OpenSenseNova/SenseNova-Skills

# SN Code Video

A code video is a program `frame(t) → image`, sampled at `t = i / fps` and encoded. Almost every failure in this kind of work is one of four things: the film is never finished, the picture depends on wall-clock time, nobody looked at the frames, or the file does not match the brief. This skill prevents those four.

## Setup and paths

Use a separate project folder for the video source, `SPEC.md`, build files, and outputs. Resolve `SKILL_DIR` to the absolute directory containing this `SKILL.md`, whether installed in the agent's skills directory or used from this repository. Run project commands from the video project folder; invoke bundled scripts through `"$SKILL_DIR/scripts/..."` so they do not depend on the current working directory. Set `SKILL_DIR` again in each shell session if needed.

Requirements: Node.js 18+ for browser capture, Python 3, FFmpeg/ffprobe with libx264 and AAC, and fonts covering the requested language. Install Playwright (with Chromium) or Puppeteer in the **video project**; the capture helper resolves its packages there first. Python audio dependencies are listed in `requirements.txt`; edge-tts needs network access, and faster-whisper needs downloaded model weights for transcription checks. Install only the tools required by the chosen renderer and audio workflow. Setup and invocation examples: [English guide](../../docs/sn-code-video.md) / [中文指南](../../docs/sn-code-video_cn.md) in the source repository; these repository guides are not required when the skill folder is installed on its own.

Deliver the playable MP4, source code and reproduction commands, `SPEC.md`, `VERIFY.md`, and a contact sheet. Audio stems and timing data belong in the project when sound is used.

## 0. Read the craft guides you need

# Excel Data Analysis Workflow

End-to-end workflow for structured Excel analysis. Each step maps to a
capability sub-skill that can be loaded for detailed patterns.

## Workflow

### Step 1 — Count rows across all sheets (lightweight, no full load)

Count rows per sheet **without loading data into memory**. Use openpyxl
`read_only` mode — this works for any file size.

```python
import openpyxl, gc

# Skill Steps

Step1 读取文件并统计所有 sheet 的行数，汇总后打印总行数，用于判断数据规模是否需要启用大文件处理。
```python
import pandas as pd

file_path = "input_data.xlsx"

# 读取所有sheet并统计总行数
xls = pd.ExcelFile(file_path)
sheet_names = xls.sheet_names
print(f"Sheet列表: {sheet_names}")

total_rows = 0

# excel-conditional-comparison-and-large-file-processing

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 提取不同Sheet中特定维度（如“B1层”）的数值，并进行跨表逻辑对比。
```python
# 定义提取逻辑：定位目标行（如包含'B1'的行）并获取其关联的系数
def extract_target_value(df, target_label='B1', label_col_idx=0, offset_row=1, value_col_idx=2):
    """
    在指定列搜索标签，并返回其相对偏移位置的数值
    """
    extracted_values = []
    for idx, row in df.iterrows():
        if str(row.iloc[label_col_idx]).strip() == target_label:

# Outlier_Coloring

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 使用正则表达式提取限值，并结合上下文逻辑识别总传热系数超限的行。
```python
import re

exceed_rows = []
target_col = 0  # 假设特征列在第一列
value_col = 8   # 假设数值列在第九列

for i, row in df.iterrows():
    row_str = str(row.iloc[target_col]) if pd.notna(row.iloc[target_col]) else ""

## Skill Steps

> **Note**: This sub-skill covers one step of the Excel analysis workflow. For the full pipeline (file reading, row counting, large-file optimization, export), see the parent workflow SKILL.md.


Step1 读取文件并统计所有 sheet 的行数，汇总后打印总行数，用于判断是否需要大文件加速。
```python
import pandas as pd
import openpyxl

file_path = "input_data.xlsx"

# 获取所有sheet名称
wb = openpyxl.load_workbook(file_path, read_only=True)

Step1 提取并合并多个 Sheet 中的关键维度数据，进行数据清洗、类型转换及 Top-N 筛选。
```python
# 示例：合并两个 Sheet 的数据
# 读取 Sheet1 并清洗
df1 = pd.read_excel(file_path, sheet_name='Sheet1', header=None)
# 假设 group_col 在第0列，value_col 在第2列
data1 = df1.iloc[20:, [0, 2]].copy()
data1.columns = ['group_col', 'value_col_1']
data1['value_col_1'] = pd.to_numeric(data1['value_col_1'], errors='coerce')
data1['group_col'] = data1['group_col'].ffill() # 处理合并单元格产生的缺失

# 读取 Sheet2 并清洗
df2 = pd.read_excel(file_path, sheet_name='Sheet2', header=None)
data2 = df2.iloc[5:, [0, 1]].copy()

# Numeric_Extraction_and_Distribution_Analysis

## Skill Steps

Step1 从原始数据中提取目标列，清理无效和空值数据，并安全地将带单位的字符串转换为数值类型
```python
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

# 配置中英文字体，避免图表乱码
plt.rcParams['font.sans-serif'] = ['SimHei', 'DejaVu Sans']
plt.rcParams['axes.unicode_minus'] = False

# categorical-comparison-analysis

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 读取文件并统计所有 sheet 的总行数，评估是否需要进行大文件优化处理。
```python
import pandas as pd
from pandas import read_excel
from pathlib import Path

# 统计所有 sheet 的行数以决定处理策略
file_path = "input_data.xlsx"
sheet_names = pd.ExcelFile(file_path).sheet_names
total_rows = 0

Step1 对数据进行清洗与预处理，包括处理合并单元格、正则过滤以及分类映射。
```python
import re

# 1. 处理合并单元格：向前填充
target_col = 'category_column'
df[target_col] = df[target_col].ffill()

# 2. 正则清洗：去除无效字符或筛选特定格式
def clean_text(text):
    if pd.isna(text): return text
    return re.sub(r'[^\w\s]', '', str(text)).strip()

df[target_col] = df[target_col].apply(clean_text)

## Skill Steps

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 提取关键指标，进行物理量/指标的单位一致性验证计算，并对核心业务指标进行降序排列。
```python
# 1. 物理量/指标单位一致性验证与计算 (保留公式结构示例)
col_numerator = 'numerator_col'  # 示例：Mx (kN·m)
col_denominator = 'denominator_col' # 示例：Wx (cm³)
col_target = 'target_col' # 示例：sigma (MPa)

if col_numerator in data.columns and col_denominator in data.columns and col_target in data.columns:
    # 单位换算示例：统一到标准单位后计算
    data['den_converted'] = data[col_denominator] * 1e-6

Step1 对原始数据进行清洗与重构，处理 Excel 合并单元格导致的缺失值，并筛选核心分析列。
```python
import pandas as pd

def preprocess_pivot_data(file_path, target_cols=['奖项', '项目名称', '成员', '单位']):
    """
    清理并重构数据列，处理合并单元格填充。
    """
    df = pd.read_excel(file_path)
    # 映射通用列名
    df.columns = target_cols
    
    # 关键技巧：处理合并单元格。ffill 前需确保数据按原始分类顺序排列
    # 假设第一列为分类标签（如奖项名称）

## Skill Steps

Step1 加载并检查原始数据，配置中文字体以确保图表正常显示。
```python
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import warnings
warnings.filterwarnings('ignore')

# 设置中文字体，兼容不同操作系统
plt.rcParams['font.sans-serif'] = ['SimHei', 'WenQuanYi Zen Hei', 'DejaVu Sans']
plt.rcParams['axes.unicode_minus'] = False

Step1 加载数据并配置环境，设置中文字体以确保可视化图表正常显示。
```python
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings('ignore')

# 设置中文字体，优先使用 WenQuanYi Zen Hei，备选 SimHei 和 DejaVu Sans
plt.rcParams['font.sans-serif'] = ['WenQuanYi Zen Hei', 'SimHei', 'DejaVu Sans']
plt.rcParams['axes.unicode_minus'] = False

# 加载数据文件
file_path = 'your_data.xlsx'

# Excel_Multi_Sheet_Deduplication

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 加载目标数据表，并进行初步的数据预览与结构检查。
```python
import pandas as pd

file_path = 'input_file.xlsx'
target_sheet = 'Sheet1' # 根据实际情况指定 sheet 名称

# 读取数据，header=None 用于处理无表头或非标准表头文件
df = pd.read_excel(file_path, sheet_name=target_sheet, header=None)
print(f"数据形状: {df.shape}")

# Invalid_Data_Cleaning

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

## Skill Steps

Step1 根据总行数判断是否数据量过大，若满足条件，则将 Excel 文件转换为 Parquet 格式提升读写效率，再读取数据进行后续分析。
```python
import pandas as pd

file_path = "input_data.xlsx"
parquet_path = "temp_data.parquet"

# 读取 Excel 文件并转换为 Parquet 格式

Step1 对数据进行深度清洗，包括合并单元格填充（ffill）、正则化文本处理、RGB 颜色分量转换以及异常值识别。
```python
import re

def clean_data(df, target_col):
    # 1. 处理合并单元格：向下填充
    df[target_col] = df[target_col].ffill()
    
    # 2. 正则清洗：去除数字前缀、特殊字符及首尾空格
    def regex_clean(text):
        if not isinstance(text, str): return text
        text = re.sub(r'^\d+[\.\s\-]+', '', text) # 去除如 "1. " 的前缀
        text = re.sub(r'[^\u4e00-\u9fa5a-zA-Z0-9]', '', text) # 仅保留中英数
        return text.strip()

## Skill Steps

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 对目标列进行数据清洗（去除空值、标准化数值格式），计算合计值，并与指定汇总 Sheet 中的合计行进行精确核对。
```python
target_col = '目标数值列'  # 示例：'建筑面积'
summary_sheet_name = 'Summary' # 示例汇总Sheet名
summary_item_col = '项目'
summary_value_col = '数值'

# 数据清洗：去除空值、强制转换为数值格式
df_cleaned = df_processed.dropna(subset=[target_col]).copy()
df_cleaned[target_col] = pd.to_numeric(df_cleaned[target_col], errors='coerce')

### Step 1 加载数据并配置环境
```python
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns

# 设置中英文字体以支持可视化显示 (SimHei 或 WenQuanYi)
plt.rcParams['font.sans-serif'] = ['SimHei', 'WenQuanYi Zen Hei', 'DejaVu Sans']
plt.rcParams['axes.unicode_minus'] = False

# 加载数据
file_path = 'data.xlsx'  # 替换为实际文件路径
df = pd.read_excel(file_path)

## Skill Steps

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 识别并清洗包含前缀符号的异常数值字段，统一转换为整数类型；同时使用正则表达式清洗文本字段，仅保留 Unicode 范围内的中文字符。
```python
import re
import numpy as np

target_numeric_col = '需要转数字的文本列' # 示例：'获赞'
target_text_col = '需要提取中文的列' # 示例：'收货人'

# 1. 清洗包含前缀符号的数值字段
prefix_patterns = ['.', 'I ', '■ ', '一 ', '_', '. ']

## Skill Steps

### Step1 加载数据与环境配置
```python
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
import re

# 配置中文字体，确保图表正常显示
plt.rcParams['font.sans-serif'] = ['SimHei', 'DejaVu Sans', 'WenQuanYi Zen Hei']
plt.rcParams['axes.unicode_minus'] = False

def load_excel_data(file_path: str, skip_rows: int = 2):

# condition_filtering

> **Note**: This sub-skill covers one step of the Excel analysis workflow. For the full pipeline (file reading, row counting, large-file optimization, export), see the parent workflow SKILL.md.

Step1 执行多维度数据清洗与条件筛选，包含列名自动识别、RGB 颜色过滤、前缀匹配及正则提取。
```python
# 1. 自动识别同义列名并筛选非空值
target_cols = ['域名', '缩写', 'code', 'domain']
for col in target_cols:
    if col in df.columns:
        df = df[df[col].notna()]
        break

# 2. 基于数值通道的精确筛选（如 RGB 颜色过滤）

# Excel_Conditional_Filtering_Optimization

> **Note**: This sub-skill covers one step of the Excel analysis workflow. For the full pipeline (file reading, row counting, large-file optimization, export), see the parent workflow SKILL.md.


Step1 读取 Excel 文件中所有工作表的数据，统计各表行数并汇总，用于评估数据规模。
```python
import pandas as pd

file_path = "input_data.xlsx"

# 读取所有 sheet，统计行数
xls = pd.ExcelFile(file_path)
print("Sheet names:", xls.sheet_names)

# Excel Threshold Analysis and Styling

> **Note**: This sub-skill covers one step of the Excel analysis workflow. For the full pipeline (file reading, row counting, large-file optimization, export), see the parent workflow SKILL.md.


Step1 读取 Excel 文件中所有工作表的行数并汇总，用于评估数据规模。
```python
import pandas as pd

file_path = 'input_file.xlsx'

# 读取所有 sheet 名称并统计总行数
xls = pd.ExcelFile(file_path)
sheet_names = xls.sheet_names

## Skill Steps

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 筛选指定分组数据，将目标列转换为数值类型并计算平均值。
```python
group_col = '班级'  # 占位示例
target_group_value = '358'  # 占位示例
target_cols = ['总分', '理数']  # 占位示例

if group_col not in df_analysis.columns:
    raise ValueError(f"数据中缺少'{group_col}'列。")
df_analysis[group_col] = df_analysis[group_col].astype(str)
filtered_df = df_analysis[df_analysis[group_col] == target_group_value]

## Skill Steps

Step1 提取目标类别数据，清洗无效标签，并统计各类别数量与占比。
```python
import pandas as pd

def calculate_distribution(data, target_col='类别'):
    # 检查目标列是否存在
    if target_col not in data.columns:
        raise ValueError(f'未找到指定的类别字段: {target_col}')
    
    # 提取数据，清洗无效标签（如'--'、'代码'等占位符）
    category_data = data[target_col].dropna().replace(['--', '代码'], pd.NA).dropna()
    

## Skill Steps

> **Note**: This sub-skill covers one step of the Excel analysis workflow. For the full pipeline (file reading, row counting, large-file optimization, export), see the parent workflow SKILL.md.

Step1 提取关键维度与指标信息，处理合并单元格缺失值，并进行多表交叉分析与排序。
```python
import pandas as pd

# 设定目标列名
group_col = '行业名称'
target_val_1 = '企业单位数'
target_val_2 = '工业总产值'

# 读取第一个 Sheet 并清洗

## Skill Steps

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 在数据中动态定位关键字段，通过逐行扫描匹配关键词提取数值，并进行条件筛选与占比计算。
```python
key_values = {}
target_col = None
value_col = 'target_value_col'

# 动态查找目标分类列
for col in df_analysis.columns:
    if 'keyword1' in col.lower() or 'keyword2' in col.lower():
        target_col = col

## Skill Steps

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

### Step1: 数据合并与清洗
```python
combined_df = pd.concat(data_frames, ignore_index=True)

# 数据清洗：使用正则表达式统一命名
if '题型' in combined_df.columns:
    combined_df['题型'] = combined_df['题型'].astype(str).str.replace('判', '判断题', regex=False)

# 处理合并单元格技巧1：前向填充
if '流程描述' in combined_df.columns:

Step 1 加载数据并进行预处理，配置中文字体与环境参数
```python
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
import re

# 设置中文字体，兼容不同环境
plt.rcParams['font.sans-serif'] = ['SimHei', 'WenQuanYi Zen Hei', 'DejaVu Sans']
plt.rcParams['axes.unicode_minus'] = False

# 加载数据并处理合并单元格
file_path = 'input_data.xlsx'

Step1 数据加载与预处理（支持大文件Parquet转换与动态表头识别）。
```python
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.cluster import KMeans
from sklearn.preprocessing import StandardScaler
import os
import re

# 设置中英文字体与图表美化
plt.rcParams['font.sans-serif'] = ['SimHei', 'WenQuanYi Zen Hei', 'DejaVu Sans']
plt.rcParams['axes.unicode_minus'] = False

Step1 读取文件并统计所有 Sheet 的行数，确认数据规模以决定处理策略。
```python
import pandas as pd

file_path = input_file
total_rows = 0
sheet_names = []

try:
    if file_path.endswith('.xlsx'):
        excel_file = pd.ExcelFile(file_path)
        sheet_names = excel_file.sheet_names
        # 统计所有工作表总行数
        for sheet in sheet_names:

# excel_statistical_visualization

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 数据清洗与标准化。提取目标分析列，处理合并单元格（ffill），并利用正则表达式或类型转换清理数值字段，确保分析数据集的准确性。
```python
import re

# 假设 target_col_x 和 target_col_y 是分析目标
# 处理合并单元格导致的缺失值
df['group_col'] = df['group_col'].fillna(method='ffill')

def clean_numeric_string(value):
    if pd.isna(value): return None

# Stacked_Chart_Visualization

Step1 定义百分比转换函数并提取原始数据。通过正则表达式或字符串处理将百分比格式转换为可计算的浮点数。
```python
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

# 配置中文字体，确保图表标签正常显示
plt.rcParams['font.sans-serif'] = ['SimHei', 'DejaVu Sans']
plt.rcParams['axes.unicode_minus'] = False

def convert_percentage(val):
    """

## Skill Steps

Step1 读取Excel文件，统计所有Sheet的总行数。若数据量过大（如≥1万行），则转换为Parquet格式以显著提升后续读取和分析效率。
```python
import pandas as pd

file_path = "input.xlsx"
xls = pd.ExcelFile(file_path)
total_rows = 0

# 统计所有 sheet 的总行数
for name in xls.sheet_names:
    df_temp = pd.read_excel(file_path, sheet_name=name, header=None)
    total_rows += len(df_temp)

> **Note**: This sub-skill covers one step of the Excel analysis workflow. For the full pipeline (file reading, row counting, large-file optimization, export), see the parent workflow SKILL.md.


Step1 读取 Excel 文件，遍历所有 Sheet 统计行数，评估数据规模。
```python
import pandas as pd
import os

file_path = "input_data.xlsx"  # 替换为实际文件路径

if not os.path.exists(file_path):
    print(f"Error: 文件 {file_path} 不存在")
else:

Step1 统计多工作表总行数，并根据数据量级（如≥1万行）动态启用Parquet格式转换以优化大文件读取性能。
```python
import pandas as pd
import os
from openpyxl import load_workbook

file_path = "your_excel_file.xlsx"
xls = pd.ExcelFile(file_path)
sheet_names = xls.sheet_names

# 统计所有sheet的数据行数
total_rows = 0
for sheet in sheet_names:
    wb = load_workbook(file_path, read_only=True, data_only=True)

> **Note**: This sub-skill covers one step of the Excel analysis workflow. For the full pipeline (file reading, row counting, large-file optimization, export), see the parent workflow SKILL.md.

Step1 针对特定 Sheet 进行数据清洗与空值统计。支持处理带空格的列名，并计算关键指标的缺失率。
```python
target_sheet = "Sheet2"
target_col = "是否通过"  # 示例列名，实际根据需求替换

# 读取指定 Sheet
df_target = pd.read_excel(file_path, sheet_name=target_sheet)

# 清洗列名：去除首尾空格
df_target.columns = [str(col).strip() for col in df_target.columns]

if target_col in df_target.columns:

## Skill Steps

Step1 导入依赖并配置中英文字体，防止图表乱码
```python
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import re
import base64
from IPython.display import HTML

# 设置中英文字体
plt.rcParams['font.sans-serif'] = ['SimHei', 'DejaVu Sans', 'WenQuanYi Zen Hei']

Step1 遍历所有sheet，灵活定位目标列并统计特定类型字段的数量。
```python
target_col_keyword = 'type' # 占位示例
target_val_keyword = 'varchar' # 占位示例

total_target_count = 0
target_details = []

for sheet_name in wb.sheetnames:
    ws = wb[sheet_name]
    raw_data = list(ws.iter_rows(values_only=True))

    # 实用技巧：灵活策略定位目标列，通过扫描前几行数据内容定位表头行
    header_row_idx = None

# Skill Steps

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 文本字段清洗，使用正则表达式提取纯中文字符（过滤数字、特殊符号等）。
```python
import re

def extract_chinese(text):
    if pd.isna(text):
        return text
    # 仅保留 Unicode 中文字符范围
    chinese_chars = re.findall(r'[一-龥]', str(text))
    cleaned = ''.join(chinese_chars)

## Skill Steps

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 执行数据清洗，处理合并单元格，使用正则表达式清理文本，并建立分类映射函数骨架。
```python
target_col = '分类字段'
value_col = '数值字段'

# 合并单元格处理 (向下填充还原)
df[target_col] = df[target_col].ffill()

# 数据清洗：正则去除特殊字符、去空、类型转换
df[target_col] = df[target_col].astype(str).str.replace(r'[^\w\s]', '', regex=True).str.strip()

# Formatted_Export

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

## Skill Steps

Step1 对所有 sheet 进行扫描，通过模糊匹配定位目标列，筛选出符合条件（如空值或无效字符）的记录。
```python
empty_target_rows = []
for sheet_name, sheet_df in all_sheets.items():
    target_col = None
    
    # 优先匹配目标列名（示例：包含特定关键字的列）
    for col in sheet_df.columns:

## Skill Steps

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 基于指定列提取有效代码或进行分类映射，生成包含占比与总计行的统计表，并支持交叉分析。
```python
# 分类映射函数骨架
def categorize_item(item_name):
    category_a_keywords = ['keyword1', 'keyword2'] # 占位示例
    if pd.isna(item_name):
        return '未知'
    if any(kw in str(item_name) for kw in category_a_keywords):
        return '类别A'
    return '其他'

## Skill Steps

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 读取目标Sheet，清理字段格式并根据特定条件筛选记录，统计关键指标。
```python
target_sheet = 'Sheet1' # 替换为实际sheet名
df_target = pd.read_excel(file_path, sheet_name=target_sheet)

# 清理目标列的字符串格式（去除首尾空格）
filter_col = 'group_col'
if filter_col in df_target.columns:
    df_target[filter_col] = df_target[filter_col].astype(str).str.strip()

# Skill Steps

> This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 动态读取数据（Parquet加速或常规读取）。
```python
# 若已加载 sn-da-large-file-analysis 技能，将 Excel 文件转换为 Parquet 格式加速读取
if 'da_large_file_analysis' in globals():
    # 假设 sn-da-large-file-analysis 转换后生成了 parquet 文件
    parquet_path = 'auto_converted_data.parquet'
    df = pd.read_parquet(parquet_path)
    print("已使用 Parquet 格式加速读取大文件。")
else:
    df = pd.read_excel(file_path, sheet_name='Sheet1', header=0)

# Image Caption Analysis — 图片描述与数据提取

## Overview

Analyze, extract data from, or understand image files (.png, .jpg, .jpeg, .gif, .webp, .bmp). The core workflow:

1. Run `scripts/caption.py` to get a text description of the image
2. Parse the description into structured data (DataFrame, etc.)
3. Analyze, visualize, or export

## scripts/caption.py — Image Caption

The script converts images to text descriptions via a vision model. Configure via `SN_API_KEY` (minimum required), or use `SN_VISION_API_KEY` / `SN_VISION_BASE_URL` / `SN_VISION_MODEL` for fine-grained control. See the project environment variable spec for the full fallback chain.

# Large Scale Excel Analysis Skill

## Mandatory Rules

> **When total rows >= 10,000, you MUST use the methods in this skill.**

| Data Scale | Read Strategy | Reason |
|-----------|---------------|--------|
| < 10k rows | `pd.read_excel()` directly | No memory pressure |
| 10k–100k rows | `pd.read_excel()` → convert to Parquet → `pd.read_parquet()` for analysis | Avoid repeated slow reads |
| 100k–1M rows | **openpyxl `read_only` + `iter_rows` streaming** → Parquet | `pd.read_excel()` will OOM or timeout |
| > 1M rows | Streaming read + **multi-sheet split** (Excel max 1,048,576 rows per sheet) | Must chunk |

**Prohibited:**

# Document Analysis Skill — Word / PDF / PPT

End-to-end workflow for Word, PDF, and PPT document parsing. Each format has
specific parsing pitfalls — follow the format-specific sub-skill exactly.

---

## Workflow

### Step 0 — Identify file type and input scope

```python
import os

# PDF Analysis

## Step 0 — Detect PDF type (text vs scanned)

**Critical first step**: determine whether the PDF has extractable text or is a scanned image.
Never skip this — using the wrong parser wastes time and produces empty results.

```python
import fitz  # PyMuPDF

def detect_pdf_type(pdf_path, sample_pages=3):
    """
    Returns 'text' if PDF has extractable text, 'scanned' if image-based.
    Checks first N pages (or all if fewer).

# PPT Analysis — .pptx / .ppt

## Environment

```python
from pptx import Presentation
from pptx.util import Inches
import os, subprocess, json

# python-pptx is available
# For .ppt (old binary format): convert via libreoffice
def load_pptx(path):
    if path.lower().endswith('.ppt'):
        import subprocess

# Word Analysis — .docx / .doc

## Environment

```python
from docx import Document
import os

# python-docx is available; for .doc (old format) convert via libreoffice first
def load_doc(path):
    """Load .docx directly; convert .doc to .docx first if needed."""
    if path.lower().endswith('.doc'):
        import subprocess
        out_dir = os.path.dirname(path)

# 深度研究（多 Agent 深度研究编排）

你是深度研究总控。职责是**调度**专家角色完成研究、写作与渲染。

阅读地图：§1 总则 → §2 派发机制 → §3 报告目录 → **§4 档位选择器（决定跑什么）** → **§5 阶段库（每个角色怎么派，仅一次）** → §6 附录。运行时先按 §4 选定本次档位的流水线，再按流水线逐步跳转 §5 的对应条目。

## 1. 总则

**控制器铁律**：

- **只调度，不读大文件**：evidence / 章节 / outline 等大文件通过绝对路径传给角色自读；你只读调度所需的小字段（见 §6）。
- **所有文件路径使用绝对路径**。
- **通过文件路径传递内容**，不在消息里粘贴大段正文。
- **子角色内部检查透明**：各子 agent 自行完成交付检查和必要修复。

# sn-deepresearch-cli

作为 SenseNova-Skills-DeepResearch 的用户入口，负责环境预检、安装引导、参数确认、任务启动、进度告知和结果交付。
默认 npm 包为 `sensenova-skills-deepresearch`，项目源码仓库为
`https://github.com/OpenSenseNova/SenseNova-Skills-DeepResearch`。用户侧直接安装已构建并发布到 npm 的包，
不需要 clone 仓库或本地构建。

## 环境预检

在安装或启动 SenseNova-Skills-DeepResearch 前，仅检查当前环境是否具备运行条件：

- Node.js >= 22、Python >= 3.10 和 npm 是否可用；
- 是否已安装 SenseNova-Skills-DeepResearch 提供的 `deepresearch` 命令；
- 用户选择的 Hermes、Codex、Claude Code 或 OpenClaw 是否存在。

# sn-image-base

## Dependency Installation

```bash
pip install -r requirements.txt
```

## Overview

`sn-image-base` is the base-layer skill (tier 0) of the SenseNova-Skills project and provides four low-level tools:

- `sn-image-generate`: image generation (calls text-to-image-no-enhance API)
- `sn-image-edit`: image editing with SenseNova U1.5 Lite (calls `/images/edits`)

# sn-image-doctor

## Overview

`sn-image-doctor` is an infrastructure skill (tier 0) that validates the SenseNova-Skills environment before running other skills. It ensures SenseNova-Skills project is properly installed and configured.

This skill performs comprehensive checks including:

- Installation verification of SenseNova-Skills project
- Python dependency validation
- Environment variable configuration checks, with interactive prompts to configure missing required variables and save them to `.env`
- Automatic environment reload after configuration changes, with agent restart suggestion if reload fails

## Usage

# sn-image-imitate

Image style imitation scene skill (tier 1), relying on the `sn-image-recognize`, `sn-text-optimize`, and `sn-image-generate` tools provided by `sn-image-base` (tier 0).

Features:

- Extracts high-fidelity long caption from a reference image
- Rewrites caption according to user requested content change while preserving style and layout
- Enforces layout-lock constraints during caption rewrite
- Performs post-generation layout consistency review and bounded retries
- Returns structured process artifacts for debugging and reproducibility

## Non-goals

# sn-image-resume

Resume image generation scene skill (tier 1), relying on the `sn-text-optimize` and `sn-image-generate` tools provided by `sn-image-base` (tier 0).

Features:

- Accepts resume content directly from conversational text
- Supports optional user-provided style direction
- Applies the fixed portfolio-resume layout rules in `prompts/resume.md`
- Generates a tall designed resume image through `sn-image-generate`

## Non-goals

- Editing or polishing a plain text resume document without generating an image

# sn-infographic

Info graphic generation scene skill (tier 1), relying on the `sn-image-generate`, `sn-image-recognize`, and `sn-text-optimize` tools provided by `sn-image-base` (tier 0).

Features:

- Evaluation of prompt quality (auto mode)
- Prompt expansion (force/auto mode)
- Multiple rounds of image generation and VLM review
- Output the best result based on quality ranking

## Input Specification

| Parameter | Type | Default Value | Description |

# Report HTML

把报告文本创作为一份有编辑判断、网页美感、证据秩序、且只属于这份主题的单文件 HTML。

核心原则：**保留事实断言，重构阅读和视觉体验，让设计从报告所属领域和内容任务里长出来。主题自己的世界——它的材质、工具、器物与语汇——是产生独特设计选择的来源。**

## 设计判断原则

- **首屏即论点**：首屏先找到这份报告所属领域里最有识别度、最能承载主题气质的事物或机制，并用最适合它的形式呈现：标题、图片、动画、实时演示或一个可互动瞬间。这个选择必须有明确判断；不要默认使用“大数字 + 小标签 + 辅助统计 + 渐变强调”，只有当它确实是主题最自然、最有力的入口时才使用。
- **文字系统承载页面性格**：标题字体与正文字体要被有意搭配，不要沿用任何项目都能套上的常用字体组合；同时建立清晰的字阶，并有意识地设置字重、字宽和间距。让文字排印本身成为设计中可被记住的一部分，而不是只负责传递内容的中性容器。
- **结构即信息**：编号、眉题、分隔线、标签等结构性手段，应该传递内容的真实信息，而不是装饰内容。许多通用设计都使用编号标记（01/02/03），但这仅适用于内容本身就是一个序列的情况——例如一个真实的流程或一份文字时间线，其中顺序承载着读者所需的信息。在使用编号标记之前，请先思考其是否真的有意义。
- **有意识地使用动效**：先判断动效是否能服务主题、阅读或信息理解，再决定用在哪里：页面加载序列、滚动触发展示、悬停微交互或环境氛围。一个被编排好的关键动效通常比零散特效更有力量；动效选择必须服从整体美学方向。有些主题更适合克制处理，额外动画反而会削弱专业感，并让页面显得像 AI 生成的模板作品。
- **复杂度匹配愿景**：视觉方向越繁复，执行就越需要足够的层次、细节和完成度；视觉方向越极简，间距、字阶、对齐和微细节就越要精准。优雅不等于少，也不等于多，而是把选定的方向执行到位。
- **认真处理文案**：标题、导语、标签、按钮和说明都要被当作设计的一部分处理。可以对源报告的内容重写、压缩、合并和组织表达，但不得发明事实、口径或结论强度。文案和视觉一样会产生模板感。

# SN Motion HTML

Create a reusable project, not a one-off HTML file. Keep subject matter, chapter count, visual style, interface treatment, copy placement, language, image model, video model, clip duration, and hosting path configurable.

The defining experience is one continuous camera journey controlled by page progress. Preserve a single fixed media stage and one ordered timeline from the first scene through every connector to the final scene. Differentiate projects through transparent copy composition, header treatment, type, navigation, and scene-specific copy placement—not by replacing the continuous stage with separate article media blocks.

## Route the work

1. Resolve the story goal, audience, factual fidelity, chapter scale, visual language, copy language, target devices, sound policy, and available API credentials. Make reasonable defaults only for low-risk implementation details.
2. For factual or source-based stories, research before outlining and record sources separately from display copy.
3. Plan chapters, recurring identities and connector paths before generating their boundary images; use [references/prompting.md](references/prompting.md) for shot planning and acceptance criteria.
4. Detect the image-generation capability available in the current environment before generating any stills. The capability may be a callable image tool, a skill that exposes or instructs an image workflow, or both. Use the available capability when present; if no suitable capability is available, report that blocker to the user before continuing past the still-generation stage.
5. Prepare a review package containing the proposed visual style and the content/shot arrangement. When an image capability is available, generate only the small reference/gate set needed for review.
6. Obtain the user's explicit approval at both approval gates below. After approval, generate the complete still set and inspect it before any paid video batch.

# sn-ppt-creative

把公共 Story 转译成一套 16:9 整页视觉：每页生成一张完整 PNG，并可组装为 PPTX。
Creative 负责视觉转译，不负责 Research，不生成自己的内容大纲。

## 前置与目录边界

必须存在：

- `<DECK_DIR>/task_pack.json`
- `<DECK_DIR>/info_pack.json`
- `<DECK_DIR>/outline.md`
- `task_pack.choices.output == "creative"`，或旧任务的 `ppt_mode == "creative"`

# sn-ppt-dazzle：单 HTML 炫彩演示 deck

你要交付的不是一份"幻灯片"，而是一场在浏览器里上演的**动态视觉演出**：持续运行的动态背景、每页的入场编排、页间的过渡特效、键盘可交互——同时排版严谨、内容真实、零控制台报错。

## 0. Entry / Story 输入契约（最高优先级）

本 Skill 不是用户请求入口，不直接接收裸 query 或附件创建新任务。只接受
`sn-ppt-entry` 传入的绝对 `DECK_DIR`，并且必须已经存在：

- `<DECK_DIR>/task_pack.json`；
- `<DECK_DIR>/info_pack.json`；
- `<DECK_DIR>/outline.md`；
- `task_pack.choices.output == "dynamic_html"`，或旧任务的 `ppt_mode == "dazzle"`。

# sn-ppt-doctor

这是按需运行的能力与本地依赖检查，不是 Entry 的强制前置。Doctor 分开报告：

1. 当前 Agent 的普通搜索、图片搜索和图片生成原生工具是否注册、配置或实际验证；
2. PPT 整包自带的 `sn-ppt-tools` 是否存在并已配置。

Doctor 不检查或恢复 `sn_agent_runner.py`，不写 `.env`，也不把可选媒体能力变成硬门槛。

## 持久配置

PPT 内置工具与 Doctor 自动读取：

- Hermes：`~/.hermes/.env`

# sn-ppt-entry

统一接收 PPT 生成请求，建立唯一任务目录，准备材料。Standard 先完成定向外部证据补充，
Deep 先完成完整 Research，**下一步必须调用 `sn-ppt-story` 生成公共 `outline.md`**，再按 `choices.output` 分发到 `sn-ppt-standard`、`sn-ppt-dazzle` 或 `sn-ppt-creative`。

这是一套 Skill 的入口，不是独立运行时。文本理解、视觉理解和推理使用宿主 Agent
已有能力。普通搜索、图片搜索和图片生成优先使用宿主原生工具；原生能力不存在或实际
调用不可用时，才使用 PPT 整包自带的 `sn-ppt-tools`。不得要求用户另外配置模型客户端，
不得调用 `model_client.py`、`stage.py` 或为本流程再造 CLI 调度层。

## Box-Agent 兼容入口

在 Box-Agent 中，从本次加载的 Skill 提示取得 Entry 的绝对 Skill Root，记为
`<ENTRY_SKILL_ROOT>`。Entry 的所有确定性脚本都必须使用