Skip to content

Launch from Ticket

The hands-on ClusterCode workflow: paste a ticket, get a DevBox prepared for it — repo cloned, branch created, ticket context written, agent CLI installed — and take the terminal yourself. To have ClusterCode do the work instead, see Run an agent on demand.

From the orchestrator dashboard, click Launch or ask Nucleus:

“Launch a DevBox for https://github.com/org/repo/issues/42

Supported formats:

  • https://github.com/owner/repo/issues/123
  • https://dev.azure.com/org/project/_workitems/edit/456
  • Trello and Monday.com — coming soon

The system pre-selects a worker for you, and you can override it. Leaving it on auto-select hands the choice to the placement rules — which pack onto the fullest host that still fits, not the least-loaded one. See how a target is chosen.

Pick an AI agent:

  • Claude Code (default)
  • GitHub Copilot
  • Codex

Click Launch. Within 30 seconds the DevBox is ready:

  1. Ticket details are fetched from the provider
  2. A CLAUDE.md is generated with full context
  3. The repository is cloned into the DevBox
  4. The agent CLI is installed and the ticket context is in place

The agent is ready — it is not running yet. A launch gives you a prepared environment, not an autonomous worker. To start work, pick one:

You wantDo this
To drive the agent yourselfOpen the live terminal and run claude (or codex, copilot)
The agent started for you on every launchAdd a Post-Launch Workflow to your NUCLEUS.md
ClusterCode to plan and execute the whole ticketStart a Run with the ticket URL instead of launching a DevBox

From the dashboard:

  • Watch the live terminal output
  • See resource usage (CPU, memory, disk)
  • Intervene at any time via the web terminal

A DevBox does not open a PR on its own — the agent working inside it does, when you ask it to. Once the work is done, from the live terminal:

  • gh pr create --draft opens the draft PR
  • Update the ticket from the terminal, or let a Run handle it

If you want ticket → PR with no manual steps, that’s a Run, not a launch.

  • Use specific, well-written tickets — the AI agent’s output quality depends on the input context
  • Enable host networking if the DevBox needs to expose ports for local testing
  • You can launch multiple DevBoxes in parallel across different workers