Agent skill · Research
CSV analyst
Answers questions about a CSV or spreadsheet: profiles the columns first, computes with code when a code tool is available and otherwise shows each calculation, states caveats about missing or odd data, and returns findings with a chart suggestion. Use when asked to analyze, summarize or chart tabular data, or to answer a question from a CSV.
The csv-analyst/ folder with SKILL.md.
The skill itself. Agents read the name and description first and load the rest only when a task matches.
csv-analyst/SKILL.md
---
name: csv-analyst
description: "Answers questions about a CSV or spreadsheet: profiles the columns first, computes with code when a code tool is available and otherwise shows each calculation, states caveats about missing or odd data, and returns findings with a chart suggestion. Use when asked to analyze, summarize or chart tabular data, or to answer a question from a CSV."
---
# CSV analyst
Answer the question from the data, and show enough working that someone can check every number.
## Steps
1. Profile the data before answering: row count, each column's type, how many values are missing, the range or top values, and anything odd (duplicates, totals rows, mixed units, dates in two formats).
2. Restate the question as the calculation it needs ("average order value per month = sum of total / count of orders, grouped by month of order_date"). If the question is ambiguous, pick the most likely reading and say so.
3. Compute. With a code tool, use it and keep the code short and readable. Without one, show each calculation with the rows or sums it uses; for more than about 200 rows, say the result is approximate unless you can sum exactly.
4. Check: re-add a total another way, or spot-check three rows by hand. Note any number you couldn't check.
5. Write the answer: the direct answer in one or two sentences, then a short table of the numbers behind it, then caveats.
6. Suggest one chart that answers the question (type, x, y, grouping) and, if useful, a Vega-Lite spec for it.
## Rules
- Every number in the answer comes from a shown calculation. No estimates presented as results.
- Don't silently drop rows. Say how many were excluded and why.
- Correlation is not cause: say "moved together", not "drove".
- Keep personal data out of the answer unless the question needs it.