Headless Excel Workbook Builder
Create reviewable Excel files in managed-agent sessions without an open Office application.
The skill limits its stated effect to writing ./out/<name>.xlsx locally, requires no external service, credentials, or elevated permissions, and says not to append to an existing workbook without explicit instruction. However, it lacks user-confirmation rules, sensitive-data handling, data-flow disclosure, rollback guidance, and dependency-security controls, so points are deducted.
The core instructions, output path, and openpyxl example are broadly consistent and define useful input, formula, and checks conventions. However, Python/openpyxl availability is not pinned or checked, and abnormal-input, save-failure, formula-recalculation, and validation-failure feedback is unspecified; static evidence cannot justify a higher score.
The target headless/managed-agent scenario, output form, and non-fit live-Office scenario are reasonably clear. Input boundaries, supported workbook complexity, semantic trigger conditions, and Chinese/localization support are not defined, limiting adaptability.
The document has clear sections for the output contract, construction example, conventions, and non-fit cases. Repository context provides an Apache-2.0 license, CI validation, and contribution guidance, but the skill itself lacks versioning, changelog, named maintenance ownership, dependency installation notes, and troubleshooting guidance, so it is not full marks.
It can plausibly produce a directly collectible xlsx artifact and adds useful financial-model conventions such as Inputs, formulas, color coding, and Checks. The example is minimal and contains no verified representative output, formula-result validation, or complex-input handling, so manual review remains material.
There is a revision-pinned source, a committed example, and repository-level plugin-validation and secret-scan workflows. There are no tests covering xlsx-author key paths, committed generated artifacts, or third-party execution evidence; the assessment therefore relies mainly on static documentation and is capped accordingly.
- Confirm that Python and a compatible openpyxl version are installed; the skill provides no installation or version-pinning guidance.
- Formulas, formatting, balance checks, and final artifact integrity still require human verification; cleanup, retry, and rollback behavior on failure is unspecified.
- The skill targets headless managed environments; use the documented alternative when live Excel tools are available.
What it does & when to use it
xlsx-author is designed for managed-agent or CMA-style headless sessions that need to deliver an Excel workbook as a file artifact. It uses Python and openpyxl to create and save .xlsx files instead of editing a live Excel workbook. The skill requires output under ./out/<name>.xlsx and defines conventions for inputs, formulas, links, named ranges, and balance checks. It fits automated financial-modeling workflows that lack an open Office application.
Creates ./out/ when needed; writes and runs a short Python script; uses openpyxl to create or edit workbooks, worksheets, cells, styles, and formulas; marks hardcoded inputs, formulas, and links to other sheets or files in blue, black, and green; keeps inputs on an Inputs tab; uses named ranges for values referenced from decks or memos; adds a Checks tab with TRUE/FALSE checks for balance-sheet balances and cash-flow-to-cash ties; saves the workbook as ./out/<name>.xlsx.
- A managed-agent user needs a financial workbook generated without an open Excel application.
- A financial analyst needs inputs, calculations, links, and checks organized for workbook review.
- An automation pipeline must return an .xlsx artifact rather than operate a live Office workbook.
- A model builder needs explicit balance-sheet and cash-flow reconciliation checks in the output file.
Pros & cons
- Produces a file-based .xlsx artifact for headless managed-agent workflows.
- Provides concrete conventions for inputs, formulas, links, named ranges, and checks.
- Uses Python and openpyxl for workbook generation.
- Requires calculations to use formulas and encourages explicit balance checks.
- The source covers workbook generation, not live Excel editing or advanced charting capabilities.
- Requires a shell, writable filesystem, Python, and openpyxl.
- No test suite, platform validation matrix, or failure-recovery process is documented.
- When mcp__office__excel_* tools are available, the README positions them as the live-workbook option; this skill is the headless file-producing fallback.
How to install
The skill is located at plugins/agent-plugins/gl-reconciler/skills/xlsx-author/ in the anthropics/financial-services repository. The README says to install through Cowork by opening Settings → Plugins → Add plugin, pasting https://github.com/anthropics/financial-services, and selecting the desired agent. For Claude Code, run claude plugin marketplace add anthropics/financial-services, then claude plugin install gl-reconciler@claude-for-financial-services. The source does not document a separate xlsx-author installation command.
How to use
Use it in a managed-agent or CMA headless session when requesting an Excel file artifact. Example: Create a financial-model workbook and save it as ./out/model.xlsx. The skill requires the final response to include the relative path so the orchestration layer can collect the file. SKILL.md does not define a dedicated slash command or fixed trigger syntax.
Compared to similar skills
The source explicitly contrasts xlsx-author with mcp__office__excel_* tools: xlsx-author creates a workbook file on disk, while those tools drive a user's live workbook. No other concrete competitor comparison is provided.