← 返回 Skills 市场
vbdspd

analyst-fundamentals

作者 vbdspd · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
82
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install analyst-fundamentals
功能描述
A股基本面深度分析工具。当用户提供股票代码或要求分析股票基本面时调用。
使用说明 (SKILL.md)

股票基本面分析 Skill(完整版)

定位:纯基本面分析工具,不做投资决策建议

分析哲学:宁可错过,不可做错。投资是认知的变现。


触发条件

立即调用本 Skill 当:

  • 用户提供股票代码(如 002594SH600519300032
  • 用户要求分析某只股票的基本面
  • 用户询问某公司的财务状况、经营情况

执行流程总览

用户输入股票代码
    ↓
┌─────────────────────────────────────────────────────────────┐
│ 阶段一:数据获取(16并行子Agent,一次性全量派发)              │
│ • sessions_spawn 启动16个子Agent(mode=run,并行执行)       │
│ • 每个子Agent独立browser,抓取1个F10模块写入对应md文件       │
│ • 全部完成后主Agent汇总                                      │
└─────────────────────────────────────────────────────────────┘
    ↓
┌─────────────────────────────────────────────────────────────┐
│ 阶段二:分析生成(6并行子Agent)                               │
│ • sessions_spawn 一次性启动6个子Agent(mode=run)            │
│ • 每个Agent读取2-5个原始数据文件                              │
│ • 按投资哲学框架输出结构化分析报告                            │
│ • B1-B6并行执行,完成后主Agent汇总                           │
└─────────────────────────────────────────────────────────────┘
    ↓
┌─────────────────────────────────────────────────────────────┐
│ 阶段三:报告汇总(5并行子Agent + 主Agent汇总)                  │
│ • sessions_spawn 一次性启动5个子Agent(mode=run)            │
│ • C1-C5并行:三大核心问题/五法验证/风险提示/画像优劣势/跟踪指标│
│ • 主Agent读取11个分析文件,合成最终report.md                  │
└─────────────────────────────────────────────────────────────┘

阶段一详解

16个F10模块并行抓取

# 模块 Hash 文件
1 公司概况 gsgk gsgk.md
2 经营分析 jyfx jyfx.md
3 财务分析 cwfx cwfx.md
4 核心题材 hxtc hxtc.md
5 操盘必读 cpbd cpbd.md
6 同行比较 thbj thbj.md
7 股东研究 gdyj gdyj.md
8 分红融资 fhfx fhfx.md
9 公司高管 gsgg gsgg.md
10 资金流向 zjlx zjlx.md
11 资本运作 zbyz zbyz.md
12 公司大事 gsds gsds.md
13 资讯公告 zxgg zxgg.md
14 股本结构 gbjg gbjg.md
15 关联个股 glgg glgg.md
16 龙虎榜单 lhbd lhbd.md

子Agent任务消息格式

每个子Agent接收以下格式的任务:

请抓取以下F10数据模块:

股票代码:{代码}
市场:SZ(深圳以00/3开头)/ SH(上海以6开头)
URL:https://emweb.securities.eastmoney.com/pc_hsf10/pages/index.html?type=web&code={市场}{代码}&color=b#/{hash}
输出文件:C:\Users\pd\.openclaw\workspace-analysts\data\{代码}\{hash}.md

执行步骤:
1. browser.start()(如未启动)
2. browser.open(url)
3. browser.snapshot(compact=true)
4. 提炼关键数据,写入输出文件
5. 完成后回复"模块{hash}.md已完成"

主Agent执行步骤(阶段一)

1. mkdir -p data/{代码}
2. sessions_spawn 一次性启动16个子Agent(mode=run)
   - 每个Agent分配1个模块(hash)
   - runtime: "subagent"
   - runTimeoutSeconds: 120
3. 等待所有子Agent完成

阶段二详解

6个并行分析任务

任务 文件 输入数据 输出
B-1 快速画像 b1_quick_profile.md cpbd+gsgk quick_profile_analysis.md
B-2 行业护城河 b2_industry_moat.md gsgk+hxtc+thbj industry_moat_analysis.md
B-3 财务质量 b3_financial_quality.md cwfx+jyfx financial_quality_analysis.md
B-4 公司治理 b4_governance.md gsgg+gdyj+gbjg governance_analysis.md
B-5 估值资本 b5_valuation.md thbj+fhfx+zbyz valuation_analysis.md
B-6 资金情绪 b6_fund_market.md zjlx+lhbd+glgg+gsds+zxgg fund_market_analysis.md

sessions_spawn 执行方式(阶段二)

// 6个并行(一次性全量派发)
sessions_spawn × 6:
  task: "\x3C读取对应提示词文件bX_xxx.md内容>"
  runtime: "subagent"
  mode: "run"
  runTimeoutSeconds: 300

阶段三详解

5个并行汇总任务

任务 文件 核心输出
C-1 三大核心问题 c1_three_questions.md 三维度判断结论表
C-2 五大经验法则 c2_five_rules.md 五法验证表(✅/⚠️/❌)
C-3 风险提示 c3_risk_warnings.md P0/P1/P2/P3分级风险清单
C-4 画像优劣势 c4_summary_profile.md 公司画像+核心优势+主要关注点
C-5 跟踪指标 c5_tracking_metrics.md 关键指标+预警条件+跟踪频率

主Agent汇总

读取6个分析文件 + 5个汇总文件
→ 填充 report_synthesis.md 模板
→ 输出 data/{代码}/report.md

输出文件结构

data/{股票代码}/
├── gsgk.md ~ lhbd.md          # 16个原始数据
├── quick_profile_analysis.md     # B-1
├── industry_moat_analysis.md   # B-2
├── financial_quality_analysis.md # B-3
├── governance_analysis.md        # B-4
├── valuation_analysis.md        # B-5
├── fund_market_analysis.md      # B-6
├── c1_three_questions.md        # C-1
├── c2_five_rules.md            # C-2
├── c3_risk_warnings.md         # C-3
├── c4_summary_profile.md        # C-4
├── c5_tracking_metrics.md       # C-5
└── report.md                    # 最终报告

投资哲学框架(内置于每个提示词)

三大核心问题

问题 对应分析维度 判断标准
好生意? 行业护城河+财务质量 ROE>15%、毛利率>30%、现金流健康、护城河≥8分
好管理层? 公司治理 管理团队>4分、任职>5年、持股、高质押\x3C30%
好价格? 估值资本运作 PE分位\x3C50%、PEG\x3C1.5、股息率>3%

五大经验法则

法则1:ROE核心 → 连续3年>15%优秀
法则2:现金流 → 净现比>1利润质量好
法则3:管理层 → 无违规+专注主业+利益一致
法则4:估值 → PE分位\x3C50%合理,>70%高危
法则5:风险排查 → 存贷双高/现金流持续为负/质押>50% → 一票否决

一票否决项

□ 存贷双高
□ 经营现金流连续2年为负
□ 大股东质押>50%
□ 频繁关联交易
□ 审计非标
□ 商誉占净资产>30%
□ 毛利率远高于同行
□ 应收账款持续异常

sessions_spawn 参数说明

mode="run"(推荐用于一次性任务)

{
  "task": "任务描述",
  "runtime": "subagent",
  "mode": "run",
  "runTimeoutSeconds": 120
}

关键约束

  • mode="run"一次性任务,发完即结束,无需 thread 参数
  • mode="session"持久会话,需要 thread=true(当前webchat不支持)
  • 所有阶段均使用 mode="run",避免使用 mode="session"
安全使用建议
Key things to consider before installing: - Function vs requests: The skill does what it claims (scrapes EastMoney F10 and synthesizes reports), but review whether you want the agent to spawn many subagents and run browser snapshots — this can be resource-heavy. - Persistent workers: The docs contain conflicting instructions; phase1_guide describes launching persistent session workers with no timeout. Ask the publisher why persistent workers are needed and prefer mode="run" (one-shot) unless you explicitly want long‑lived workers. - Filesystem paths: The skill hard-codes a Windows path (C:\Users\pd\...). If your agent runs on Linux/macOS or under a different user, the skill may fail or write files to unexpected locations; request a configurable workspace path or that the skill use a relative workspace path provided by the platform. - Network and data exposure: The skill scrapes external web pages (EastMoney) using browser snapshots. Confirm your environment's network policies and whether snapshots may include any sensitive context. There are no explicit exfiltration endpoints in the instructions, but persistent subagents increase exposure risk. - Testing recommendations: Run the skill in a sandboxed/test agent with non-sensitive inputs first. Monitor spawned subagents, their lifetime, and the directory where files are written. If you don't want persistent sessions, require the skill be modified to use only mode="run" with bounded timeouts. - If you need higher assurance: request the publisher remove mode="session" examples, make the data path configurable, and document expected runtime capabilities (which browser/tooling it requires) and supported OSes.
功能分析
Type: OpenClaw Skill Name: analyst-fundamentals Version: 1.0.0 The 'analyst-fundamentals' skill bundle is a legitimate multi-agent workflow designed for deep fundamental analysis of A-share stocks. It orchestrates 16 sub-agents to scrape financial data from EastMoney (securities.eastmoney.com) and subsequently processes this data through various analytical frameworks (e.g., DuPont analysis, Porter's Five Forces). The skill operates within a defined local workspace (C:\Users\pd\.openclaw\workspace-analysts\) and uses standard OpenClaw tools for browser automation and file management. No indicators of data exfiltration, malicious command execution, or unauthorized persistence were found across the SKILL.md or the extensive prompt templates.
能力评估
Purpose & Capability
Name/description (A股基本面深度分析) match the instructions: the SKILL.md describes fetching F10 pages from EastMoney, extracting 16 modules, running 6 analysis tasks and 5 report-aggregation tasks. The required actions (browser.open, snapshot, parse, write markdown reports) are appropriate for that purpose. Minor mismatch: the skill writes to a Windows-specific path (C:\Users\pd\... ) while the registry metadata does not declare any OS restriction — this may fail or behave unexpectedly on non-Windows hosts.
Instruction Scope
Most instructions stay within the stated purpose (web-scrape F10, parse, analyze). However, there are two problematic instruction-level issues: (1) inconsistent guidance about sessions_spawn mode — the top-level flow emphasizes mode="run" (one-shot subagents), but phase1_guide explicitly shows launching four persistent workers with mode="session" and runTimeoutSeconds:0 (no timeout). Persistent session workers are broader in scope and can keep running/receiving work beyond a single invocation; this elevates lifecycle and privacy risks. (2) Hard-coded filesystem paths (C:\Users\pd\.openclaw\workspace-analysts\...) assume Windows and a particular user-home layout, which is brittle and may cause unintended reads/writes on hosts that do match that path.
Install Mechanism
Instruction-only skill with no install spec and no code files — lowest install risk. The skill relies on platform capabilities (browser API, sessions_spawn/subagent runtimes) but does not pull external binaries or archives.
Credentials
The skill declares no required environment variables, credentials, or config paths. All runtime activity described uses platform primitives (browser snapshot, subagents) and local file writes under the skill's workspace. There are no explicit requests for unrelated credentials or secrets.
Persistence & Privilege
Although registry flags (always=false) are normal, the instructions include an explicit pattern to start persistent session-mode workers (mode="session" with runTimeoutSeconds:0). Persistent subagent sessions can increase blast radius (long-lived background processes that could be repurposed or exfiltrate data). This is inconsistent with other parts of the document that recommend mode="run" for one-shot tasks. The presence of both patterns is a substantive behavioral inconsistency and a risk factor.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install analyst-fundamentals
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /analyst-fundamentals 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.0
- Initial release of analyst-fundamentals, an A股基本面深度分析工具。 - Supports automated, multi-stage stock data gathering and structured fundamental analysis given a stock code or basic analysis request. - Executes in three main stages: 16-module F10 data parallel scraping, 6 parallel fundamental analyses, and 5 report synthesis tasks, all orchestrated with sessions_spawn in run mode. - Outputs a comprehensive report using a standardized, investment-philosophy-driven framework. - Strictly for fundamental analysis, with no investment advice or recommendations.
元数据
Slug analyst-fundamentals
版本 1.0.0
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

analyst-fundamentals 是什么?

A股基本面深度分析工具。当用户提供股票代码或要求分析股票基本面时调用。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 82 次。

如何安装 analyst-fundamentals?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install analyst-fundamentals」即可一键安装,无需额外配置。

analyst-fundamentals 是免费的吗?

是的,analyst-fundamentals 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

analyst-fundamentals 支持哪些平台?

analyst-fundamentals 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 analyst-fundamentals?

由 vbdspd(@vbdspd)开发并维护,当前版本 v1.0.0。

💬 留言讨论