Headless Excel Workbook Builder
Generate structured Excel files in managed-agent sessions without an open Office app.
The skill constrains output to a local ./out/<name>.xlsx file and explicitly avoids driving a live Excel workbook, reducing external effects. However, it requires Bash/Python execution and an unpinned openpyxl dependency, with no sensitive-data handling, user-confirmation, permission-boundary, rollback, or dependency-security guidance, so points are deducted.
The output path, example script, and modeling conventions make the happy path understandable. There are no tests, input validation, failure handling, dependency-installation instructions, or diagnosable error feedback; under static calibration the score remains below 10.
The target managed-agent/CMA headless scenario is clear, and the non-fit live-workbook case is stated. Input scope, non-fit boundaries, trigger precision, Chinese-language support, and environment compatibility are not defined, so points are deducted.
The document provides purpose, an output contract, an example, conventions, and an alternative-tool path; the repository also supplies Apache-2.0 licensing and basic CI. Skill-specific versioning, changelog, maintenance ownership, installation/troubleshooting guidance, and known limitations are missing, so points are deducted.
It can produce a directly collectible .xlsx artifact and gives useful conventions for inputs, formulas, links, and checks. However, there is only a minimal example and no representative-output verification, complex-workbook coverage, or comparative-benefit evidence; static calibration limits the score to 7 or below.
The skill text is auditable, and repository CI validates plugin structure. There is no committed test suite covering the key workbook-generation paths and no third-party execution evidence, so the conclusion relies mainly on static source review and is scored conservatively.
- The deployer should pin and review Bash/Python and openpyxl dependencies and restrict filesystem writes.
- Handling, redaction, logging, and cleanup policies for sensitive financial data are unspecified.
- Generated formulas, formatting, and Checks results still require human review; repository CI does not verify workbook correctness.
What it does & when to use it
xlsx-author is designed for headless Managed Agent or CMA sessions that must deliver an Excel workbook as a file artifact. It uses Python and openpyxl to create .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 checks. It fits automated workflows that need collectible workbook files, but not workflows centered on editing an already open workbook.
Creates ./out/ when needed; writes and runs a short Python script; uses openpyxl to create worksheets, write inputs, and add formulas; and saves the workbook to ./out/<name>.xlsx. It specifies blue formatting for hardcoded inputs, black for formulas, and green for links to another sheet or file. It also requires formulas in calculation cells, named ranges for values referenced from a deck or memo, a Checks sheet exposing TRUE/FALSE results, and one model per file.
- A managed-agent workflow must produce a financial model in an environment without an open Office application.
- A team needs an .xlsx artifact for an orchestration layer to collect rather than edits to a live Excel workbook.
- A modeling process needs inputs centralized on an Inputs tab and calculations driven by formulas.
- An automated workbook-generation task needs balance checks for statements, cash flow, or related model ties.
Pros & cons
- Defines a clear output contract that orchestration layers can collect.
- Provides concrete conventions for inputs, formulas, links, named ranges, and checks.
- Designed for environments without an open Office application.
- Requires formula-driven calculation cells and exposes checks as TRUE/FALSE.
- Produces a workbook file on disk and does not replace live Excel workbook editing.
- Requires Python, Bash, and openpyxl; the source gives no version pinning or lockfile.
- The source provides no test suite or platform verification record.
- Enforces one model per file and does not append to an existing workbook unless explicitly requested.
How to install
The source does not document a standalone installation command for xlsx-author. It is located inside the KYC Screener plugin in the anthropics/financial-services repository. The README documents collection-level installation by opening Cowork Settings → Plugins → Add plugin and pasting https://github.com/anthropics/financial-services, then selecting the desired agents or verticals; a zip of a directory under plugins/ may also be uploaded. No standalone installation procedure is provided for this skill.
How to use
Invoke the skill in a headless Managed Agent or CMA session with a request such as: “Create an Excel model with Inputs, calculation, and Checks worksheets in the headless environment, and save it to ./out/model.xlsx.” The runtime must be able to execute Bash and Python and have openpyxl installed. The final response should return the relative path ./out/model.xlsx.