Domino skills reference

A Domino skill is a packaged set of instructions that teaches a coding assistant how to carry out a specific Domino platform task, such as submitting a Job, registering a model, or deploying an App. You describe the task and the assistant selects the matching skill, so you work in terms of the capability rather than the API. Skills are optimized for Claude Code. Copilot and Codex run in Domino as general-purpose coding agents, but they do not reliably pick up Domino-specific skills.

What the plugin ships

Skills are published in the public domino-claude-plugin repository, alongside several other Claude Code components:

Component	What it provides
Skills	Roughly 23 skills covering Jobs, Flows, Datasets, environments, model endpoints, monitoring, governance, tracing, and the Domino SDKs.
Slash commands	`/domino-app-init`, `/domino-debug-proxy`, `/domino-experiment-setup`, and `/domino-trace-setup`. Invoked by name, unlike skills.
Agent definitions	Preconfigured agents for debugging, deployment, and Project setup.
Output styles	Alternative response styles for MLOps and for learning Domino.
Example hooks	Sample Claude Code hooks. Nothing is active until you copy one into your own configuration.
Bundled MCP server	A Model Context Protocol (MCP) server that runs Domino Jobs, checks Job status and results, and syncs files with DFS-based Projects.

The repository is public, carries an MIT license, and accepts pull requests. See Contribute a skill.

Preinstalled in Workspaces

The Domino Standard Environment (DSE) ships Claude Code, Codex, and GitHub Copilot, together with the Domino Claude plugin and its skills, so a Workspace needs no setup:

Agent	Loads Domino skills
Claude Code	Yes
Codex	Not by default
GitHub Copilot	Not by default

Codex and Copilot handle general coding work in a Workspace, and Troubleshoot coding assistants records the current limitation on Domino-specific skills for both. Which agents are available in which Workspace IDE is covered by Use AI coding assistants. To authenticate any of the three once and reuse the credential in every Workspace, see Reuse credentials. Skills load when a coding assistant starts inside a Domino Workspace, so there is nothing to install or configure. Describe what you want done:

Train a model on the diabetes dataset and track the experiment.

Plugin skills do not appear in the /skills list. Claude Code activates them from the context of your request and reports the loaded set in its startup message.

For the catalog of skills the DSE provides, see the skills catalog.

Use the skills elsewhere

Each skill is a directory under skills/ holding a SKILL.md file and, where a skill carries more detail than one file should, supporting Markdown files beside it. That content is plain Markdown, so it travels to places Claude Code’s plugin mechanism does not reach:

Component	Outside Claude Code
Skills, meaning `SKILL.md` and its supporting Markdown	Portable. Any agent that reads instruction files from your Project or home directory can consume the text as written.
Bundled MCP server	Portable. It is a standalone Python program that speaks the Model Context Protocol, so any MCP client can run it. The `.mcp.json` file that registers it is a Claude Code convention.
Marketplace and plugin installation	Claude Code only.
Slash commands, agent definitions, output styles, and hooks	Claude Code only. Each one depends on a Claude Code feature that has no cross-agent equivalent.

Portable is not automatic. Nothing loads the skill files into Codex, Copilot, or a third-party agent for you, and the mechanism that matches your request to a skill belongs to Claude Code rather than to the Markdown, so another agent needs the file placed where it already looks for instructions. To standardize a team on a different agent, pin agent versions, or bundle extra command-line tools and MCP servers into an image, build a custom Compute Environment. Clone the plugin repository in the environment’s Dockerfile instructions to put the skill files in every Workspace that uses the environment, and pin a commit when you need the set to stay fixed.

Contribute a skill

Anyone can propose a skill by opening a pull request against domino-claude-plugin. CONTRIBUTING.md in the repository is the authority on the process, which runs as follows.

Fork the repository and branch

Fork the repository, clone your fork, and create a feature branch.

Create the skill directory

Add a directory under skills/ and write a SKILL.md file in it. The file opens with a YAML block carrying name and a description that states what the skill does and includes the keywords that should trigger it. Keep SKILL.md under 500 lines and move the detail into supporting files in the same directory.

Add the skill to plugin.json.

Follow the skill authoring standards

CONTRIBUTING.md sets five rules that reviewers enforce:

Fetch a short-lived bearer token from http://localhost:8899/access-token. Never use DOMINO_USER_API_KEY, not even as a fallback, because Domino API keys are deprecated.
Read hosts and identifiers from the environment variables Domino injects, such as DOMINO_API_HOST, DOMINO_REMOTE_FILE_SYSTEM_HOSTPORT, DOMINO_PROJECT_ID, and DOMINO_RUN_ID, rather than writing a hostname or a placeholder into an example.
Show REST calls instead of the python-domino SDK. The domino-data-sdk and python-sdk skills are the exceptions, since they document the SDKs.
Verify every endpoint against the cluster’s own API spec at $DOMINO_API_HOST/assets/public-api.json before you document it.
Run each documented payload against a live deployment and confirm a 2xx response.

Test the plugin locally

Load your branch with claude --plugin-dir /path/to/domino-claude-plugin, confirm the skill triggers, and check that every file and link it references resolves.

Open the pull request

Update the repository’s README.md when you add a feature, record your smoke-test results in the pull request description, and request review from the maintainers.

Install the plugin yourself

Install the plugin directly when you run a coding assistant outside a Domino Workspace, when you need to pin a version, or when you build a custom Compute Environment that standardizes on different tooling. Prerequisites:

Claude Code 1.0.33 or later.
The uv package manager, for the bundled MCP server.
DOMINO_API_KEY and DOMINO_HOST set in your shell. Outside a Workspace, the MCP server has no other way to reach your deployment.

Use the marketplace install unless you are evaluating the plugin or developing against it.

Marketplace, recommended

Registers the plugin through Claude Code’s marketplace so it survives across sessions.

Clone the repository into a marketplace directory

Run the following:

git clone https://github.com/dominodatalab/domino-claude-plugin.git
mkdir -p ~/.claude/marketplaces/domino/.claude-plugin
mkdir -p ~/.claude/marketplaces/domino/plugins
mv domino-claude-plugin ~/.claude/marketplaces/domino/plugins/domino-claude-plugin

Write the marketplace manifest

Save this as ~/.claude/marketplaces/domino/.claude-plugin/marketplace.json. The owner object is required, and source must be a string path rather than a nested object:

{
  "name": "domino-marketplace",
  "owner": { "name": "Domino Data Lab", "email": "support@dominodatalab.com" },
  "plugins": [
    {
      "name": "domino-claude-plugin",
      "description": "Domino Data Lab plugin for Claude Code",
      "version": "1.0.0",
      "source": "./plugins/domino-claude-plugin",
      "category": "development"
    }
  ]
}

From inside Claude Code, run the following. Use an absolute path, because ~ may not expand:

/plugin marketplace add /home/your-username/.claude/marketplaces/domino
/plugin install domino-claude-plugin@domino-marketplace

Plugin directory, for evaluation and development

Loads the plugin from a cloned directory without registering a marketplace:

git clone https://github.com/dominodatalab/domino-claude-plugin.git
claude --plugin-dir ./domino-claude-plugin

To keep it loaded, alias the command:

echo 'alias claude="claude --plugin-dir /path/to/domino-claude-plugin"' >> ~/.bashrc

Declare the marketplace and enable the plugin in the repository’s .claude/settings.json. The path is the marketplace directory you created, not the marketplace name:

{
  "extraKnownMarketplaces": {
    "domino-marketplace": {
      "source": {
        "source": "directory",
        "path": "/home/your-username/.claude/marketplaces/domino"
      }
    }
  },
  "enabledPlugins": { "domino-claude-plugin@domino-marketplace": true }
}

Claude Code prompts each team member to install the marketplace and plugin once they trust the repository folder.

Restart Claude Code after installing, then confirm the plugin appears under the Installed tab of /plugin. To update, pull the latest commit and restart Claude Code. After a marketplace install the plugin lives at ~/.claude/marketplaces/domino/plugins/domino-claude-plugin, not where you first cloned it. After a --plugin-dir install, pull in the cloned directory.

Next steps

Get started with coding assistants: launch a Workspace and authenticate your agent.
Skills catalog: the skills available in the DSE.
Modeling assistant walkthrough: an end-to-end example of a skill doing real work.
Custom environments: pin versions, add tools, or standardize on a different agent.
Troubleshoot coding assistants: known issues, including skill availability for agents other than Claude Code.

API

CLI

Python SDK

Domino skills reference

What the plugin ships

Preinstalled in Workspaces

Use the skills elsewhere

Contribute a skill

Install the plugin yourself

Next steps

​What the plugin ships

​Preinstalled in Workspaces

​Use the skills elsewhere

​Contribute a skill

​Install the plugin yourself

​Next steps

What the plugin ships

Preinstalled in Workspaces

Use the skills elsewhere

Contribute a skill

Install the plugin yourself

Next steps