xlsx Spreadsheet Skill
Equips an agent to create, edit, and analyze Excel files professionally, with zero formula errors and preserved formulas and formatting.
recalc.py writes a Basic macro into the user's LibreOffice profile (~/.config/libreoffice or ~/Library/...), then overwrites the target file in place via calculateAll+store — a persistent side effect on the user environment with no user confirmation, backup, or rollback, and the data flow is not disclosed in the docs. No malware, credential theft, or covert exfiltration found, but confirmation, isolation, and recovery are incomplete, landing mid-scale.
The SKILL.md workflow is self-consistent and recalc.py returns structured JSON error feedback (error types, locations, formula counts), which is good failure feedback; however it assumes LibreOffice is installed and soffice on PATH (macro setup failure returns only a generic error), timeout depends on Linux timeout / macOS gtimeout with no protection if absent, and the stderr keyword-based error attribution is fragile. Static cap is 10 and no dedicated tests cover key paths, so 10 stands without further uplift.
Scenarios are clear (create/edit/analyze xlsx, formula recalculation, financial-modeling conventions) and the trigger description is reasonably precise; but non-fit boundaries are undeclared, environment assumptions (LibreOffice, openpyxl/pandas) are stated only verbally, and Chinese-language support is not addressed. LibreOffice and the Python libraries are reachable in target environments, so no deduction for unreachable overseas services.
Docs are well layered (output standards → workflows → checklists → best practices) with common pitfalls and failure explanations; however the skill declares 'Proprietary. LICENSE.txt has complete terms' while the repository root is MIT and no skill-level LICENSE.txt is shown — a licensing contradiction — and there is no skill-level versioning, changelog, or maintenance ownership.
The workflow (pandas analysis, openpyxl formula-preserving editing, formulas over hardcoded values, mandatory recalculation verification) targets real needs and, if followed, should yield usable files; but there are no committed representative outputs, sample files, or skill-specific test evidence, so direct usability is only statically inferred and the anchor above 7 is unreachable.
Primary material (SKILL.md and recalc.py source) is auditable and claims cross-check against script behavior; but there are no tests covering the skill's key paths, no execution evidence, and no third-party corroboration. Static cap is 5 and coverage is thin, so 4.
- recalc.py writes a macro into the user's LibreOffice profile and overwrites the Excel file in place; back up target files before use, or skip the recalculation step if the side effect is unacceptable.
- The skill is marked Proprietary while the repository is MIT, and the referenced LICENSE.txt is not provided; verify licensing compliance before commercial use.
- Prerequisites (LibreOffice installed, soffice on PATH, openpyxl/pandas available) are unverified by execution; on macOS without gtimeout the script has no timeout protection.
- This is a static source review; no scripts were executed and effectiveness conclusions are inferred.
What does this skill do, and when should you use it?
This is one of 15 skills bundled in the Mini-Agent repository (MiniMax-AI/Mini-Agent, MIT-licensed), located at mini_agent/skills/document-skills/xlsx/. It instructs the agent to use pandas for data analysis, openpyxl for formulas and formatting, and a recalc.py script that drives LibreOffice to recalculate formulas and detect #REF!, #DIV/0!, and similar errors. The skill embeds financial modeling standards: industry-standard color coding, number formats, and source citation for hardcoded values. It suits anyone who needs auditable, dynamic spreadsheets delivered reliably.
Reads .xlsx/.xlsm/.csv/.tsv files with pandas for analysis and visualization; creates or edits workbooks with openpyxl, writing real Excel formulas (never hardcoded computed values) while preserving existing templates; runs recalc.py via LibreOffice to recalculate all formulas and return a JSON error report with locations for fixing; enforces financial modeling conventions — blue inputs / black formulas / green cross-sheet links, $#,##0 and 0.0% number formats, assumptions in separate cells, and 'Source: ...' comments on hardcodes.
- An analyst generating a financial or business model from scratch that must ship with zero formula errors
- A data worker exploring an unfamiliar workbook, needing quick reads, statistics, and multi-sheet inspection
- A user updating an existing template whose original formulas and layout conventions must be preserved exactly
- A finance professional needing valuation or projection models that follow investment-banking color coding and number formatting
- A developer batch-editing spreadsheets programmatically with automated verification for #REF!, #NAME?, and similar errors
What are this skill's strengths and limitations?
- Complete closed-loop workflow: read, edit, write, recalculate, and verify errors
- Mandates Excel formulas over hardcoded values, keeping output files dynamic and updateable
- Bakes in industry-grade financial modeling standards (color coding, number formats, assumption separation, source citations)
- recalc.py returns structured JSON error reports, making issues easy to locate and fix
- The SKILL.md is marked Proprietary (see LICENSE.txt), differing from the repo's overall MIT license — review terms before commercial use
- Formula recalculation hard-depends on LibreOffice; without it the recalculation step cannot run
- No test coverage evidence specific to this skill; the README's test suite targets the Mini-Agent core, not this skill
- Steps for using this skill standalone outside Mini-Agent are undocumented
How do you install this skill?
The skill ships with the Mini-Agent collection: run uv tool install git+https://github.com/MiniMax-AI/Mini-Agent.git, then the setup-config.sh (or .ps1) script to create ~/.mini-agent/config/config.yaml and fill in your MiniMax API key. For development, clone the repo, run uv sync, and initialize skills with git submodule update --init --recursive. The skill lives at mini_agent/skills/document-skills/xlsx/. Note: steps for manually installing this single SKILL.md into other Agent Skills clients are not documented in the source.
How do you use this skill?
Start the Mini-Agent CLI (mini-agent or mini-agent --workspace /path/to/project) and state a spreadsheet task in natural language, e.g. "Read sales.xlsx and build a summary sheet with SUM formulas." Whenever formulas are used, run python recalc.py output.xlsx to recalculate, then fix any issues reported in the JSON output (check error_summary when status is errors_found) and recalculate again. LibreOffice must be installed in the environment.
How does this skill compare with similar options?
The SKILL.md references Claude Skills (https://github.com/anthropics/skills) as its upstream origin; this skill is that xlsx skill as bundled with Mini-Agent. The source names no other comparable alternatives.