Agent skill

Your agent will tell you the data looks fine

It cannot measure the data it is describing, so it guesses and you find out later. This gives it checks it runs itself—on your machine, on your files, before it answers.

Choose an agent, then Open.

Node 20+ to install Python 3.8+ for the helpers

Install

Install the data-validation-rules skill with npx govalidate-lite, then ask me what I need to check.
Open in Claude

Cursor's documented link fills this sentence in. Claude, ChatGPT, and Grok are best-effort (q in the address); Copy if the composer is empty. Works anywhere the agent has a shell—desktop, web, mobile, IDE.

12 → 5 wrong claims across 12 blind-graded runs Same prompt and agent on 6 messy CSVs, scored blind against a fixed key. Without the skill: 12 wrong claims, 58/60 problems found. With it: 5 wrong claims, 57/60.

It reads your files. That’s all it does.

  • Never leaves your machine No upload, no account, no API key.
  • Never touches the original The helpers only read. A repair writes a new file beside yours.
  • Nothing you can’t read Five Python scripts, standard library only. Open them, change them.

There is a Pro skill

It adds the deterministic engine behind these checks. Not publicly available yet.

Frequently Asked Questions

What is the GoValidate agent skill?

The GoValidate agent skill (data-validation-rules) equips AI coding agents with deterministic, local checks for CSV and tabular data so they verify their own work before presenting answers.

How do I validate a spreadsheet before my Cursor or Claude agent uses it?

Run npx govalidate-lite in your project (or open https://govalidate.ai/skill/), install the data-validation-rules skill, then ask the agent to check the file before it answers. Checks run locally; the skill does not upload your data.

Is GoValidate Lite the same as govalidate.dev?

No. This skill is GoValidate Lite at govalidate.ai. Not affiliated with govalidate.dev (startup idea validation) or go-validate.com (digital identity / LA Wallet).

How do I install the data-validation-rules skill?

Run npx govalidate-lite in your project directory. It installs five read-only Python verification scripts directly into your agent directory (.claude/skills, .cursor/skills, or standard skill roots) without external dependencies.

Which coding agents support the GoValidate skill?

The skill works natively with Claude Code, Cursor, Codex, Grok, and any agent system supporting local skill manifests or CLI tool execution.

Does the agent skill upload files or data to external servers?

No. The skill runs entirely locally on your machine using standard Python libraries. No files, row contents, or API keys are ever transmitted over the network.

What checks does the skill perform on CSV data?

The skill validates column counts, detects delimiter consistency, checks for duplicate rows and primary key uniqueness, profiles data types, and flags unescaped newlines or corrupted records.

How was the skill tested?

In trial 16 (October 2026), a fresh Claude Opus agent got the same prompt for each of 6 messy CSV files: find what is wrong before the file is loaded. Each file got 2 runs with the skill and 2 without, 12 runs each way. A separate agent scored every answer blind against a fixed answer key of 60 planted problems, without knowing which answers used the skill.

What did the trial find?

With the skill, the 12 answers contained 5 wrong claims and found 57 of the 60 planted problems. Without it, they contained 12 wrong claims and found 58 of 60. Both found nearly every problem; the difference was in what they made up. That gap held at 95% confidence.

What did the trial not show?

It showed no reliable difference in cost or in how long the agent worked, so this page claims neither. Twelve runs per arm is a small sample, the answers were scored by one blind scorer, and the six files were easy enough that every arm found nearly every planted problem. The method, the answer keys and the per-run results are kept with the trial.