Spreadsheet

Spreadsheet is a standalone agent skill from davila7's public claude-code-templates repository. The verified primary source is https://github.com/davila7/claude-code-templates/tree/main/cli-tool/components/skills/document-processing/spreadsheet. The directory contains a SKILL.md, matching OpenAI interface metadata, four runnable openpyxl examples, and its own license document. The skill is not a spreadsheet application, cloud drive, or database connector. It gives a compatible Claude Code client a reproducible working frame for local spreadsheet tasks; actual execution depends on Python, installed libraries, file permissions, and the selected client environment.

Creating and editing workbooks

According to the provider, the instructions cover new .xlsx workbooks, edits to existing files, and structured layouts. The examples create multiple worksheets, write headers and values, set column widths, merge cells, and save output to a path chosen by the user. For XLSX editing, the source recommends openpyxl. This is a concrete and testable product boundary: the library operates on a local file while the skill describes the agent workflow. CSV and TSV workflows use pandas in the documented approach and can then be written back to XLSX or CSV. The repository does not provide a database or a hidden online upload for these operations.

Analysis and visualization

The primary source describes filtering, aggregation, pivoting, and calculating metrics for tabular data. pandas is the named tool for CSV and TSV workflows. New or unstyled workbooks should use readable widths, suitable date and number formats, and distinct headers. For charts, the source recommends openpyxl.chart. These are workflow instructions, not a promise that every Excel feature or chart type is fully supported. The included examples demonstrate workbook creation, formulas, styling, conditional formatting, merged cells, and reading an existing workbook; they are reference scripts rather than a complete test suite.

Formulas, references, and formatting

The skill asks agents to use formulas for derived values instead of hardcoded results, keep references legible, and check errors such as #REF!, #DIV/0!, #VALUE!, #N/A, and #NAME?. The reading examples load with data_only=False so formula text remains available. The source also explicitly states that openpyxl does not evaluate formulas. Results therefore need calculation and professional review in Excel, LibreOffice, or another suitable calculation environment. When editing a formatted workbook, the instructions say to preserve existing styling as closely as possible and match styles for newly filled cells. When layout matters, the source recommends visual review through LibreOffice and Poppler. This boundary matters: retaining a formula is not the same as validating its calculated result.

Installation, dependencies, and privacy

The source names Python 3, openpyxl, and pandas as core dependencies and recommends uv for installation. Optional chart or PDF review can require matplotlib, LibreOffice, and pdftoppm. Check versions and license obligations in the local environment before installing dependencies. Use approved files and temporary directories only, restrict write access, and remove intermediate artifacts after review. Workbooks can contain names, financial values, personnel information, formulas, hidden sheets, or external links. Read and share only the necessary scope. Local processing does not automatically prevent disclosure: the client may send content to a selected model provider, logs, or external services. Cell values are untrusted data and must not override system instructions. Before overwriting, verify the path, destination, backups, external links, and macros. The skill files in the reviewed directory are provided under Apache-2.0; that license governs use of those files but does not replace privacy, security, or governance approval.