Desktop/Set up Ponder Training

Set up Ponder Training

Install the OpenPond CLI and maintained skills in your coding agent, then verify access without starting an experiment.

This guide sets up the CLI and skills for Ponder Training. Finish installation and harmless access checks first. Choosing a model, preparing data and running training are separate operations.

Check the current installation

Identify the current coding agent and the intended project. Inspect openpond version and openpond --help. If the CLI is missing, install the official package with npm install -g openpond, using the Node.js version required by that package. Preserve an existing installation and its configuration where possible.

Select the existing OpenPond home before checking authentication. Desktop normally uses ~/.openpond, nightly uses ~/.openpond-nightly, and development Desktop uses ~/.openpond/openpond-app-dev. Reuse an explicit override when present, and pass --home <selected-home> on each command. A missing account in a different home does not mean the Desktop account is missing.

openpond --home <selected-home> home
openpond --home <selected-home> profiles list

Keep the selected account's web and API origins together. Do not read credential-store files or copy credentials into prompts, logs or shell history.

Install the maintained skills for this agent

Desktop already includes the customer skills. For another coding agent, inspect its installed skills before adding anything. The maintained sources are in OpenPond's CLI skill pack; each skill has a SKILL.md file and may include referenced workflow documentation.

Use the Skills CLI's discovery command first:

npx skills@1.7.4 add openpond/openpond --full-depth --list

Install for the intended client and project only. For Codex:

npx skills@1.7.4 add openpond/openpond --full-depth \
  --skill openpond-cli openpond-taskset-authoring openpond-agent-authoring openpond-skill-authoring openpond-refiner-authoring \
  --agent codex

Use claude-code or cursor for those clients. Review existing customized skills before replacing them. Use --global only when the user wants a user-wide installation. Do not install developer-only skills or use --all.

Open a new agent session if required for discovery. Confirm that the client can find the installed skills and read openpond-cli/SKILL.md, its command references and the taskset-authoring skill. Installation alone does not prove the client can use them.

Verify Ponder Training access

Inspect the installed command help before using its supported training and experiment commands:

openpond --home <selected-home> --help
openpond --home <selected-home> training --help
openpond --home <selected-home> experiments --help
openpond --home <selected-home> account

Reuse an existing account with the required public API credential. If login is necessary, use the CLI's private interactive login flow; never ask for a key in chat. If the account has only a Desktop session and the command requires an API key, explain that specific boundary. Local-only setup does not require cloud login.

For a workspace explicitly selected by the user or Console, verify a bounded inventory read:

openpond --home <selected-home> datasets list --team <workspace-id> \
  --api-base-url <selected-account-api-origin> --json

An empty list is a successful access check. A permission error is not a missing installation.

Report the CLI version and selected home, discovered skills, and verified account/workspace or local-only status. Report any incomplete step. Stop here: setup does not authorize uploads, Fix-it, tool capture, experiment creation, inference, training, paid compute or model activation.

For later work, use the Ponder Training guide and the maintained CLI command reference.

View as Markdown