Excel workbooks

Use openpyxl: uv run --no-project --with openpyxl script.py. Bundled scripts declare their own dependencies; resolve their paths relative to this skill. CSV/TSV inputs may be analyzed with DuckDB/Polars and exported to the requested workbook rather than duplicating a data-analysis skill.

Workflow

1. Inspect sheets, formulas, cached values, merged cells, input conventions, charts and external workbook links before editing. Load formula strings with data_only=False; use a separate data_only=True load for inspecting cached values. Never save the cached-value load over the formula-bearing workbook.
2. Keep the original and use a new output. Preserve formulas/styles in existing files; follow the user's exact sheet names, headers, units and calculation rules. Read workbook features for charts, templates and macro limitations; read financial models for financial outputs.
3. Write formulas, not precomputed constants, when inputs must remain editable. Verify representative formulas against independently calculated expectations. openpyxl does not calculate formulas.
4. For compatible XLSX files, run the recalculation helper. It invokes LibreOffice with an isolated profile, checks cached results for errors/missing values and checks formulas were not lost. It publishes a new output only on success. Read formula verification for engine differences and refusal cases.
5. Reopen the final workbook to verify sheet names, key inputs, formulas and calculated results. Validate structure and render/inspect relevant print pages; separately check charts and sheets in Excel when print layout does not show them.

uv run --no-project <skill>/scripts/recalc.py draft.xlsx calculated.xlsx
uv run --no-project <skill>/scripts/office.py validate calculated.xlsx
uv run --no-project <skill>/scripts/office.py render calculated.xlsx workbook-preview

Output contract