Headless Excel Workbook Builder
Create deliverable Excel files in managed-agent sessions without an open Office application.
The skill limits its stated action to creating a local ./out/<name>.xlsx artifact, requires no external connector, and distinguishes headless mode from Cowork mode. It lacks user confirmation, sensitive-data handling, rollback or cleanup guidance, and dependency-security controls, so points are deducted.
The output path, openpyxl example, and modeling conventions make the happy path understandable. There are no tests, input validation, exception handling, pinned dependency or diagnostic failure messages, so the static score remains limited.
It clearly targets headless/CMA sessions without an open Office application and identifies Cowork mode as out of scope. Inputs, complex-workbook boundaries, trigger precision, Chinese-language support, and broader environment compatibility are not specified, so points are deducted.
The document has concise sections for the output contract, example, conventions, and non-fit case. Repository-level Apache-2.0 licensing and official provenance are available, but the skill lacks versioning, changelog, maintenance ownership, installation notes, FAQs, and explicit limitations.
It can directly guide creation of a basic xlsx artifact with formulas, styles, and a known output location. The example is narrow and does not demonstrate a realistic month-end-close workbook, validation, or complex inputs, so direct usability still requires review.
The instructions are auditable, and repository CI provides plugin validation and secret scanning. Those workflows do not reproduce the skill's key workbook-generation paths, and no committed test suite or third-party execution evidence is provided, so verification is limited.
- Confirm that a compatible Python environment and openpyxl installation are available; installation and version pinning are undocumented.
- The skill writes to ./out/ but does not define overwrite behavior, failed-artifact cleanup, or recovery.
- Color, named-range, and Checks conventions are advisory only; no automated audit or formula-result verification is supplied.
- Do not treat the example as a complete month-end-close workflow; financial outputs still require human review.
What it does & when to use it
xlsx-author creates Excel workbook files in headless managed-agent or CMA environments. It uses Python and openpyxl to write the workbook to disk instead of driving a live Excel workbook. The required output location is ./out/<name>.xlsx, and the relative path must be returned in the final message. The skill also defines conventions for inputs, formulas, links, named ranges, and model checks.
Creates ./out/ when needed, writes and runs a Python script, uses openpyxl to create or modify one Excel workbook, and saves it as ./out/<name>.xlsx. The workbook may include Inputs, calculation, and Checks tabs; hardcoded inputs use blue font, formulas use black font, and links to another sheet or file use green font. Calculation cells must contain formulas, values referenced from decks or memos should use named ranges, and applicable balance checks should be included for the balance sheet, cash flow, and related ties.
- A managed-agent user needs an Excel financial model when no Office application is open.
- A financial analyst needs to generate a new workbook with input, calculation, and check tabs using Python and openpyxl.
- A modeler needs consistent color conventions for hardcoded inputs, formulas, and cross-sheet links during review.
- An orchestration workflow needs a generated .xlsx file saved on disk as a collectable artifact.
Pros & cons
- Designed for managed-agent environments without an open Office application.
- Defines a clear output path and artifact-delivery contract.
- Provides concrete conventions for inputs, formulas, links, and checks.
- Uses Python and openpyxl, making workbook generation scriptable.
- Requires Python, Bash, filesystem write access, and openpyxl.
- The source provides instructions and an example but no test suite or supported-platform matrix.
- It is not intended for interactive editing of an already open Excel workbook.
- The source does not document formula recalculation, complex-format compatibility, or error-recovery behavior.
How to install
The skill is located at plugins/agent-plugins/month-end-closer/skills/xlsx-author/SKILL.md. The README does not document a standalone installation command for this skill; in Cowork, you can open Settings → Plugins → Add plugin, paste https://github.com/anthropics/financial-services, and select the required agent or vertical. The repository is licensed under Apache-2.0.
How to use
In a headless managed-agent or CMA session, request an Excel artifact, for example: “Create an Excel workbook with Inputs, calculation, and Checks tabs and save it as ./out/model.xlsx.” The skill requires running a Python script with Bash and using openpyxl to generate the file; the final response must return the relative path ./out/<name>.xlsx. If mcp__office__excel_* tools are available, use the live Excel workflow instead of this file-producing fallback.
Compared to similar skills
The skill explicitly distinguishes itself from mcp__office__excel_* live Excel tools: xlsx-author produces a disk-based .xlsx artifact, while those tools drive a user's live workbook.