Guide · Agent skill

How to use the Arize Dataset Manager agent skill

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

The arize-dataset skill is a set of instructions that teaches an AI agent how to work with datasets in Arize, a platform for testing AI systems. A dataset here is a versioned collection of test records, called examples, used to evaluate models. This guide explains what the skill does, how to install it, how to phrase requests so the agent uses it, and what you need before it works.

Arize Dataset Managergithub/awesome-copilot · MIT Download the skill

Quick start

1Get an Arize account and API key. Sign in to Arize, make sure you have access to a space, and create an API key under the admin API Keys page.
2Install the ax CLI. Install arize-ax-cli with uv, pipx or pip, then confirm the version is 0.14.0 or higher with ax --version.
3Configure your profile. Set ARIZE_API_KEY in your shell, then check your setup with ax profiles show.
4Find your space. Run ax spaces list and note the name or ID of the space you want to use.
5Install the skill. Unzip the arize-dataset folder into ~/.claude/skills/ or a project's .claude/skills/, or upload it under Settings → Capabilities → Skills.
6Make a request. Ask the agent in plain language, for example to list datasets in your space or export a golden dataset to JSON.

What is an agent skill and what does this one do?

An agent skill is a folder with a SKILL.md file that tells an AI agent how to handle a specific kind of task. When the task matches the skill's description, the agent reads the instructions and runs the commands they describe.

Once the arize-dataset skill is installed, the agent knows how to list, look up, create, export and append examples to Arize datasets. It does this by running the ax command-line tool on your behalf. The skill covers dataset CRUD (create, read, update and delete), appending examples, exporting data, and creating datasets from local files.

List datasets in a space, with pagination
Look up dataset metadata and its version list
Export all examples, or a specific version, to a local file
Create a dataset from a CSV, JSON, JSONL or Parquet file, or from piped data
Append examples to an existing dataset
Troubleshoot login and setup problems using the skill's reference files

What do you need before installing?

The skill's requirements are listed in its compatibility note: the ax CLI and a configured Arize profile. The skill is also designed around an Arize account with access to a space, and every command needs an API key. Without an account and key the skill cannot be used.

A space is an organizational container, and datasets belong to a space. You can pass a space by name, such as my-workspace, or by base64 ID. Run ax spaces list to find yours.

An Arize account with access to at least one space
An Arize API key (in the Arize app, go to https://app.arize.com/admin and open API Keys)
The ax CLI, version 0.14.0 or higher, installed on your machine
A configured ax profile that holds your API key

How do you install the arize-dataset skill?

Download the ZIP of the skill from the awesome-copilot page on this site and unzip it. For a personal install, place the folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

In the Claude apps you can instead upload the skill under Settings → Capabilities → Skills.

Personal install: unzip into ~/.claude/skills/
Project install: unzip into the project's .claude/skills/ folder
Claude apps: Settings → Capabilities → Skills, then upload the skill

How do you set up the ax CLI and profile?

The skill tells the agent to run ax commands directly and only check setup when something fails. You can also set things up yourself first. If ax is not found, the troubleshooting guide suggests installing it with uv tool install arize-ax-cli (preferred), pipx install arize-ax-cli, or pip install arize-ax-cli.

To inspect your profile, run ax profiles show. The profile reference says to pass your key through the ARIZE_API_KEY environment variable rather than as a raw flag, so set that variable in your shell before you configure the profile.

Check the install with ax --version (needs 0.14.0 or higher)
Inspect the profile with ax profiles show
Find your space with ax spaces list

ax --version
ax profiles show
ax spaces list

How do you trigger the skill with a natural request?

You do not need to type a command. The skill loads when your request matches its description, so mention test data, evaluation examples, or use words such as create dataset, list datasets, export dataset, append examples, dataset version, golden dataset or test set.

Name the dataset and space when you can. The agent will ask for the space or project if it cannot work them out, and it will offer a list to pick from if you are unsure.

"List the datasets in my-workspace."
"Export the golden dataset for the support bot to a JSON file."
"Append these five question and answer examples to the rag-eval dataset."
"Create a dataset called hallucination-test from ./data/tests.csv in my-workspace."

What commands does the agent run?

The agent uses the commands below. Export writes a file named after the dataset and a timestamp, and it can also print to stdout so you can pipe it into other tools.

Creating a dataset needs a name, a space and a data file. Piping data through --file - avoids a temporary file. Append uses --json instead of a file.

ax datasets list --space SPACE --limit 20 (limit is 1 to 100, default 15)
ax datasets get NAME_OR_ID --space SPACE -o json
ax datasets export NAME_OR_ID --all --output-dir ./data
ax datasets create --name "My Dataset" --space SPACE --file data.jsonl
ax datasets append --json '[...]'

ax datasets list --space SPACE --limit 20
ax datasets export NAME_OR_ID --space SPACE
ax datasets export NAME_OR_ID --all
ax datasets create --name "My Dataset" --space SPACE --file data.csv
echo '[{"question": "What is 2+2?", "answer": "4"}]' | ax datasets create --name "my-dataset" --space SPACE --file -

What should you know about file formats and exports?

Choose your file format with care. CSV loses type information: dates become strings and null becomes an empty string. JSON and Parquet preserve types. A JSONL file must hold one object per line; a JSON array in a .jsonl file will fail, so use the .json extension for arrays. Parquet requires pandas and pyarrow.

The skill also includes two export rules. If an export returns exactly 500 examples, the agent re-runs it with --all, because the result is likely truncated. After exporting, the agent compares the row count with the count the server reports.

CSV: column headers become field names
JSON: an array of objects
JSONL: one object per line, not an array
Parquet: column names become field names and types are kept
System fields id, created_at and updated_at are generated by the server, so the skill never includes them when creating or appending

What safety points should you check?

This skill runs scripts and uses the network. The ax commands send requests to Arize, so a valid API key is needed and the data you send leaves your machine. The skill's instructions also tell the agent never to read .env files or search the filesystem for credentials. Arize credentials come from ax profiles, and LLM provider keys come from ax ai-integrations.

Exported examples are written to local files. They may contain private content, so do not share or commit them to a repository without checking first.

The skill runs scripts and uses the network
Keep your API key out of commands; reference the ARIZE_API_KEY variable
Review exported files before sharing or committing them
The skill only targets Arize spaces, not datasets stored elsewhere

What people use it for

Pull a golden dataset to a local JSON file for offline review
Bulk-load new question and answer examples into a test set without writing CLI flags
Create a dataset from a CSV of evaluation cases
Check a dataset's versions and latest example count before a model comparison
Troubleshoot a 401 Unauthorized error by inspecting the ax profile

Frequently asked questions

Is the arize-dataset skill free to use?

The skill is released under the MIT licence. Using it still requires an Arize account with an API key and access to a space, and the ax CLI must be installed. Check Arize's own terms for account costs.

What do I need to install the arize-dataset skill?

You need the ax CLI (version 0.14.0 or higher), a configured Arize profile with an API key, and an Arize account with access to a space. The skill cannot run without an API key.

Where do I put the arize-dataset skill folder?

Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder for one project. In the Claude apps, upload it under Settings → Capabilities → Skills.

How do I fix a 401 Unauthorized error with ax?

Run ax profiles show to check the current profile. If the key is missing or wrong, follow the profile reference to update it using the ARIZE_API_KEY variable. You can get a key from app.arize.com under admin API Keys.

Why does my export only return 500 examples?

A result of exactly 500 examples is likely truncated. Re-run the export with --all, which performs an unlimited bulk export, then compare the row count with the server's count.

Can the skill work with datasets outside Arize?

No. The skill only targets Arize spaces. For datasets stored elsewhere, you would need a different tool or workflow.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you keep evaluation or golden test sets in Arize and want an agent to list, export or bulk-load examples without memorising ax flags. Pick something else if you do not have an Arize account, need datasets stored outside Arize, or cannot safely handle exported files that may contain private data.

Download the skill