Skip to content
Shipyard

Ask for a feature. Agents build it and launch it here.

Describe a change to this site. GitLab Duo agents triage it, write the code, test it, scan it, deploy it and watch it in production. Nobody on our side writes the code.

0 in the yard now, 21 launched so far.

Requests are closed

The hackathon is being judged, and the project it submitted must not change until the winners are announced, so the agents are switched off.

The roadmap, changelog, incidents and carbon pages show everything they did up to the deadline.

How a request ships

Six steps, each run by an agent or the pipeline. A dispatcher hands work from one to the next and merges only what was reviewed.

  1. Step 1

    Received

    Dispatcher

    Filed as a GitLab issue. The triage agent picks it up within a minute or two.

  2. Step 2

    Triaged

    Triage agent

    Checked that it fits this site and wrote the acceptance criteria the change must meet.

  3. Step 3

    Building

    Planner and developer agents

    Writing the change and its tests, running them, then opening a merge request.

  4. Step 4

    In review

    Pipeline and review agent

    Tests and security scans passed. The review agent is checking the change against the criteria.

  5. Step 5

    On staging

    Pipeline

    Merged and running on staging while the pipeline smoke-tests it.

  6. Step 6

    Live

    Pipeline and monitor

    Passed a canary and a 60-second watch in production. Alerts roll it back if it misbehaves.

Every change must pass Scope policy Unit tests SAST Secret detection Dependency audit Licence check Container scan Agent review Canary Production watch

Frequently asked questions

Who writes the code?
GitLab Duo agents do: they triage each request, plan and write the change and its tests, and review it. Nobody on our side writes the code.
How is a request checked before it goes live?
Tests, security scans and the agent review come before the change is merged and tried on staging. The canary and the 60-second production watch come after the merge, and a change only counts as live once it passes them.
What happens if something breaks?
Production alerts open an incident, and the pipeline rolls the change back or fixes it forward. If automation stops, it asks a maintainer to take over, and the incidents page shows each one.

Recently launched