documents-xlsx-skill

XLSX (Excel) Document Generation Skill

Generate downloadable `.xlsx` files for structured tabular output that CSV cannot express: multiple sheets, formulas, formatted cells, frozen headers, currency / percent / date number formats. If the user says "Excel" — produce `.xlsx`, not CSV.

When xlsx vs csv

Pick xlsx if ANY of these are true:

• More than one logical table (use multiple sheets)
• Formatted cells needed (currency, percent, date, bold headers, column widths, freeze panes)
• The user said "Excel", "spreadsheet", "xlsx", "workbook"
• Formulas are required
• The output is a deliverable meant to be opened in Excel / Google Sheets

Pick csv if the output is a single flat table meant for pipeline interchange, raw data dump, or when the file will be read programmatically, not humans.

When in doubt → xlsx.

Output path convention

• Deliverables: `f'{WORKSPACE}/.xlsx'` — auto-detected and uploaded.
• Intermediate working files: `f'{WORKSPACE}/tmp_.xlsx'` — `tmp_` prefix skips auto-upload.
• Never write outside `{WORKSPACE}`. Never use `/tmp/` directly.
• One file per delivery. Multiple sheets inside one file — not multiple files.

Pattern A — single-sheet (pandas, simplest)

```python import pandas as pd df = pd.DataFrame(rows) df.to_excel(f'{WORKSPACE}/campaign-report.xlsx', index=False, sheet_name='Summary', engine='openpyxl') ```

`index=False` is almost always right. Omitting it writes the pandas index as a leading column and confuses users.

Pattern B — multi-sheet (pandas)

```python import pandas as pd with pd.ExcelWriter(f'{WORKSPACE}/performance.xlsx', engine='openpyxl') as w: summary_df.to_excel(w, sheet_name='Summary', index=False) campaigns_df.to_excel(w, sheet_name='Campaign Detail', index=False) weekly_df.to_excel(w, sheet_name='Weekly Trend', index=False) ```

Pattern C — formatted (openpyxl, for polished reports)

Use when cell colors, number formats, frozen headers, custom column widths, or formulas are required.

```python from openpyxl import Workbook from openpyxl.styles import Font, PatternFill, Alignment from openpyxl.utils import get_column_letter

wb = Workbook() ws = wb.active ws.title = 'Summary'

headers = ['Campaign', 'Spend', 'Revenue', 'ROAS'] ws.append(headers)

Bold + branded header row

header_font = Font(bold=True, color='FFFFFF') header_fill = PatternFill('solid', fgColor='006E78') for cell in ws[1]: cell.font = header_font cell.fill = header_fill cell.alignment = Alignment(horizontal='center')

Data rows

for row in rows: ws.append([row['campaign'], row['spend'], row['revenue'], row['roas']])

Currency formatting on Spend + Revenue

for r in range(2, ws.max_row + 1): ws.cell(row=r, column=2).number_format = '"$"#,##0.00' ws.cell(row=r, column=3).number_format = '"$"#,##0.00' ws.cell(row=r, column=4).number_format = '0.00"x"'

Freeze header + widen columns

ws.freeze_panes = 'A2' ws.column_dimensions['A'].width = 32 ws.column_dimensions['B'].width = 14 ws.column_dimensions['C'].width = 14 ws.column_dimensions['D'].width = 10

wb.save(f'{WORKSPACE}/performance.xlsx') ```

Formulas — write as strings starting with `=`:

```python ws['D2'] = '=C2/B2' # openpyxl does NOT evaluate; Excel computes on open ```

For pre-computed totals (recommended when the file must display a numeric value without being opened in Excel first), calculate the total in pandas and write the literal number — not a formula.

Critical rules (read before writing code)

• Sheet names: max 31 chars; no `: \ / ? * [ ]`. Sanitise with `re.sub(r'[:\\/\?\*\[\]]', '', name)[:31]`.
• Unique sheet names: duplicates raise `ValueError`. Dedupe before calling `to_excel`.
• `index=False` on `to_excel` calls unless you specifically want the index as a column.
• Engine: use `engine='openpyxl'` (consistent, vetted). Don't use xlsxwriter — openpyxl covers every case we need and is what pandas defaults to for `.xlsx`.
• Formulas are not evaluated by openpyxl — the file opens with `#VALUE!` / blank until Excel computes. If the consumer will not open in Excel, write literal pre-computed values instead.
• Hard cell cap: 1,048,576 rows × 16,384 columns per sheet. Split across sheets if exceeded.
• 25 MB upload cap. Down-sample or split if the workbook is large.
• No hardcoded data. Always load from workspace files via `dataSources` or previous saves.
• Never use CSV as a substitute when the user asked for Excel. CSV has no formatting, no sheets, no formulas.

Typical ads report (end-to-end example)

```python import pandas as pd from openpyxl import load_workbook from openpyxl.styles import Font, PatternFill, Alignment

1. Build data with pandas

df_summary = pd.DataFrame([ {'Metric': 'Total Spend', 'Value': total_spend}, {'Metric': 'Total Revenue', 'Value': total_revenue}, {'Metric': 'Overall ROAS', 'Value': overall_roas}, ]) df_campaigns = campaigns_df # from earlier analysis df_weekly = weekly_df

2. Write all sheets

out = f'{WORKSPACE}/ads-performance-report.xlsx' with pd.ExcelWriter(out, engine='openpyxl') as w: df_summary.to_excel(w, sheet_name='Overview', index=False) df_campaigns.to_excel(w, sheet_name='Campaign Breakdown', index=False) df_weekly.to_excel(w, sheet_name='Weekly Trend', index=False)

3. Re-open for formatting

wb = load_workbook(out) for sheet_name in wb.sheetnames: ws = wb[sheet_name] # Header styling for cell in ws[1]: cell.font = Font(bold=True, color='FFFFFF') cell.fill = PatternFill('solid', fgColor='006E78') cell.alignment = Alignment(horizontal='center') ws.freeze_panes = 'A2' # Auto-widen based on max content length per column for col_idx, col_cells in enumerate(ws.columns, start=1): max_len = max((len(str(c.value)) for c in col_cells if c.value is not None), default=10) ws.column_dimensions[ws.cell(row=1, column=col_idx).column_letter].width = min(max_len + 2, 40)

wb.save(out) ```

Verification checklist (agent must confirm before finishing)

• File saved under `{WORKSPACE}/`, ends in `.xlsx`, no `tmp_` prefix
• All sheet names ≤ 31 chars, no forbidden chars, unique
• Headers bolded, freeze-panes set (unless trivial single-row data)
• Currency / percent / date columns have `number_format` applied
• No hardcoded sample data in the code
• File size < 25 MB