Book a call GitHub 19

Use Nanook with AI agents

A coding agent can draft a Nanook table and the script that turns it into test data. It does that well when it has three things: a skill that knows how a decision table is built, a few rules that keep the generation script correct, and the documentation as plain text. This page lists all three for Claude Code and for other agents. None of it replaces checking what the agent produced; the last section says how.

1 · Claude Code: the skill

Since skill version 0.2.0 the skill create-equivalence-class-table ships as a Claude Code plugin, straight from the GitHub repository. Install it once in Claude Code:

/plugin marketplace add xhubio/nanook-table
/plugin install nanook@nanook

Then, in a project with Nanook and exceljs installed (in a new project, start with npm init -y and npm pkg set type=module):

npm install @xhubio/nanook-table
npm install -D exceljs
claude
/nanook:create-equivalence-class-table Login Form

The skill copies two scripts into your project: check-classes.mts recounts the coverage and reports every class without a test case of its own, generate-fixtures.mts runs Nanook and writes one JSON per test case. What a run produces, how long it took and what to watch for is on Quickstart with Claude Code. Without the plugin, copy the skill from the package (any version after 3.0.1): mkdir -p .claude/skills && cp -r node_modules/@xhubio/nanook-table/skills/create-equivalence-class-table .claude/skills/.

2 · Other agents

The skill is a plain folder with a SKILL.md in the Agent Skills format. One command installs it for Codex, Cursor, GitHub Copilot, Gemini CLI and other agents that read that format:

npx skills add xhubio/nanook-table --skill create-equivalence-class-table

Without --skill the command also offers the repository’s own development skills. We have run the skill in Claude Code only and checked the install for Codex, so treat other agents as untested. The rules block in the next section does not depend on skills and works in any agent that reads AGENTS.md.

3 · A rules block for AGENTS.md

AGENTS.md is the instruction file that Codex, Cursor, GitHub Copilot and other agents read from the project root. Paste this block into it. The markers let you replace the block later without touching the rest of the file:

<!-- BEGIN:nanook-agent-rules -->
## Nanook: test cases and test data

Test cases are defined in XLSX workbooks and turned into test data with
@xhubio/nanook-table (Node.js 22 or newer, ESM).

- Before writing a table or a generation script, read the docs for the
  installed version: node_modules/@xhubio/nanook-table/docs/
  (guide/decision-tables.md, guide/directives.md, api/processor.md).
  Website index: https://nanook.xhub.io/llms.txt
- Every generator a table calls (gen:<instanceId>:<generator>:<parameter>)
  must be registered in the DataGeneratorRegistry. The built-in
  GeneratorFaker takes a Faker path and no arguments; anything else
  needs its own generator.
- Pass the tables to TestcaseProcessor keyed by table name, not as the
  array from FileProcessor, or every ref: fails.
- In 3.0.1 the default writer throws in before(); use an inline
  InterfaceWriter.
- After generating, compare the number of test cases with the number of
  test-case columns (plus one per extra element of a range reference).
  Fewer means a generator failed: Nanook logs the error and goes on.
<!-- END:nanook-agent-rules -->

Claude Code reads CLAUDE.md. Put the line @AGENTS.md into it and Claude Code imports the same rules, so both files never drift apart.

The block points the agent to the Markdown documentation inside the package. Unlike the pages on this site, it always matches the installed version: the guide and tutorials here describe the table concepts, and the 3.x API reference is under API. Rules two to four are mistakes that cost a run its test cases on 3.0.1; the last one is how you notice. The quickstart describes them under Generate the test data and What can go wrong.

4 · The docs as plain text

For a chat window, or an agent that fetches URLs:

  • /llms.txt is the index: every docs page with one line on what it covers, in the llms.txt format.
  • /llms-full.txt is the whole documentation in one Markdown file, to paste or attach.
  • Every quickstart, guide, tutorial and module page and the 3.x API reference has a Markdown version: add .md to its address, for example /docs/quickstart/quickstart.md. The Copy as Markdown button at the top of those pages puts it on the clipboard. The 1.x API pages have none.
  • Inside a project, the package itself carries the 3.x documentation as Markdown in node_modules/@xhubio/nanook-table/docs/, matching the installed version.

5 · Check what the agent produced

An agent that writes a table and a script will report success. Two checks catch most of what goes wrong. First, open the workbook: if the skill built it, its summary row shows the coverage per sheet. Second, count: the script must report one test case per test-case column, plus one for every extra element of a range reference. Fewer means a generator failed on the way, and Nanook logs that and keeps going. The quickstart explains the count under Generate the test data.

Not yet

There is no Nanook MCP server yet. An agent with a terminal does not need one: it runs the generation script itself.

Where to go next
← Create filter processorNanook Table Overview →