Analyze a CSV in Cowork - and check the numbers

Anthropic's own examples of Cowork output include Excel spreadsheets with functional formulas and synthesized research (October 2026 docs). A CSV is squarely in that territory. Here is the documented picture - and the checking habits that matter more than the tool.

What the docs say it can do

Cowork works with your files and produces finished ones: documents, organized files, spreadsheets with working formulas. The Data plugin adds dataset querying and visualization skills on top. So "analyze this CSV" is a documented use - the shape is: give it the file, ask a specific question, get an answer plus the artifact.

The habits that keep it honest

AI arithmetic fails quietly, so verification is the workflow:

1. Ask for the method, not just the answer. "Show how you computed the total" surfaces wrong-column and wrong-filter mistakes before they cost you.
2. Spot-check with numbers you know. A row count, a sum of one column you can verify in a spreadsheet in ten seconds. If the anchor numbers are right, the rest earns trust faster.
3. Check one edge. An empty cell, a currency column, a date in a different format - ask how each was handled. The edge cases are where CSV analysis actually breaks.
4. Keep the artifact. A spreadsheet with functional formulas (the documented output) lets you re-run the math yourself; a pasted table does not.

When to add the Data plugin

For a one-off question, plain Cowork is enough. If CSV analysis is weekly work, the Data plugin's query and visualization skills are built for it - contents in the starter shortlist and install steps in add a GitHub marketplace.

Sources

• Get started with Claude Cowork - Claude Help Center
• anthropics/knowledge-work-plugins on GitHub