Excel Data Cleaning Assistant
Clean inconsistent spreadsheets before analysis.
The skill requires a pre-change issue summary, user confirmation for destructive actions, and preferably helper-column formulas, providing some least-change, transparent data flow, and auditability. Points are deducted because sensitive financial-data handling, access boundaries, backup or rollback, and external data flow are unspecified; the openpyxl dependency is unpinned and has no security guidance; recovery boundaries are unclear.
The workflow is organized into scoping, detection, proposal, application, and change reporting, and it lists common anomaly classes, making the happy path plausible. Points are deducted because the skill contains only prose instructions with no tests or sample inputs and outputs; key Office JS and openpyxl paths are incomplete; rules and failure feedback for near-duplicates, date parsing, mixed-type repair, and abnormal input are thin. The static calibration cap keeps this below 10.
The name, intended scenario, trigger phrases, and two environments—Excel and standalone xlsx—are reasonably clear for spreadsheet-cleaning preparation. Points are deducted because non-fit boundaries, output formats, header assumptions, and complex-workbook limits are absent; Chinese content, regional date formats, and mainland-China environment fit are not addressed, so semantic trigger precision remains uncertain.
The skill has a readable progressive workflow and issue table; the repository provides Apache-2.0 licensing, installation guidance, contribution paths, and CI validation signals. Points are deducted because the skill itself lacks versioning, changelog, named ownership, and an update path, and it omits dependency installation notes, complete examples, FAQs, known limitations, and troubleshooting guidance.
The skill addresses common cleaning needs including blanks, whitespace, casing, numeric text, dates, duplicates, encoding problems, and spreadsheet errors, while requiring before/after summaries and staged confirmation. Points are deducted because there is no executable implementation, representative output, or verification evidence; formula coverage is limited and results still require human review. Static calibration limits this to 7.
The repository includes plugin validation, secret scanning, and read-only CI workflows, while the license and repository provenance are traceable. Points are deducted because these checks do not cover the skill's cleaning logic, and there are no skill-level tests, fixed fixtures, third-party execution results, or corroborating sources. This supports only limited auditability and is capped at 5 under static review.
- Before processing personal, client, or transaction data, clarify whether data leaves the local environment, who can access it, and how backup and rollback work.
- Removing duplicates, filling blanks, and overwriting originals can change business meaning; preserve a versioned copy and define duplicate rules before confirmation.
- DATEVALUE, VALUE, casing normalization, and near-duplicate detection can depend on locale and business semantics; Chinese dates, amounts, percentages, and encoding anomalies require extra validation.
- The skill provides no reproducible fixtures or executable tests, so correctness cannot be assumed from the documentation alone.
What does this skill do, and when should you use it?
This skill cleans messy data in an active Excel sheet or a specified range. It checks whitespace, casing, numbers stored as text, date formats, duplicates, blanks, mixed types, encoding problems, and spreadsheet errors. It profiles dominant column types and outliers, then shows a proposed-fix summary before making changes. It prefers helper-column formulas for transparent, auditable transformations and asks for confirmation before destructive operations.
Reads the active sheet or a specified range and profiles each column's dominant type and outliers. It detects leading or trailing whitespace, double spaces, inconsistent categorical casing, numeric text, currency or comma or percent characters in number cells, mixed date formats, exact and near-duplicate rows, unexpected blanks, mixed-type columns, mojibake, non-printing characters, and common Excel errors. It produces a summary with the column, issue, count, and proposed fix, then can apply helper-column formulas such as TRIM, VALUE, SUBSTITUTE, UPPER, and DATEVALUE. It reports a before-and-after summary of changes.
- A financial analyst receives an inconsistently formatted Excel export and needs to standardize numbers, dates, and categorical fields before modeling.
- An operations user finds duplicate customer or transaction rows caused by whitespace or casing differences and needs exact and near-duplicate detection.
- A researcher has numeric text containing currency symbols, commas, or percentages and needs values converted for analysis.
- An Excel user inherits a worksheet containing mojibake, blanks, and formula errors and wants an issue inventory before editing.
- An audit-conscious user wants cleaning transformations preserved in helper columns instead of overwriting the original data.
What are this skill's strengths and limitations?
- Covers common spreadsheet problems including whitespace, casing, numeric text, dates, duplicates, blanks, mixed types, encoding, and errors.
- Shows an issue summary and proposed fixes before editing.
- Prefers helper-column formulas, making transformations more transparent and auditable.
- Provides separate guidance for Excel Office JS and standalone .xlsx files.
- Duplicate removal, blank filling, and overwriting originals require confirmation, which can slow large cleanup jobs.
- No standalone command, test suite, or specific version requirements are documented.
- The precise near-duplicate matching algorithm is not specified, and formulas are not provided for every issue type.
- The repository states that its outputs are not investment, legal, tax, or accounting advice and require professional review.
How do you install this skill?
In Cowork, open Settings → Plugins → Add plugin, paste https://github.com/anthropics/financial-services, and select financial-analysis. In Claude Code, run: claude plugin marketplace add anthropics/financial-services; claude plugin install financial-analysis@claude-for-financial-services. The README does not provide a separate installation command for clean-data-xls.
How do you use this skill?
With the financial-analysis plugin installed, provide the active worksheet or a range and use a trigger such as “clean this data,” “clean up this sheet,” “normalize this data,” “dedupe,” or “standardize this column.” Example: “Clean the data in A1:F200, show me the issue summary first, and ask before removing duplicates.” For a standalone .xlsx file, the skill specifies Python/openpyxl; inside an Excel Office Add-in / Office JS environment, it specifies Office JS.