CLI / Use with AI agents

Use Workspark with AI agents

Give Claude, Codex, or OpenCode grounded access to your Workspark data. Your agent can log accomplishments, help you prep a self-review, check on your OKRs, and read your org data, all from real output, never fabricated.

What you need

Two pieces work together to connect an agent to Workspark.

The CLI

ws

A single binary that talks to Workspark. The grounded connection, the agent's hands.

The Skill

workspark/skills

Instructions that teach the agent to drive ws correctly. The agent's playbook.

= Your agent can log accomplishments, prep reviews, and update OKRs, grounded in real data with no hallucinations.

How you install the two pieces depends on your agent. Jump to your path:

Claude

Easiest — CLI + Skill in one plugin

On Claude, a single plugin bundles both the ws binary and the Skill. Nothing else to install.

1

Download and upload the plugin

Download the Workspark plugin file, then upload it into Claude. One-click install from the Workspark marketplace is coming soon.

Upload it into Claude (desktop app / Cowork)

  1. Open Customize then Personal plugins.
  2. Click the + button.
  3. Choose Create plugin then Upload plugin.
  4. Select the workspark-0.2.2.plugin file you just downloaded.
2

Allow network egress to Workspark (required in Cowork)

The wsbinary runs inside Claude's sandbox, which blocks outbound network connections by default. Without allowlisting Workspark's domains, the CLI cannot reach the API and every command will fail with a network error.

In Claude settings, go to Settings then Capabilities then Allow network egress. Turn the toggle on. Then, under Domain allowlist and Additional allowed domains, add both of these:

app.workspark.io
workspark.io

Security note:Claude's settings panel shows a caveat that network egress carries risk. Only add domains you trust.app.workspark.iois the Workspark API andworkspark.iocovers the sign-in flow.

3

Authenticate

The plugin's Skill installs the bundled ws binary automatically the first time Claude uses it. Then it needs to sign in to your Workspark account. Two flows: Claude Cowork in the sandbox, or Claude Code on your machine.

Claude Cowork

The Cowork sandbox can't open a browser, so Claude walks you through it. Just ask Claude to log in:

  1. Tell Claude “Log in to Workspark.”
  2. Claude gives you a URL. Open it on your own machine and approve the sign-in.
  3. Come back and tell Claude “continue” or “done”. Claude validates the session and you're ready to go.

The sign-in link expires after a few minutes. If it lapses, just ask Claude to start the login again.

Claude Code

Ask Claude to run auth. It calls ws auth login, which opens the browser on your machine to sign in. Just tell Claude “Log in to Workspark.”

4

Try it

Once authenticated, tell Claude what you want to do. Some prompts to start with:

  • Log what I shipped today as an accomplishment.
  • Help me prep my self-review for the current cycle.
  • Which of my OKRs are stale?
  • Who on my team has been most active this month?

Claude confirms before taking any write action and never fabricates accomplishment entries, review content, OKR updates, or survey responses. Everything it does is grounded in what the CLI actually returns.

Codex, OpenCode & other harnesses

Add both pieces

For any other agent, install the two pieces yourself: the CLI binary, then the Skill that teaches the agent to use it.

1

Install the CLI

Install the ws binary so the agent has something to call. The simplest option works on every platform:

Terminal
npm install -g @workspark/cli

Prefer a shell script, PowerShell, or manual binaries? See the Install the CLI guide for all options, plus how to authenticate and set your organisation.

2

Install the Skill

Add the Workspark Skill to your agent harness so it knows how to drive the CLI correctly. One command:

Terminal
npx skills add workspark/skills

Prefer to browse it first? The Skill is open source at github.com/workspark/skills.

This installs the Workspark Skill for your harness, including supported agents such as OpenAI Codex and OpenCode. Once installed, the agent pulls in the Skill when you ask it to work with Workspark.

Restart the agent harness after installing the Skill. Most harnesses only load Skill and instruction files at startup, so a running session won't pick up the new Skill until you restart it.