create-spreadsheet

name: create-spreadsheet title: Create Spreadsheet description: Generate an Excel (.xlsx) workbook or a CSV file — tables, financial models, data exports, formatted reports with charts. Use when the user wants a spreadsheet or CSV. Built with openpyxl + pandas (Python).

When NOT to use: a Google Sheet (use the googlesheets skill), a PDF table (use create-pdf), or a Word doc (use create-docx).

target: docgen_agent

Create Spreadsheet

When to use

The deliverable is .xlsx or .csv. For Google Sheets use the googlesheets skill.

Important: values, not live formulas

There is no spreadsheet recalc engine in the sandbox. If you write a formula like =SUM(...), Excel shows it as blank until the user opens and recalculates. So:

• Compute results in Python (pandas/numpy) and write the final values. This is the default and it always opens correctly.
• Only add a literal =FORMULA when the user explicitly wants an editable formula — and when you do, also write the computed value in an adjacent labeled cell so the number is visible immediately.

How it works

A template here is a Python program (openpyxl + pandas) that writes the workbook. You adapt the data and structure, then the build script runs it and validates the file.

Skill directory: /workspace/integrations/docgen/agent/skills/create-spreadsheet — templates/, reference.md, and scripts/ live here; read them by absolute path. Your work goes in ./scratch/.

Workflow (.xlsx)

1. Work in a job dir: mkdir -p ./scratch/xlsx-<short-name>.
2. Adapt the template — don't start blank. read /workspace/integrations/docgen/agent/skills/create-spreadsheet/templates/report.py, then write the edited program to ./scratch/xlsx-<name>/sheet.py. Read /workspace/integrations/docgen/agent/skills/create-spreadsheet/reference.md for openpyxl building blocks and the values rule.
3. Build + validate:

   bash /workspace/integrations/docgen/agent/skills/create-spreadsheet/scripts/build.sh ./scratch/xlsx-<name>/sheet.py ./scratch/xlsx-<name>/out.xlsx
   

   Prints OK: <path> (sheets=N) or ERROR: ....
4. Fix loop: on ERROR, read the message, fix, re-run. Cap at 5 attempts.
5. Deliver: mv ./scratch/xlsx-<name>/out.xlsx ./artifacts/<final-name>.xlsx, then report the full path.

CSV (no build needed)

CSV is plain text — write it directly with bash and deliver:

python3 - <<'PY'
import csv
rows = [["name","amount"], ["Acme", 1200], ["Globex", 980]]
with open("./artifacts/<final-name>.csv", "w", newline="") as f:
    csv.writer(f).writerows(rows)
PY

No toolchain, no validation step — just confirm the file is in ./artifacts/ and report the path.

Toolchain note

The build script provisions openpyxl + pandas on first use; the first workbook may take a little longer, then it's fast.

Never

• Never write = formulas as the only representation of a computed number (no recalc → blank cells). Write the value.
• Never leave the file in ./scratch/. Never skip build/validate for .xlsx.

Templates

• report.py — formatted workbook: styled header, data rows, a computed totals row, number formats, column widths, and a bar chart.