Quick start with a coding agent

This quick start does the same as Quick start in the app, but your own coding agent does the work. You install the Artemis skills, which teach the agent to drive Artemis through the Artemis CLI, and then ask it to run a first Discovery on quickstart, a small Python project made for trying Discovery with a local model. The skills work with Claude Code, Codex, Cursor and GitHub Copilot.

Before you start

• Install Artemis Desktop.
• Set up your model and check that the indicator beside Models is green.
• Choose a project. This page uses quickstart, a small Python project made for trying Discovery with a local model, but you can use one of the Example projects or any project you like. For another project, use its own install, test and benchmark commands in place of the quickstart's. The agent imports the repository for you. Fork it first so that the agent can push the result to your fork.

1. Install the Artemis skills

Install the skills for your coding agent, then start or reload the agent so that it loads them. For updates and the list of skills, see Artemis Skills.

Claude Code

Install the Artemis plugin in either of these ways:

From your terminal

claude plugin marketplace add turintech/skills
claude plugin install artemis@skills

Inside Claude Code

/plugin marketplace add turintech/skills
/plugin install artemis@skills

Then start Claude Code, or run /reload-plugins if it is already running. Check the installation with /plugin list.

Codex

With Node.js available, install the skills at user scope:

npx skills add turintech/skills --agent codex --global --skill '*'

Then restart Codex.

Cursor

With Node.js available, install the skills at user scope:

npx skills add turintech/skills --agent cursor --global --skill '*'

Then restart Cursor.

GitHub Copilot

With Node.js available, install the skills at user scope:

npx skills add turintech/skills --agent github-copilot --global --skill '*'

Then restart your editor.

2. Connect your agent

1. In Artemis Desktop, open Platform Settings → Connect Your Agent.
2. Copy the setup prompt and paste it into your coding agent. With the skills loaded, the agent installs the Artemis CLI and points it at your Artemis Desktop deployment.
3. Create an API key under Platform Settings → API Keys and enter it in your own terminal when the agent asks you to sign in. Do not paste it into the conversation.

3. Prompt your agent to start a Discovery

Ask the agent to use the quickstart skill, and give it the repository, model and benchmark:

Use the Artemis quickstart skill to run my first Discovery.

- Repository: https://github.com/<your-github-user>/quickstart
  (my fork of artemisTurintech/quickstart)
- Model: my local model, Local/Qwen3-Coder-30B-A3B. Use it for every
  model choice, including the review models.
- Runner: set one up on this machine if none is online
- Script: setup with "python -m py_compile slow.py" and
  "python test_slow.py", benchmark with "python benchmark.py"
- Objective: reduce total_ms by speeding up the functions in slow.py
  without changing their results. All tests must keep passing.
- Generate 5 versions.

Pause whenever you need a credential or an account choice from me.
When the run finishes, compare the best version with the Baseline
and explain what it changed.

4. Follow along

The agent works through the setup and the run with the matching skills, and stops for the steps only you can do:

Stage	Skill	What you may need to do
Set up a Runner	runner-setup	Approve installing and starting the Runner on your machine.
Connect GitHub	project-import	Connect GitHub under Platform Settings → Git if no Git key exists yet. A key is needed for import even for a public repository. See Git Integration.
Import the project and create the Script	project-import, repo-command-setup	Nothing, unless the Script fails on the Runner.
Start the Discovery Run	discovery-start	Nothing. The agent waits for the Baseline and checks that exploration started.
Evaluate the result	discovery-inspect	Read the agent's summary and decide whether to keep the change.

With a local model, each Version can take a while. You can also follow the run in Artemis Desktop under Discover.

5. Evaluate the Version and push it to your repository

Evaluate. Ask the agent to compare the best Version with the Baseline using the discovery-inspect skill: how much each metric changed, whether all tests still pass, and what the code change does. You can also open the run in Artemis Desktop under Discover → Versions to see each Version's scores and diff. See Versions.

Push upstream. When a Version is worth keeping, ask the agent to push it to your repository. Each Version lives on its own Branch, and the agent can either:

• publish the Branch as a Git branch on your repository, or
• open a Pull Request from it, so the change can be reviewed before it is merged.

Your repository is not changed until you publish the Branch or open the Pull Request. To push to a repository that belongs to someone else, such as artemisTurintech/quickstart, import your fork of it.

The CLI commands behind these steps

The agent runs commands like these. You can also run them yourself.

# Sign in and check what is available
# Connect Your Agent already set the deployment URL, so only the key is needed
artemis login --api-key <your-api-key>
artemis status
artemis model list
artemis runner list
artemis key list

# Import the project
artemis project import `
  --git-url https://github.com/<your-github-user>/quickstart.git `
  --key-id <key-id> --name "quickstart"

# Select the Runner and create the Script
artemis project runner set --project quickstart --runner <runner>
artemis project scripts create --project quickstart --name "benchmark" --default `
  --setup-cmd "python -m py_compile slow.py" --setup-cmd "python test_slow.py" `
  --benchmark-cmd "python benchmark.py"

# Start the Discovery Run
artemis discovery create --project quickstart `
  --task "Reduce total_ms by speeding up the functions in slow.py without changing their results. All tests must keep passing." `
  --model <model> --review-model <model> `
  --target-files slow.py `
  --versions 5 --script <script-id> --runner <runner>

# Evaluate the Versions against the Baseline
artemis discovery get <run-id>
artemis discovery versions list <run-id>
artemis discovery metrics <run-id> --stats

# Push the best Version upstream: publish its Branch, or open a Pull Request
artemis changeset list --project quickstart
artemis changeset publish <changeset-id> --project quickstart
artemis changeset pr <changeset-id> --project quickstart --title "perf: faster slow.py"

See the Artemis CLI reference for every command.

Next steps

• Ask the agent to steer the run with the discovery-steer skill, for example to extend it with more Versions.
• Try the same steps on one of the Example projects, or on your own repository.