解冻，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_worth:
+    flow_content: |
+      ReadLocalThsOp(tag="worth")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取盈利预测信息，例如：业绩预测、业绩预测详表、研报评级，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_news:
+    flow_content: |
+      ReadLocalThsOp(tag="news")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取新闻公告信息，例如：新闻与股价联动、公告列表、热点新闻列表、研报列表，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_concept:
+    flow_content: |
+      ReadLocalThsOp(tag="concept")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取概念题材信息，例如：常规概念、其他概念、题材要点、概念对比，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_position:
+    flow_content: |
+      ReadLocalThsOp(tag="position")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取主力持仓信息，例如：机构持股汇总、机构持股明细、被举牌情况、IPO获配机构，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_finance:
+    flow_content: |
+      ReadLocalThsOp(tag="finance")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取财务分析信息，例如：财务诊断、财务指标、指标变动说明、资产负债构成、财务报告、杜邦分析，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_bonus:
+    flow_content: |
+      ReadLocalThsOp(tag="bonus")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取分红融资信息，例如：分红诊断、分红情况、增发机构获配明细、增发概况、配股概况，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_event:
+    flow_content: |
+      ReadLocalThsOp(tag="event")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取公司大事信息，例如：高管持股变动、股东持股变动、担保明细、违规处理、机构调研、投资者互动，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true
+
+  crawl_ths_field:
+    flow_content: |
+      ReadLocalThsOp(tag="field")
+    enable_cache: false
+    cache_expire_hours: 1
+    description: "通过A股股票代码获取行业对比信息，例如：行业地位、行业新闻，最后返回和query相关的信息。"
+    input_schema:
+      code:
+        type: string
+        description: "stock code"
+        required: true

finance_mcp/core/agent/conduct_research_op.py

@@ -74,7 +74,7 @@ def build_tool_call(self) -> ToolCall:
         )
 
     async def async_execute(self):
-        """Run the multi-step research loop and produce a final answer.
+        """Run the multistep research loop and produce a final answer.
 
         The method performs the following high-level steps:
 

finance_mcp/core/crawl/__init__.py

@@ -13,10 +13,12 @@
 """
 
 from .crawl4ai_op import Crawl4aiOp, Crawl4aiLongTextOp
+from .read_local_ths_op import ReadLocalThsOp
 from .ths_url_op import ThsUrlOp
 
 __all__ = [
     "Crawl4aiOp",
     "Crawl4aiLongTextOp",
     "ThsUrlOp",
+    "ReadLocalThsOp",
 ]

finance_mcp/core/crawl/read_local_ths_op.py

@@ -0,0 +1,64 @@
+import json
+from pathlib import Path
+from typing import Dict
+
+from flowllm.core.context import C
+from flowllm.core.op import BaseAsyncOp
+from loguru import logger
+
+
+@C.register_op()
+class ReadLocalThsOp(BaseAsyncOp):
+    # Class-level cache: {tag: {code: tool_result}}
+    _cache: Dict[str, Dict[str, str]] = None
+
+    def __init__(self, tag: str = "", **kwargs):
+        super().__init__(**kwargs)
+        self.tag: str = tag
+        # Initialize class-level cache if not exists
+        if ReadLocalThsOp._cache is None:
+            ReadLocalThsOp._cache = {}
+
+    def _load_cache(self) -> Dict[str, str]:
+        """Load all crawl_ths_{tag}*.json files and build code->tool_result mapping."""
+        tool_cache_dir = Path("tool_cache")
+        pattern = f"crawl_ths_{self.tag}*.json"
+        matching_files = list(tool_cache_dir.glob(pattern))
+
+        total_files = len(matching_files)
+        logger.info(f"Found {total_files} files matching pattern '{pattern}'")
+
+        result_dict = {}
+        for idx, file_path in enumerate(matching_files, 1):
+            logger.info(f"Loading file [{idx}/{total_files}]: {file_path.name}")
+
+            with open(file_path, 'r', encoding='utf-8') as f:
+                data = json.load(f)
+
+            file_records = 0
+            for item in data:
+                code = item['tool_args']['code']
+                result_dict[code] = item['tool_result']
+                file_records += 1
+
+            logger.info(f"  → Processed {file_records} records from {file_path.name}")
+
+        logger.info(f"✓ Successfully loaded {len(result_dict)} total records for tag={self.tag}")
+        return result_dict
+
+    async def async_execute(self):
+        """Read tool_result for self.context.code from cached data."""
+        # Load cache if not exists
+        if self.tag not in ReadLocalThsOp._cache:
+            ReadLocalThsOp._cache[self.tag] = self._load_cache()
+
+        # Get code from context
+        code = self.context.code
+        if not code:
+            self.context.response.answer = f"No code={code} found in context."
+            logger.info(self.context.response.answer)
+            return
+
+        # Get tool_result from cache
+        tool_result = ReadLocalThsOp._cache[self.tag].get(code)
+        self.context.response.answer = tool_result