Skip to content

CANFAR Agent Skill

Use your coding agent to launch a notebook, plan a batch workflow, move research data, or find answers in the CANFAR documentation. The official canfar skill provides instructions for these tasks and is maintained in the CANFAR repository.

Install the skill

You need a local coding agent, Node.js with npm, and Git. The skills installer requires Node.js 22.20.0 or newer as of September 2026; see its current requirements.

In your project's terminal, run:

npx skills add opencadc/canfar --skill canfar

This uses the skills.sh installer to install the canfar skill from the official repository. If npm asks to download the skills package, confirm to continue. Select your coding agent and choose the installation scope:

Scope When to choose it
Project You want the skill available in the project where you ran the command.
Global You want it available to your selected agent across your projects on this machine.

To select Global directly, add --global:

npx skills add opencadc/canfar --skill canfar --global

If the installer asks for an installation method, keep its recommended symlink option, or choose Copy when your environment does not support symlinks. The installer reference explains both methods.

The skill installs instructions for your agent. To operate CANFAR, you also need the CANFAR client and an account with access. You can ask documentation questions before setting those up.

Use the skill in your agent

Choose the tab for your local agent. Each tab gives an optional command to select that agent during installation, followed by a prompt to enter in its chat. You only need to install once for your chosen scope.

In the Codex command-line interface (CLI) or editor extension, type $ and select canfar, or find it through /skills. Then send:

$canfar explain where I should save my CANFAR results so they survive the end of a session.

See Codex skill invocation.

In Claude Code, invoke the skill by name:

/canfar explain where I should save my CANFAR results so they survive the end of a session.

See Claude Code skills.

In VS Code chat, type /, select canfar, and add your request:

/canfar explain where I should save my CANFAR results so they survive the end of a session.

See VS Code skill commands.

In Cursor Agent chat, type /, select canfar, and add your request:

/canfar explain where I should save my CANFAR results so they survive the end of a session.

See Cursor skills.

These agents can also select the skill when a request matches its description. Invoking it explicitly is a useful first check that the agent can find it.

Ask for what you want

After selecting the skill, describe the task and the result you expect. Replace SERVER, PROJECT, and SESSION_ID with your own values.

Start a notebook

Start a CANFAR notebook on cadc for my work.
Use an available astronomy image, wait until the Session is Running,
and give me its connection link and ID.

Plan a batch run

Plan a CANFAR batch run for reduce.py over the FITS files
in /arc/projects/{PROJECT}/raw. Use four replicas and save outputs
in /arc/projects/{PROJECT}/reduced. Check how the script will reach
the remote Sessions and show me the plan before submitting it.

Copy results

Copy ./results from my laptop to arc:/projects/PROJECT/results
and check that the expected files arrived.

Diagnose a Session

Find out why CANFAR Session SESSION_ID is still Pending
and explain what I should do next.

Include the Science Platform Server, project, input and output paths, and any Container Image or package versions your analysis requires. Say when you want a plan before execution. Your agent uses the tools and permissions available in the environment where it runs.

Complete login yourself

The skill tells the agent to check your installed client and use its command help. When you need to authenticate, it asks you to log in through your Identity Provider: the Canadian Astronomy Data Centre (CADC) or the SKA Regional Centre Network (SRCNet). Run the appropriate command in your own terminal:

canfar login cadc

canfar login srcnet

Enter your password or approve the device yourself, then tell the agent to continue. Keep credentials out of chat, scripts, and shared logs. See Authentication and servers for setup and re-login guidance.

Check the result

The skill asks the agent to report the Session IDs, observed states, and verified output paths. A created Session has been accepted; it still needs to become Running. For a task that produces data, successful completion also requires checking its output on persistent storage.

Review those results before requesting cleanup. The skill instructs the agent to keep cleanup within the work you authorized.

Update or remove the skill

List installed skills to check that canfar is present:

npx skills list

Choose the scope you used when installing to update only CANFAR:

Run from the same project directory:

npx skills update canfar --project

npx skills update canfar --global

To uninstall it, use npx skills remove canfar from the project directory, or npx skills remove canfar --global for a global installation. These commands follow the skills CLI reference.

If the skill does not appear

  1. Run npx skills list and check the reported agent and installation scope.
  2. For a Project installation, open that project in your agent. Confirm that installation ran on the same machine as the agent.
  3. Use the skill picker described in your agent's tab. If Codex has not picked up a change, restart it. Restart Cursor to trigger discovery at startup.
  4. If the installer rejects a command or reports an unsupported Node.js version, check the current installer requirements and options.

For CANFAR workflow help, continue with Getting started, Data access, or Support.