✦ New • Free AI Skill
Draft and validate GitLab CI pipelines without leaving your editor.
Draft and validate GitLab CI pipelines without leaving your editor.
A free AI skill that drafts and validates your .gitlab-ci.yml in your local editor. Works with Cursor, VS Code, Claude Code, and any agent you already use.
New! Actively shipping improvements, share your feedback with us.
0GitLab accounts needed to validate locally
6+AI Agents and editors supported
MITOpen source (skill & CLI)
The Shift
From commit-to-validate, to validate-to-commit
Pipelines are the only part of the modern dev stack you still can't validate locally. The skill drafts the YAML in your editor. glci (an experimental GitLab project) runs it against the real runner before you push. You stop using your remote pipeline as a debugger and your git history as a typo log.
Before
Commit. Push. Wait. Fail. Repeat.
- Write the YAML by hand from memory or docs
- Commit and push to a branch to find out if it works
- Wait 8–12 minutes for a remote runner
- Fail on a typo, a missing variable, or a misnamed job
- Edit, push, repeat 3–4 times
- Leave a trail of "fix CI" commits in your history
With the skill + glci
Draft. Validate. Push when it's green.
- Ask the agent to draft a pipeline from your repo
- Run
glci showto inspect the job graph - Run
glci runto execute every job in real Docker - Fix what fails — in seconds, not minutes
- Push once, with a pipeline you've already seen pass
- Keep your git history about your code, not your YAML
The shift isn't a faster pipeline. It's a different relationship with your pipeline. The same one you already have with your application code.
Get started
Two steps. About five minutes
No GitLab account needed to validate locally. You don't push anything until you choose to.
01~1 minute
Add the skill to your editor
Drop the skill into Claude Code, Cursor, VS Code, OpenCode, or Codex. The agent now knows GitLab CI/CD — syntax, best practices, your stack.
02~3 minutes
Ask, run, push
"Write a CI pipeline for this project." Review the YAML the agent drafts. Run glci run. Push when it's green.
What you need
- A compatible editor or agentClaude Code, Cursor, VS Code, OpenCode, Codex, or anything that loads markdown skills.
- A projectAny codebase, hosted anywhere. The agent reads from your local working directory and drafts a suggested pipeline automatically.
- Docker running locallySo glci can validate and execute jobs in real containers.
- A GitLab project(when you're ready to run CI on every push)The skill and glci validate locally without it; you'll need one when you want pipelines to run in the cloud.
Install
Works with your existing AI agents
A CLI to run pipelines locally, and a skill to draft them in your editor. Install in either order.
Add the skill to your editor
Pick your agent
Download the skill and learn where to place files that Cursor discovers.
git clone https://gitlab.com/gitlab-org/ci-cd/gitlab-ci-skill.git ~/.cursor/skills/gitlab-ci-skill
Reload Cursor. The agent uses the skill automatically when prompted: "Write a CI pipeline for this project."
Why GitLab
More than a runner. More than this skill
You don't want to log into the UI to write a pipeline. This skill keeps you in your editor for as long as it can. But sometimes you have to come back, whether it's a broken pipeline, an MR review, a deploy gone sideways. When you do, the platform's already wired up. Code, pipelines, registry, secrets, and deploys live in one place. Whatever editor or agent you use plugs in. The same security checks run on AI-written code as on yours.
One data model. Open at the edges.
Code, pipelines, packages, security findings, deploys, releases — all on the same system, connected rather than synced. Whatever editor, agent, or model you bring plugs in via MCP and works from the same authoritative view. Open at the edges, governed in the middle.
Context is what separates fast AI from trustworthy AI.
Agents without context write code that looks right and breaks production because they can't see what depends on a change or what already exists. GitLab's knowledge graph keeps a live map of how your code, pipelines, deploys, and security findings connect, so questions about blast radius or downstream impact get answered in seconds, not days. Any agent can read from it.
Governance that's structural, not bolted on.
AI-written code goes through the same security scans, the same approvals, and the same audit trail as code you wrote. Agents have scoped identities, behavioral policies, and full chain-of-custody. Bring your own model, your own cloud, your own agent — all governed by the same fabric.
From traditional to autonomous, same platform.
Some of your teams will keep writing code by hand. Others will direct agents on specific tasks. A few will run agents autonomously on lower-risk work. All three live on the same data model and governance so teams advance at their own pace, with no re-platforming as their AI maturity changes.
We're actively building this skill, and the team behind it wants it to fit how you actually work. Tell us what's clicking and what's getting in the way. Share your feedback with us.
Going further
Two more places GitLab's AI meets your CI
The GitLab CI Skill is purpose-built for drafting and validating new pipelines in your editor. When your CI/CD work changes shape, GitLab has companion products for the other moments.
Free AI Skill · MigrationView the migration skill
Coming from GitHub Actions?
The GitHub Actions migration skill reads your .github/workflows/ and converts them to idiomatic GitLab CI/CD, flagging anything that needs a manual decision. Same editors, same workflow.
GitLab Duo Agent PlatformMeet CI Expert Agent
Pipeline getting complex?
CI Expert Agent lives inside GitLab Duo Agent Platform with full project context: reading live job logs, optimizing build times, debugging flaky jobs, and working across multi-project pipelines. For when your pipelines stop being something you write and start being something you operate.