/sn-da-excel-workflow

name: sn-da-excel-workflow
description: "Excel 数据分析多步编排器。覆盖：(1) 读取多 Sheet Excel 文件并统计行数，(2) 大文件检测（≥10k 行自动 Parquet 优化），(3) 数据清洗（缺失值、文本标准化、无效字符），(4) 条件筛选与分类提取，(5) 跨 Sheet 统计聚合，(6) 导出 Excel/CSV 并提供下载链接。覆盖从数据读取到报告生成全流程，按步骤编排 capability 子 skill。**遇到以下任一情况就主动使用本 skill，不要自行写几行 pandas 就回答**：①用户出现触发词：Excel 分析 / 表格分析 / 数据分析 / 数据清洗 / 数据统计 / 数据筛选 / 数据可视化 / 数据导出 / 汇总统计 / 透视表 / 分组统计 / 交叉分析 / 趋势分析 / 对比分析 / 异常值检测 / 去重 / 缺失值处理 / Excel 报告 / 生成报表 / analyze Excel / data analysis / data cleaning / pivot table；②用户上传或指定了 .xlsx / .xls / .csv 文件并要求分析、清洗、统计或可视化；③任务涉及多 Sheet 读取、条件筛选、分类汇总、图表生成中的任意一项；④用户要求导出带格式的 Excel 报告或下载链接。仅不用于：不涉及表格数据的纯文本处理、图片分析（使用 sn-da-image-caption）、单个公式计算的简单问答。"

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.

import openpyxl, gc

wb = openpyxl.load_workbook(file_path, read_only=True, data_only=True)
total_rows = 0
sheet_info = {}
for name in wb.sheetnames:
    ws = wb[name]
    row_count = sum(1 for _ in ws.iter_rows(min_row=2, values_only=True))
    total_rows += row_count
    sheet_info[name] = row_count
    print(f"Sheet '{name}': {row_count} rows")
wb.close()
print(f"总行数={total_rows}")

⚠️ **Do NOT use `pd.read_excel()` to count rows** — it loads all data into memory, which will OOM on large files.

→ capability: `excel-reading/multi-sheet-reading`

Step 2 — Large file gate (CRITICAL — choose strategy by row count)

| total_rows | Strategy | What to do | |-----------|----------|------------| | < 10k | Direct read | `df = pd.read_excel(file_path, sheet_name=target_sheet)` | | 10k – 100k | Parquet cache | `pd.read_excel()` once → `df.to_parquet()` → all later reads from Parquet | | **>= 100k** | **STOP. Load `sn-da-large-file-analysis` skill** | Read its SKILL.md, then follow its streaming read + Parquet pattern. **Do NOT use `pd.read_excel()` at all** — it will OOM or timeout on 100k+ rows. |

**For >= 100k rows:**

read_file(path="<skills_base>/sn-da-large-file-analysis/SKILL.md")

Then use `stream_excel_to_parquet()` from that skill — it reads via openpyxl `iter_rows` in 50k-row chunks with constant memory.

**For 10k – 100k rows (only):**

import pandas as pd
parquet_path = "/tmp/_auto_parquet.parquet"
df = pd.read_excel(file_path, sheet_name=target_sheet)
df.to_parquet(parquet_path, engine="pyarrow")
del df; gc.collect()
df = pd.read_parquet(parquet_path)

→ capability: `excel-reading/large-excel-reading`

Step 3 — Inspect schema & data types

Preview target sheet structure. **For large files (>= 10k rows), only read a small sample — never full load just to inspect.**

# For any file size — read only first N rows for inspection
df_head = pd.read_excel(file_path, sheet_name=target_sheet, nrows=20)
print(f"Columns: {df_head.columns.tolist()}")
print(f"Dtypes:\n{df_head.dtypes}")
print(df_head.head(10))

→ capability: `excel-reading/range-reading`

Step 4 — Data cleaning

Handle missing values, normalize text, clean invalid characters.

# Missing values
null_count = df[col].isna().sum()

# Text cleaning: keep only Chinese characters
import re
def clean_text(val):
    if pd.isna(val): return val
    return "".join(re.findall(r"[\u4e00-\u9fff]", str(val))) or ""

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

⚠️ **Large file rule**: When `total_rows >= 100k`, do NOT use `df.apply(lambda...)`. Use vectorized operations or `np.where()` instead. See `sn-da-large-file-analysis` skill for the vectorized cheat sheet.

→ capabilities:

• `excel-data-cleaning/missing-value-handling`
• `excel-data-cleaning/invalid-data-cleaning`
• `excel-data-cleaning/text-normalization`

Step 5 — Filter & extract

Apply condition or category filters, aggregate results.

# Condition filter
mask = df[col].astype(str).str.strip() == target_value
filtered = df[mask]

# Category extraction (for headerless layouts)
df_raw = pd.read_excel(file_path, sheet_name=sheet, header=None)
# Walk rows to find category markers, collect items until next marker

→ capabilities:

• `excel-data-filtering/condition-filtering`
• `excel-data-filtering/category-filtering`
• `excel-data-filtering/threshold-filtering`

Step 6 — Export results

Save filtered/cleaned data as Excel or CSV. Provide download link.

output_path = "/mnt/data/result.xlsx"
result_df.to_excel(output_path, index=False)
print(f"[Download](sandbox:{output_path})")

→ capabilities:

• `excel-result-export/single-sheet-export`
• `excel-result-export/formatted-export`

Key rules

• **Always count rows first** — gate large-file logic on the 10k threshold.
• **>= 100k rows → MUST load `sn-da-large-file-analysis` skill** — do not attempt to handle with `pd.read_excel()`.
• **Column names may contain spaces** (e.g. `'是否通 过'`) — use exact string indexing.
• **Headerless sheets** — use `header=None` and positional indexing.
• **Prohibited on large files (>= 100k rows)**:
• `pd.read_excel()` for full load (use streaming read → Parquet)
• `df.apply(lambda...)` or `df.iterrows()` (use vectorized ops or `itertuples()`)
• `fc-list`, `find ... fonts`, `subprocess` to search fonts, or `pip install` (use fixed font paths below)
• Printing all unique values or full DataFrames (use `.head()`, `.value_counts().head()`)

CJK Font Setup (mandatory for charts)

When generating charts with matplotlib, **copy this block as-is**. Do NOT search for fonts.

import os
import matplotlib
import matplotlib.pyplot as plt
import matplotlib.font_manager as fm

_FONT_PATHS = [
    '/mnt/afs_agents/SimHei.ttf',
    '/mnt/afs_agents/mnt/data/SimHei.ttf',
    os.path.expanduser('~/.fonts/SimHei.ttf'),
    '/usr/share/fonts/truetype/wqy/wqy-zenhei.