Setup

You run the discovery and the build in the same place: a workbench on your machine with Claude Code in it. One template puts everything you need into it. You set this up once and then stay there.

Note

acme is a sample workbench name. Use your own, and replace it everywhere below.

Set it up

1. Install Monoceros. You run one script, and Docker has to be running. Installation has it for macOS, Linux and Windows.

2. Create the workbench from the discovery template. acme is the name you give it. Every later command takes that name.

   monoceros init acme --template=discovery-atlassian

   The template puts three things into the workbench: Claude Code with the discovery plugin, the three build roles, and the Atlassian CLI. The skills use that CLI to read and write Jira and Confluence.

   Then init asks you for the values those features need: your Atlassian site, your account email, the API token, and a Claude API key if you sign in with one. You can leave an answer empty and fill the value in later. It goes into acme.env for this workbench, or into ~/.monoceros/monoceros-config.env for all of them. The one to get right now is the Atlassian token.

3. Bring the repository along, if the project already has one. Not every project starts from nothing. If the code for it already exists somewhere, name that repository here and it comes into the workbench.

   monoceros init acme --template=discovery-atlassian --with-repos=https://github.com/acme/guestbook.git

4. Build it. The first run pulls the image, so it takes a few minutes.

   monoceros apply acme

5. Attach the Claude Code app over SSH. You then work in the app instead of the terminal. You do not install anything inside the container and you do not sign in there. Claude Code has the steps.

6. Write the brief. Say what you want to build, and the right skill starts.

   I'd like to build something that keeps my freelance invoices in order.
   Let's write a brief for it.

From here you work through the seven skills to a finished backlog, and the same workbench builds it. Discovery has them in order.

The Atlassian token

The skills write to Jira and Confluence as you, with your rights and no more. They need three values for that, and init asks you for all three.

Value	What it is
ATLASSIAN_INSTANCE	Your site host, yoursite.atlassian.net. The CLI uses it to find your cloud ID.
ATLASSIAN_EMAIL	The account that writes the pages and issues.
ATLASSIAN_API_TOKEN	A scoped API token with Teamwork Graph as the app.

This pre-filled link opens the token form with Teamwork Graph and every scope selected. Atlassian shows the value once, so copy it before you leave the page.

init writes your answers to acme.env, next to the workbench. You can also put them into ~/.monoceros/monoceros-config.env and share them across every workbench you build. If a value is in both files, the one in acme.env counts. Rovo Dev and Bitbucket need two more tokens; the Atlassian feature page has all of them.

Caution

Do not put the token into the yml. The ${…} placeholders in the feature block read your env file when you apply, and the yml is the file you commit. See Workbench configuration.

Diagrams

Two pages come with a diagram: the domain model and the technical frame. The skill draws it and attaches it to the page as an image, so your Confluence needs no extra app.

Attaching the picture needs a terminal. Claude Code has one, and so does a workbench. In a chat without a terminal, the skill gives you the picture and tells you which heading it goes under. The page holds everything even without it, because the tables and the text carry the content.

Calling a skill by name

You do not have to name a skill. Say what you want, and the right one starts: “let’s work out the personas” begins that step. To call one directly, use its slash command.

/discovery-brief-atlassian

Every skill page has an example prompt for both cases, the one without a name and the one for a new session.

Keeping it current

The plugin is not pinned to a version. Every monoceros apply fetches the marketplace again and updates the plugin with it. So a change on our side reaches you the next time you apply. See Reproducibility and upgrades.

Running the discovery in claude.ai instead

You can install the plugin in claude.ai or in the desktop app and run the dialog there. Open Customize and go to the Plugins tab. Add the marketplace https://github.com/getmonoceros/monoceros-discovery.git, then install Monoceros Discovery for Atlassian. Anthropic describes the steps in Use plugins in Claude. Connect the Atlassian connector there as well, because claude.ai has no shell to run a CLI in.

This way works up to the project files, and it stops there. That skill writes the README and the agent instructions into the project directory, and claude.ai cannot write files. You get the content as Markdown blocks and save it yourself. The build cannot run there at all, because the planner, the implementer and the reviewer are a workbench feature.

So you move into a workbench either way. Doing it first is the shorter way.