financial-report-analyzer

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

财报分析技能 (Financial Report Analyzer)

Financial Report Analyzer Skill

本技能旨在帮助 DB-GPT 系统化地分析上市公司财报，通过提取核心数据、计算财务比率、生成可视化图表并结合业务背景，产出高质量的财务分析报告。

This skill is designed to help DB-GPT systematically analyze the financial reports of listed companies. By extracting core data, calculating financial ratios, generating visual charts and combining with business background, it produces high-quality financial analysis reports.

核心工作流程

Core Workflow

数据提取与结构化：
- 使用
```
execute_skill_script_file
```
  工具执行
```
scripts/extract_financials.py
```
  脚本，传入财报文件路径（
```
file_path
```
  参数），自动提取营收、净利润、资产、负债等核心数值。
- 脚本支持 PDF 文件（通过 pdfplumber 解析）和纯文本文件，返回 JSON 格式的结构化数据。
财务比率计算：
- 使用
```
execute_skill_script_file
```
  执行
```
scripts/calculate_ratios.py
```
  ，传入 Step 1 的 JSON 数据。
- 自动计算毛利率、净利率、ROE、资产负债率等关键指标，输出 30 个模板占位符键值。
- 参考
```
references/financial_metrics.md
```
  确保指标定义的准确性。
- 系统会自动保存返回的 JSON 结果（
```
react_state["ratio_data"]
```
  ），后续 html_interpreter 会自动合并。
图表生成：
- 使用
```
execute_skill_script_file
```
  执行
```
scripts/generate_charts.py
```
  ，传入 Step 1 的 JSON 数据。
- 自动生成 3 张可视化图表：
  - ```
  financial_overview.png
```
  ：核心财务指标对比柱状图
- ```
profitability.png
```
    ：盈利能力指标横向条形图
  - ```
  asset_structure.png
```
  ：资产结构环形饼图
- 系统会自动将图片复制到静态目录并记录 URL 映射（
```
react_state["image_url_map"]
```
  ），后续 html_interpreter 会自动合并。
深度分析：
- 遵循
```
references/analysis_framework.md
```
  提供的框架，从盈利质量、偿债风险、营运效率和现金流四个维度进行深度剖析。
- 结合"经营情况讨论与分析"章节，解释业绩变动的核心驱动因素。
- 撰写以下 7 段分析文本：
  - ```
  PROFITABILITY_ANALYSIS
```
  ：盈利能力分析
- ```
SOLVENCY_ANALYSIS
```
    ：偿债与风险分析
  - ```
  EFFICIENCY_ANALYSIS
```
  ：营运效率分析
- ```
CASHFLOW_ANALYSIS
```
    ：现金流与利润质量分析
  - ```
  ADVANTAGES_LIST
```
  ：核心优势列表（HTML
```
  <li>
```
  格式）
- ```
RISKS_LIST
```
    ：主要风险列表（HTML
```
<li>
```
    格式）
  - ```
  OVERALL_ASSESSMENT
```
  ：综合评价

渲染报告：

调用

html_interpreter

，使用

template_path

模式：

json

{
  "template_path": "financial-report-analyzer/templates/report_template.html",
  "data": {
    "PROFITABILITY_ANALYSIS": "LLM撰写的盈利能力分析...",
    "SOLVENCY_ANALYSIS": "LLM撰写的偿债分析...",
    "EFFICIENCY_ANALYSIS": "LLM撰写的营运效率分析...",
    "CASHFLOW_ANALYSIS": "LLM撰写的现金流分析...",
    "ADVANTAGES_LIST": "<li>优势1</li><li>优势2</li>",
    "RISKS_LIST": "<li>风险1</li><li>风险2</li>",
    "OVERALL_ASSESSMENT": "LLM撰写的综合评价..."
  },
  "title": "XX公司 2023年度财报分析报告"
}

重要：
```
data
```
字典中只需传入你撰写的 7 段分析文本！后端会自动合并：
- Step 2 的 30 个数据指标（COMPANY_NAME、REVENUE、NET_PROFIT 等）
- Step 3 的图表 URL（CHART_FINANCIAL_OVERVIEW、CHART_PROFITABILITY、CHART_ASSET_STRUCTURE）
绝对不要在
```
data
```
中包含数据指标或图表路径，否则 JSON 过大会导致截断。

完成：
- 调用
```
terminate
```
  返回 1-2 句话的简短摘要。
- 报告会以卡片形式展示在左侧面板，用户点击卡片即可在右侧面板查看完整报告。

Data Extraction and Structuring:
- Use the
```
execute_skill_script_file
```
  tool to run the
```
scripts/extract_financials.py
```
  script, passing in the financial report file path (the
```
file_path
```
  parameter), to automatically extract core values such as revenue, net profit, assets, liabilities, etc.
- The script supports PDF files (parsed via pdfplumber) and plain text files, returning structured data in JSON format.
Financial Ratio Calculation:
- Use
```
execute_skill_script_file
```
  to run
```
scripts/calculate_ratios.py
```
  , passing in the JSON data from Step 1.
- Automatically calculate key indicators such as gross profit margin, net profit margin, ROE, asset-liability ratio, etc., outputting 30 template placeholder key-value pairs.
- Refer to
```
references/financial_metrics.md
```
  to ensure the accuracy of indicator definitions.
- The system will automatically save the returned JSON result (
```
react_state["ratio_data"]
```
  ), which will be automatically merged by the html_interpreter later.
Chart Generation:
- Use
```
execute_skill_script_file
```
  to run
```
scripts/generate_charts.py
```
  , passing in the JSON data from Step 1.
- Automatically generate 3 visual charts:
  - ```
  financial_overview.png
```
  : Bar chart comparing core financial indicators
- ```
profitability.png
```
    : Horizontal bar chart of profitability indicators
  - ```
  asset_structure.png
```
  : Donut chart of asset structure
- The system will automatically copy the images to the static directory and record the URL mapping (
```
react_state["image_url_map"]
```
  ), which will be automatically merged by the html_interpreter later.
In-depth Analysis:
- Follow the framework provided in
```
references/analysis_framework.md
```
  to conduct in-depth analysis from four dimensions: profit quality, solvency risk, operational efficiency and cash flow.
- Combine the "Management Discussion and Analysis" section to explain the core driving factors of performance changes.
- Write the following 7 sections of analysis text:
  - ```
  PROFITABILITY_ANALYSIS
```
  : Profitability Analysis
- ```
SOLVENCY_ANALYSIS
```
    : Solvency and Risk Analysis
  - ```
  EFFICIENCY_ANALYSIS
```
  : Operational Efficiency Analysis
- ```
CASHFLOW_ANALYSIS
```
    : Cash Flow and Profit Quality Analysis
  - ```
  ADVANTAGES_LIST
```
  : List of Core Advantages (in HTML
```
  <li>
```
  format)
- ```
RISKS_LIST
```
    : List of Main Risks (in HTML
```
<li>
```
    format)
  - ```
  OVERALL_ASSESSMENT
```
  : Comprehensive Assessment

Report Rendering:

Call

html_interpreter

using the

template_path

mode:

json

{
  "template_path": "financial-report-analyzer/templates/report_template.html",
  "data": {
    "PROFITABILITY_ANALYSIS": "Profitability analysis written by LLM...",
    "SOLVENCY_ANALYSIS": "Solvency analysis written by LLM...",
    "EFFICIENCY_ANALYSIS": "Operational efficiency analysis written by LLM...",
    "CASHFLOW_ANALYSIS": "Cash flow analysis written by LLM...",
    "ADVANTAGES_LIST": "<li>Advantage 1</li><li>Advantage 2</li>",
    "RISKS_LIST": "<li>Risk 1</li><li>Risk 2</li>",
    "OVERALL_ASSESSMENT": "Comprehensive assessment written by LLM..."
  },
  "title": "XX Company 2023 Annual Financial Report Analysis"
}

Important: Only pass the 7 sections of analysis text in the
```
data
```
dictionary! The backend will automatically merge:
- 30 data indicators from Step 2 (COMPANY_NAME, REVENUE, NET_PROFIT, etc.)
- Chart URLs from Step 3 (CHART_FINANCIAL_OVERVIEW, CHART_PROFITABILITY, CHART_ASSET_STRUCTURE)
Never include data indicators or chart paths in
```
data
```
, otherwise the JSON will be too large and cause truncation.

Completion:
- Call
```
terminate
```
  to return a short summary of 1-2 sentences.
- The report will be displayed as a card in the left panel, and users can click the card to view the full report in the right panel.

完整流程示例

Complete Workflow Example

Step 1: execute_skill_script_file(skill_name="financial-report-analyzer", script_file_name="extract_financials.py", args={"file_path": "/path/to/report.pdf"})
  → 返回 JSON: {"revenue": 10500000000, "net_profit": 1200000000, ...}  （记为 raw_data）

Step 2: execute_skill_script_file(skill_name="financial-report-analyzer", script_file_name="calculate_ratios.py", args=<raw_data>)
  → 返回 30 个模板键值，系统自动记录到 react_state["ratio_data"]

Step 3: execute_skill_script_file(skill_name="financial-report-analyzer", script_file_name="generate_charts.py", args=<raw_data>)
  → 生成图表，系统自动复制到 /images/ 并记录 URL 映射

Step 4: （LLM 自行撰写 7 段深度分析文本）

Step 5: html_interpreter(template_path="financial-report-analyzer/templates/report_template.html", data={仅包含 7 段分析文本}, title="报告标题")
  → 后端自动合并数据指标 + 图表 URL + 分析文本，渲染完整报告

Step 6: terminate(result="简短摘要")

Step 1: execute_skill_script_file(skill_name="financial-report-analyzer", script_file_name="extract_financials.py", args={"file_path": "/path/to/report.pdf"})
  → Returns JSON: {"revenue": 10500000000, "net_profit": 1200000000, ...}  (recorded as raw_data)

Step 2: execute_skill_script_file(skill_name="financial-report-analyzer", script_file_name="calculate_ratios.py", args=<raw_data>)
  → Returns 30 template key-value pairs, which are automatically recorded to react_state["ratio_data"] by the system

Step 3: execute_skill_script_file(skill_name="financial-report-analyzer", script_file_name="generate_charts.py", args=<raw_data>)
  → Generates charts, the system automatically copies them to /images/ and records the URL mapping

Step 4: (LLM writes 7 sections of in-depth analysis text on its own)

Step 5: html_interpreter(template_path="financial-report-analyzer/templates/report_template.html", data={contains only 7 sections of analysis text}, title="Report Title")
  → Backend automatically merges data indicators + chart URLs + analysis text to render the complete report

Step 6: terminate(result="Short summary")

资源使用说明

Resource Usage Instructions

脚本（均通过
```
execute_skill_script_file
```
执行）：
- ```
scripts/extract_financials.py
```
  ：接收
```
file_path
```
  参数，读取财报文件（支持 PDF 和文本格式），提取核心财务数据。
- ```
scripts/calculate_ratios.py
```
  ：计算财务比率，输出 30 个模板占位符键值。系统自动记录结果。
- ```
scripts/generate_charts.py
```
  ：生成 3 张可视化图表（matplotlib），系统自动处理图片复制。
- ```
scripts/fill_template.py
```
  ：（备用）接收
```
ratio_data
```
  、
```
chart_paths
```
  、
```
analysis
```
  三个参数，读取 HTML 模板并替换所有占位符。正常情况下不需要使用此脚本，因为 html_interpreter 的 template_path 模式会自动完成模板填充。

参考：

```
references/financial_metrics.md
```
：包含公式定义。
```
references/analysis_framework.md
```
：包含分析逻辑。

模板：
- ```
templates/report_template.html
```
  ：最终交付报告的 HTML 模板（必须严格遵循，不得删减章节或修改表格结构）。由 html_interpreter 的 template_path 参数自动读取并填充。
- ```
templates/report_template.md
```
  ：Markdown 版本，仅供参考结构说明。

Scripts (all executed via
```
execute_skill_script_file
```
):
- ```
scripts/extract_financials.py
```
  : Accepts the
```
file_path
```
  parameter, reads financial report files (supports PDF and text formats), and extracts core financial data.
- ```
scripts/calculate_ratios.py
```
  : Calculates financial ratios and outputs 30 template placeholder key-value pairs. The system automatically records the results.
- ```
scripts/generate_charts.py
```
  : Generates 3 visual charts (using matplotlib), and the system automatically handles image copying.
- ```
scripts/fill_template.py
```
  : (Backup) Accepts three parameters
```
ratio_data
```
  ,
```
chart_paths
```
  ,
```
analysis
```
  , reads the HTML template and replaces all placeholders. Normally, this script does not need to be used because the template_path mode of html_interpreter will automatically complete template filling.
References:
- ```
references/financial_metrics.md
```
  : Contains formula definitions.
- ```
references/analysis_framework.md
```
  : Contains analysis logic.
Templates:
- ```
templates/report_template.html
```
  : HTML template for the final delivered report (must be strictly followed, do not delete sections or modify table structures). It is automatically read and filled by the template_path parameter of html_interpreter.
- ```
templates/report_template.md
```
  : Markdown version, for structural reference only.

注意事项

Notes

必须使用
execute_skill_script_file
执行脚本（不要用 shell_interpreter），因为
```
execute_skill_script_file
```
会自动处理图片复制和数据记录。
脚本提取可能受排版影响，建议在计算前人工核对提取的关键数值。
始终关注"非经常性损益"，以评估公司核心业务的真实盈利能力。
对比至少三年的历史数据，以识别趋势。
```
generate_charts.py
```
依赖 matplotlib，请确保环境中已安装该库。

Must use
execute_skill_script_file
to run scripts (do not use shell_interpreter), because
```
execute_skill_script_file
```
will automatically handle image copying and data recording.
Script extraction may be affected by typesetting, so it is recommended to manually verify the extracted key values before calculation.
Always pay attention to "non-recurring gains and losses" to evaluate the real profitability of the company's core business.
Compare at least three years of historical data to identify trends.
```
generate_charts.py
```
depends on matplotlib, please ensure that this library is installed in the environment.