CSV Whisperer
Open, fix and clean any messy CSV — broken, old or multi-file — then compare, reconcile and chart it.
SKILL.md in the open agentskills.io standard — works directly in Claude, ChatGPT/Codex, Cursor, Copilot & more.
CSV Whisperer is an AI data skill for tabular data that opens and fixes any broken or messy CSV: it auto-detects delimiter, encoding and column types, repairs garbled characters, profiles the data, compares and merges files, spots anomalies, and builds the right chart. Two-stage: quick triage first, then a clean read.
What this skill does
- Two-stage: quick triage first (format/encoding/age/structure/risks), then a clean read
- Detects junk preambles, footer/total rows, stacked tables & disguised/corrupt files
- Old/legacy formats: ISO-8859, Windows-1252, Mac-Roman, CR-only, fixed-width, dBase/.xls
- Reliably detects dialect & encoding (semicolon, decimal comma, UTF-8 BOM, mojibake)
- Profiles columns, preserves precision, warns about Excel mangling & CSV injection
- Multi-file: join, diff two versions, reconciliation, schema drift
- Reshape & export: wide↔long, split/merge columns, CSV→JSON/SQL/Excel (round-trip safe)
- Trust checks: PII detection, k-anonymity (GDPR) + Benford fraud indicator
- Special cases: time-series gaps/duplicates, geodata, unit normalization, long-tail bucketing, DATEV/payroll preset
- Multi-valued cells: untangled only on request — flags it with a real example and asks
- Builds the right, honest chart for the question
- Cleaned RFC 4180 CSV with a change log + data contract (Frictionless Table Schema)
Description
CSV Whisperer is a skill for reading and cleaning messy CSV files — it auto-detects the delimiter, encoding, and column types of any CSV (even old, legacy or broken exports), profiles the data, reconciles it, and turns it into charts. A data Swiss Army knife for tabular data. Two-stage: first a short triage (format, encoding, line endings, age/origin, structural quirks, risks, parse plan), then a clean read — so no file gets wrecked by wrong assumptions. Handles German Excel dialects (semicolon + decimal comma), encoding repair (UTF-8 BOM, Latin-1, mojibake), old/legacy formats (ISO-8859/Windows-1252/Mac-Roman, CR-only, fixed-width, dBase/.xls, two-digit years, accounting numbers), junk preambles/footer/total rows, stacked tables and disguised/corrupt files. Plus multi-file (join, diff, reconciliation, schema drift), reshape & export (wide↔long, CSV→JSON/SQL/Excel, round-trip safe), trust checks (PII, k-anonymity, Benford) and special cases (time-series gaps, geodata, unit normalization, long-tail bucketing, DATEV/payroll preset). Multi-valued cells are never split automatically — it asks first with a real example. Profiles columns, preserves precision, warns about Excel mangling and CSV injection, delivers honest charts, a cleaned RFC 4180 CSV with a change log or a data contract (Frictionless Table Schema). Invents no values and changes nothing silently.
Examples
What it does not do
- Invents no values and changes nothing silently (always a change log)
- Won't auto-split multi-valued cells — asks first with a real example
- Won't guess separator/encoding or ambiguous centuries without saying so
- Loses no leading zeros / precision; no misleading charts
- Benford is an indicator, not proof; won't spread PII unasked; not a full database/ETL
Compatibility & tech
- Tested (internal)
- 3 scenarios
- Recommended runtime
- A model with a code/analysis tool for exact parsing and real charts.
- Modes
- Triage · Understand/Profile · Clean · Chart · Q&A · Schema/Contract · Multi-file · Reshape/Export · Audit (PII/Fraud) · Special cases
- Inputs
- CSV · TSV · Excel export · Fixed-width · Legacy export · Multiple files · Pasted table
- Output format
- Quick triage + data profile + insights; on request a chart, cleaned CSV, join/diff, reshape/export, trust check or Table Schema
- Subcategory
- Data analysis / visualization
- License
- Proprietär
Security profile
Runs entirely on your machine with your own AI — no external runtime, no running costs.
Only instructions, templates and references — no executable scripts.
Works offline with what you provide — does not call external services on its own.
What you get
- csv-whisperer-1.6.0/4 files
- .claude-plugin/marketplace.json
plugins/csv-whisperer/skills/csv-whisperer/4 files
- SKILL.md
- manifest.json
references/2 files
- csv-whisperer-checklist.md
- recipes.md
- .agents/skills/csv-whisperer/→ universal — same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
- LICENSE.txt
Installation
Also works as a chat prompt
No AI tool? Paste it into Claude, ChatGPT or Gemini and use the method right away.
This skill ships no scripts, so the prompt carries its full method.
Installing is the full version — it triggers automatically, runs its scripts and loads references as needed. As a chat prompt you drive the method by hand.
Unlock to copy the ready-to-paste prompt — then in “My Skills”.
Reviews
No reviews yet – be the first.
Note
Most accurate with a code/analysis tool (Claude Analysis, ChatGPT Data Analysis); without one it works from a sample and flags that.
Changelog
- v1.6.022.06.2026Platform-native output: now actively uses your platform's native formatting — tables, checklists, code blocks and clickable/button options, with a numbered fallback on non-interactive platforms. Plus internal metadata that prepares the upcoming agent builder.
- v1.5.018.06.2026Reworked to the AUTHORING v2 standard: realistic input/answer examples (incl. a calibration case), copy checklist(s) and internal evals; cross-references added.
- v1.4.018.06.2026Added references/recipes.md: deeper detail reference via progressive disclosure; SKILL.md stays lean.
- v1.3.010.06.2026Multi-valued cells only on request (ask with a real example); special-cases module (time series, geodata, unit normalization, long-tail bucketing, DATEV preset); k-anonymity added.
- v1.2.010.06.2026Modules: preamble/footers/stacked & disguised files; multi-file (join/diff/reconciliation/schema drift); reshape & export; trust checks (PII + Benford).
- v1.1.010.06.2026Two-stage quick triage; old/legacy module (legacy encodings, CR-only, fixed-width, dBase/.xls, two-digit years, accounting numbers).
- v1.0.010.06.2026Robust dialect/encoding/type detection (incl. DE locale & mojibake), profiling, anomalies, honest charts, cleaning with a log, Frictionless schema, Excel/CSV-injection warnings.
Frequently asked questions
What does CSV Whisperer do?
CSV Whisperer is an AI data skill for tabular data that opens and fixes any broken or messy CSV: it auto-detects delimiter, encoding and column types, repairs garbled characters, profiles the data, compares and merges files, spots anomalies, and builds the right chart. Two-stage: quick triage first, then a clean read.
How do I open a broken or corrupted CSV file?
CSV Whisperer reliably opens even old, broken, or multi-part exports with junk header rows: it auto-detects the delimiter, encoding, and column types and works in two steps — a quick triage first, then a clean read.
How do I fix CSV encoding problems and garbled characters?
CSV Whisperer auto-detects your file's encoding (UTF-8, Latin-1, Windows-1252) and reads it correctly, so you stop seeing garbled characters or columns split by the wrong delimiter.
How do I compare or merge two CSV files?
CSV Whisperer joins, diffs, and reconciles multiple CSV files, surfaces mismatches and anomalies between them, and then reshapes the data for export.
How do I make a chart from CSV data?
CSV Whisperer profiles your CSV, finds the patterns in it, and builds the right chart on request — straight from the data it read in.
Which AI tools does CSV Whisperer work with?
Claude (Projects & system prompt) + Analysis tool · ChatGPT (Custom GPT & Data Analysis) · Any LLM (pasted data)
How do I use CSV Whisperer?
CSV Whisperer is a SKILL.md in the open agentskills.io standard: install it with one command (npx) or download it and add it to your AI tool — Claude (Projects), ChatGPT (Custom GPT), Cursor, Copilot, Gemini CLI and more. No code needed.