Skip to content

GitLab Adds Setup Skill to Pair Coding Agents With Duo

GitLab published recommended configurations for Claude Code, Codex, Cursor and other agents, plus an installable skill that turns on automatic review and pipeline repair.

Draftship Editorial4 min read

GitLab Duo coding agent setup arrives as an installable skill

GitLab published a guide to using coding agents with GitLab on November 19, 2026, aimed at teams that already write code with Claude Code, Codex, Cursor, Devin or OpenCode, sometimes more than one of them. Alongside the guidance it shipped something installable: a setup skill that connects a coding agent to GitLab and configures automatic review and failed-pipeline repair. The argument behind GitLab Duo coding agent setup is consistency. When review, CI checks and security scans depend on each engineer's setup, skill or spare time, the post says, even changes in the same project move at different speeds and get different levels of scrutiny.

What the setup skill configures

You run the skill against a GitLab group or a single project. GitLab says the agent checks what is already switched on, shows one plan, and changes nothing until you approve it. Someone with the Owner role can set up the whole group; a person with the Maintainer role can start with one project. On the Ultimate tier the plan can also opt in to security triage and remediation.

The setup is not tied to your editor staying open. Once configured, the work keeps running in GitLab after you close your coding session. An optional MR follow-up skill then checks what is already running and brings review and pipeline feedback back into that session.

Review, CI repair, security fixes and issue handoff

  • Review. Duo gives a change a second look when the merge request is marked ready and posts feedback on the diff. Your agent can read the comments, fix the ones you agree with and reply on each thread.
  • CI repair. When a pipeline fails, an automation starts Fix CI/CD Pipeline and links the session on the MR. Duo suggests a change, opens a fix MR, or explains what it found if it cannot propose a reliable fix. The fix still has to pass CI.
  • Security. On Ultimate, Duo flags likely false positives from static application security testing and secret detection, and opens MRs with fixes for eligible SAST findings, usually about 10 minutes after a scan.
  • Issues. An assigned issue can be handed to the Developer flow, which reads the issue and the repository, writes the change and opens an MR in your name.

Approvals stay policy driven. Rules can require a person to approve every MR, or add approvers when a scan finds a vulnerability. GitLab states that if Duo approves an MR, that approval does not count toward required approvals. Composite identity ties each agent action to both the agent and the person who started it, and limits access to what both are allowed to do.

The price GitLab puts on a review

GitLab says a Duo review costs a flat $0.25 at list price, so 1,000 reviews come to $250. It compares that with Anthropic's Code Review at an average of $15 to $25 per review, which would put the same volume at $15,000 or more. Those are GitLab's own figures and its own comparison, not an independent measurement, and list price is not necessarily what a given contract pays.

What changes for a team already using an agent

The practical shift is that review and CI follow-up stop being personal scripts. Duo works from the repository, the MR discussion, CI logs and scanner results, so context does not have to be reassembled each time work moves between a tool and a person. Teams also pick the model for each agentic flow from Anthropic, OpenAI, Google or open-weight providers. Getting started means pasting an install instruction for the skills repository into your agent and running the setup command for your group, then approving the plan.

Limits and open questions

The post does not say which paid tiers include the setup skill itself versus only the Ultimate security features, whether the skill works on self-managed GitLab, or how it behaves when two different agents touch the same merge request. It names Devin and OpenCode among the supported agents but does not reproduce the individual recommended configuration for each one in the body of the post. The $0.25 figure is described as list price, so effective cost under an existing agreement is unstated. The pricing comparison comes from GitLab, which is selling the cheaper option.

For teams building apps with AI

A review gate only helps once code lives somewhere with a merge step. If you are generating an app rather than maintaining a large repository, that step may be premature. Draftship builds and hosts an app from a plain-language description, with a built-in database, user sign-in and hosting, and paid plans add source code access and GitHub push. That push is the point where a review workflow like the one GitLab describes could sit in front of changes. Draftship is not affiliated with GitLab, and nothing here suggests it integrates with GitLab Duo.

Primary sources

This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.