Research workflow

CSV to insight report

Profiles a pasted CSV, answers your question with each calculation shown, checks every number against its working, and returns a short report with a chart suggestion.

Prompt chainEvaluator–optimizerRuns in Claude Code, Codex, ChatGPT, Claude.ai

Running it here

Every step runs here with Claude when you're signed in, and in your own agent.

  1. TriggerData and questionStart with {{csv}} and {{question}}.→ manual
  2. PromptProfile the dataProfile {{csv}}: row count, each column's type, missing values, ranges or top values, and anything odd (totals rows, duplicates, mixed units, date formats). Then restate {{question}} as the calculation it needs.
  3. SkillCompute the answerAnswer {{question}} from {{csv}} using the plan in {{profile}}. Show each calculation with the numbers it uses, then a table of results, then caveats. Suggest one chart (type, x, y, grouping).→ csv-analyst
  4. Review loopCheck every numberRecompute every number in {{report}} from {{csv}}. Pass only if each one matches a shown calculation, totals add up, excluded rows are stated, and no sentence claims a cause the data can't show. Fix any wrong number and its sentence.
  5. OutputReturn the reportReturn {{report}} with the chart suggestion.→ chat

Good for

Anyone with an export (sales, survey, usage, ads) and a question about it, who wants the answer with every number traceable to a calculation.

Last reviewed

Skills this uses

  • 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.

Install as a skill

The whole workflow as one SKILL.md. Install it in Claude Code, Codex or Copilot and ask your agent to run “CSV to insight report”.

Get the csv-to-insight-report skill

What each step does

  1. 1. Data and question (Trigger)

    Start with {{csv}} and {{question}}.

  2. 2. Profile the data (Prompt)

    Profile {{csv}}: row count, each column's type, missing values, ranges or top values, and anything odd (totals rows, duplicates, mixed units, date formats). Then restate {{question}} as the calculation it needs.

  3. 3. Compute the answer (Skill)

    Loads the csv-analyst skill.

    Answer {{question}} from {{csv}} using the plan in {{profile}}. Show each calculation with the numbers it uses, then a table of results, then caveats. Suggest one chart (type, x, y, grouping).

  4. 4. Check every number (Review loop)

    Recompute every number in {{report}} from {{csv}}. Pass only if each one matches a shown calculation, totals add up, excluded rows are stated, and no sentence claims a cause the data can't show. Fix any wrong number and its sentence.

  5. 5. Return the report (Output)

    Return {{report}} with the chart suggestion.

Use it in your own agent

Fill in the inputs

Paste up to about 300 rows with the header row. For bigger files, run the exported skill in an agent with a code tool.

Open in ClaudeOpen in ChatGPT2 inputs use their example until you fill them in.

Save as .claude/skills/csv-to-insight-report/SKILL.md for Claude Code, or .agents/skills/csv-to-insight-report/SKILL.md for Codex and other Agent Skills tools. Then ask: “run the CSV to insight report workflow”.

SKILL.md
---
name: csv-to-insight-report
description: Profiles a pasted CSV, answers your question with each calculation shown, checks every number against its working, and returns a short report with a chart suggestion. Use when asked to run the "CSV to insight report" workflow.
---

# CSV to insight report

Profiles a pasted CSV, answers your question with each calculation shown, checks every number against its working, and returns a short report with a chart suggestion.

## Inputs

Ask for any of these that the user hasn't given:

- `csv` (e.g. "month,channel,signups,paid
2026-08,search,410,52
2026-08,social,380,21
2026-09,search,455,61
2026-09,social,402,19
2026-10,search,498,70
2026-10,social,350,14")
- `question` (e.g. "Which channel converts signups to paid better, and is the gap growing?")

## Steps

Run the steps in order. Refer to saved results by name. Don't skip a step that says STOP.

1. **Data and question** — Start: Start with {{csv}} and {{question}}.
2. **Profile the data** — Profile {{csv}}: row count, each column's type, missing values, ranges or top values, and anything odd (totals rows, duplicates, mixed units, date formats). Then restate {{question}} as the calculation it needs. Save the result as `profile`.
3. **Compute the answer** — Load the `csv-analyst` skill, then: Answer {{question}} from {{csv}} using the plan in {{profile}}. Show each calculation with the numbers it uses, then a table of results, then caveats. Suggest one chart (type, x, y, grouping). Save the result as `report`.
4. **Check every number** — Review loop (max 3 passes): Recompute every number in {{report}} from {{csv}}. Pass only if each one matches a shown calculation, totals add up, excluded rows are stated, and no sentence claims a cause the data can't show. Fix any wrong number and its sentence. Save the result as `report`.
5. **Return the report** — Finish: Return {{report}} with the chart suggestion.

## Done when

Every step has run, each approval was given, and you've reported what was produced and where it was saved.

Want different steps? Open it in the workflow builder.