Office XLSX

office-xlsx — native XLSX with openpyxl

You build editable .xlsx workbooks with Python openpyxl: real formulas the recipient can recalc, explicit number formats and fonts, and native charts. You can also read and parse existing workbooks. No LibreOffice.

Runtime contract

• uv must be on PATH. All Python runs through uv run; the runtime injects the package-mirror environment variables so uv resolves openpyxl from the internal mirror. You do not configure the mirror.
• If uv is missing, stop and report exactly: office-xlsx requires 'uv' on PATH, but 'uv --version' failed. This is an environment problem — uv should be provisioned by the PawWork runtime. Do not fall back to system pip; report the missing uv instead. Do not attempt pip install or any other fallback.
• The skill directory ships read-only inside the app bundle. Never write into it. Work in a fresh directory.

SKILL_DIR below means the skill's base directory as a plain filesystem path — copy the line labeled "Base directory as a plain filesystem path" from the skill-load output. Do not use the file:// URL form; if only a file:// URL is available, strip the file:// prefix first.

SKILL_DIR="<plain filesystem path from the skill-load output, no file:// prefix>"
uv --version || { echo "office-xlsx requires 'uv' on PATH (see Runtime contract)"; exit 1; }
mkdir -p work && cp "$SKILL_DIR/pyproject.toml" work/pyproject.toml   # pins openpyxl==3.1.5

Run every uv run command from work/.

Authoring rules (hard floor — do not relax)

• Real formulas, not baked-in numbers. When a cell is a computed result (totals, ratios, growth, margins), write the Excel formula string (ws["B10"] = "=SUM(B2:B9)"), never the pre-computed value. The recipient must be able to change an input and see downstream cells recalc. The delivery gate fails a data model with zero formula cells when --require-formula is set.
• Formula injection: only YOUR computed cells may be formulas. Any value that comes from user-supplied data (CSV imports, pasted text, names, notes) and starts with =, +, -, or @ must be stored as text, never as a formula — e.g. a user field =HYPERLINK("http://evil","x") would otherwise become a live formula. Force text with cell.value = raw; cell.data_type = "s"; cell.number_format = "@". When you genuinely need a link, use the cell.hyperlink attribute, never the HYPERLINK() formula. The gate FAILs on formulas calling HYPERLINK / WEBSERVICE / IMPORT* / DDE-style patterns.
• Explicit number formats. Set cell.number_format for every money / percent / date column (e.g. "#,##0.00", "0.0%", "yyyy-mm-dd"). Do not rely on the default General format for financial output.
• Explicit fonts and styles. Set header rows with an explicit Font(bold=True, size=...) and, where it aids reading, PatternFill / Alignment / Border. Do not leave headers visually identical to body rows.
• Freeze panes and widths. Freeze the header row (ws.freeze_panes = "A2") and set sensible column_dimensions[col].width so columns are not clipped.
• Charts are native objects. When the task calls for a chart, build it with openpyxl.chart (BarChart, LineChart, PieChart, ...) bound to a Reference over the data range — a real chart object, not a pasted image.
• One concern per sheet. Split inputs / model / outputs across named sheets rather than cramming everything into Sheet1.

Minimal shape:

from openpyxl import Workbook
from openpyxl.styles import Font
wb = Workbook()
ws = wb.active
ws.title = "Model"
ws["A1"] = "Quarter"; ws["B1"] = "Revenue"
ws["A1"].font = ws["B1"].font = Font(bold=True, size=12)
ws["A2"] = "Q1"; ws["B2"] = 100
ws["A3"] = "Q2"; ws["B3"] = 140
ws["A4"] = "Total"; ws["B4"] = "=SUM(B2:B3)"    # real formula
ws["B2"].number_format = ws["B3"].number_format = ws["B4"].number_format = "#,##0"
ws.freeze_panes = "A2"
wb.save("out.xlsx")

Reading / parsing

load_workbook(path) reads formulas; load_workbook(path, data_only=True) reads the last cached computed values. Use data_only=True when you need results, data_only=False when you need to inspect or preserve formulas. To edit an existing workbook, load_workbook(path), change cells (respecting the formula-injection rule above for any user-supplied values), then wb.save('out.xlsx').

Gate — structure check (before delivering)

Run the structure gate and fix every reported gap by rebuilding from your generator — never hand-edit the zip — until it prints OK:

uv run python "$SKILL_DIR/scripts/check_xlsx.py" out.xlsx --min-sheets 1 [--require-formula] [--require-chart]

Pass --require-formula for any calculation / model task and --require-chart when the task calls for a chart. The gate confirms the workbook opens, has populated cells, flags injection-style formulas, and (when required) contains real formula cells and chart parts. Charts are counted from the package's xl/charts/chart*.xml parts — a stable OOXML signal — so the gate does not depend on openpyxl's chart read-back.

The workbook is not done until this gate prints OK. The machine check decides completion, not your own judgement.