Xlsx Spreadsheet Skill
Enables AI agents to create, edit, and analyze Excel files using real formulas instead of hardcoded values, so spreadsheets stay dynamic.
Also a bundled copy of Anthropic's official xlsx skill — content is trustworthy but the version may lag the official source.
What does this skill do, and when should you use it?
This is the xlsx skill from the built-in Skills collection in the Open Cowork desktop app, located at .claude/skills/xlsx/. It guides an AI agent to use pandas for data analysis and openpyxl for formula and formatting work, with a recalc.py script that leverages LibreOffice to recalculate formulas and scan for errors. It enforces strict output standards: zero formula errors on delivery, preservation of existing template styles, and industry-standard color-coding and number-formatting rules for financial models. Note that the skill itself is licensed as Proprietary (per LICENSE.txt), unlike the MIT license of the overall repository.
Reads and analyzes .xlsx/.xlsm/.csv/.tsv files via pandas; creates and modifies workbooks with openpyxl, writing Excel formulas (e.g., =SUM, =AVERAGE) rather than Python-computed hardcoded values; runs recalc.py, which invokes LibreOffice to recalculate all formulas and return a JSON error report (#REF!, #DIV/0!, etc.); applies financial-modeling conventions such as blue inputs, black formulas, green cross-sheet links, and standard number formats (thousands-separated currency, parenthesized negatives, 0.0x multiples); requires source documentation for any hardcoded values.
- A finance professional wants an AI to build a dynamic financial model with assumptions in separate cells and cell-referenced formulas, not baked-in numbers
- A data analyst receives a multi-sheet Excel report and needs the AI to read, summarize, and modify it without destroying existing formulas
- An operations user needs to edit a colleague's template and must strictly match the original formatting rather than imposing a standardized style
- A developer needs to batch-process CSV/TSV data and export it as a formatted xlsx with live formulas
- An audit-style workflow requires the AI to detect and fix every formula error before delivering the file
What are this skill's strengths and limitations?
- Mandates Excel formulas over hardcoded results, so outputs recalculate when source data changes
- recalc.py auto-configures LibreOffice and pinpoints every formula error via JSON, closing the verification loop
- Includes investment-banking-grade modeling conventions: color standards, number formats, separated assumptions, source citations
- Clearly delineates pandas (analysis) vs openpyxl (formulas/formatting) and warns that saving with data_only=True permanently destroys formulas
- Formula recalculation hard-depends on LibreOffice; without it the verification step cannot run
- The skill is Proprietary-licensed (LICENSE.txt), differing from the repo's MIT license — review full terms before commercial use
- recalc.py is documented for Linux and macOS only; Windows support is unconfirmed
- No automated test suite; edge-case coverage relies on the agent following the checklist
- No standalone usage examples or versioning information provided apart from the Open Cowork app context
How do you install this skill?
The skill ships as part of the Open Cowork repository's Skills collection at .claude/skills/xlsx/. With Open Cowork it is built in (install via Homebrew: brew tap OpenCoworkAI/tap && brew install --cask --no-quarantine open-cowork). To use it with another Agent Skills-compatible client, copy the xlsx folder into that client's skills directory, including the recalc.py script. The full terms in the separate LICENSE.txt are not provided in the source material.
How do you use this skill?
- Ensure Python (pandas, openpyxl) and LibreOffice are installed. 2. Give the agent a spreadsheet task, e.g., "Read sales.xlsx in this folder and add a summary sheet with SUM formulas by region." 3. The skill directs the agent to pick the right tool (pandas or openpyxl), use formulas instead of hardcoded values, save, then run python3 recalc.py output.xlsx to recalculate. 4. Inspect the returned JSON; if status is errors_found, fix the locations listed in error_summary and recalculate until zero errors are delivered.
How does this skill compare with similar options?
The Open Cowork README lists this skill alongside its siblings pptx, docx, pdf, and skill-creator as built-in Skills; comparisons drawn there against Claude Cowork apply to the Open Cowork app as a whole, not to this skill alone. The source provides no direct comparison with other xlsx skills.