1. CLI
  2. Personal agents

CLI

Personal agents

Use Stubber from a coding agent such as Claude Code, Codex or OpenCode, with prompts you can paste in

A coding agent that can run terminal commands can work in Stubber through the CLI. Ask it in plain language to build a WhatsApp bot, find out why a stub is stuck, or add an action to a template, and it reads your templates, makes the change on the draft branch, runs a test stub, checks the stubposts and tells you what happened.

The agent learns Stubber from the Stubber skill: instructions and reference notes on templates, contexts, actions, notifications, AI chats and how to test a change. The CLI installs it.

WARNING

A coding agent sends what it reads to its AI provider: your org's templates, stubs, messages and customer data. Use it under an account your organization has approved, such as a business or enterprise plan whose terms keep that data out of model training. If you're unsure, ask your organization before pointing an agent at its Stubber data.

Set up

  1. Install the CLI.

  2. Sign in yourself. This opens a browser, so don't leave it to the agent:

            stubber auth login
    
          
  3. Install the skill:

            stubber skill install
    
          

    It goes where Claude Code, Codex and OpenCode look for skills. See Agent skill for the options.

  4. Start your agent in a folder of its own, for example ~/stubber-work. The agent keeps template copies, patch files and test inputs there.

Or paste the setup prompt below and let the agent do steps 1 and 3.

Prompts to give your agent

Copy these into your agent's chat. Replace the parts in <angle brackets>.

Set up Stubber

        Set me up to work in Stubber with the stubber CLI.

1. Check whether `stubber` is installed (`stubber --version`). If not, install it:
   - Linux or macOS: curl -fsSL https://cli.stubber.systems/install.sh | bash
   - Windows (PowerShell): irm https://cli.stubber.systems/install.ps1 | iex
2. Install or update the Stubber skill: `stubber skill install`. Then load the stubber-cli skill.
3. Run `stubber auth whoami`. If I'm not signed in, ask me to run `stubber auth login` myself and wait for me.
4. List the orgs I can use and ask me which one to work in, then select it with `stubber scope org <uuid>`.

      

Working rules

Paste this at the start of a session, or save it in your agent's project instructions file (CLAUDE.md for Claude Code, AGENTS.md for Codex and OpenCode):

        When you work in Stubber:
- Use the stubber CLI and the stubber-cli skill. Look things up with `stubber explain` and the library instead of guessing field names.
- Keep your working files in one folder per org and template, e.g. <org_domain>/<template_name>/.
- Change the draft branch only. Test on a draft stub and read its stubposts before telling me a change works.
- Ask me before you publish to live, delete anything, or send a message to a real person.
- Before publishing, diff draft against live and tell me about any changes that aren't yours.
- When you finish, give me the Stubber links to what you changed or created.

      

Find a template

        In Stubber org <org name or domain>, find the template that handles <what it does>. Show me its states and main actions, and which channels it talks on.

      

Build something new

        In Stubber org <org name or domain>, build a template that <describe the process>.
It should talk to people on <webchat / WhatsApp / email>.
Build it on draft, run a test stub through the main path, and show me the stubposts. Don't publish until I say so.

      

Change a template

        In the Stubber template <template name>, <describe the change, e.g. "add an action in the pending state that lets an admin reject the application with a reason">.
Change draft only, test it on a draft stub, then tell me what you changed and how you tested it.

      

Debug a stub

        Stubber stub <stubref> isn't doing what I expect: <what happened, what should have happened>.
Read the stub and its stubposts, find the cause, and suggest a fix. Don't change the template yet.

      

Agents without skills

If your agent doesn't support skills but can run commands and fetch web pages, point it at the skill directly:

        Read https://docs.stubber.com/skills/stubber-cli/SKILL.md and the reference files it links to, then follow it to work in Stubber with the stubber CLI.

      

Keep the skill current

We add to the skill often. Run stubber skill install again every so often, and after stubber self-update, to get the latest.


Previous <- CLI