OnboardIQ

• 

  OnboardIQ triggered via @onboardiq command in GitLab Duo Chat - generates personalized onboarding map instantly

• 

  Agent surfaces real team members, commit history, and the 5 most important files to read first - pulled live from the repo

• 

  Live commit history and in-flight MRs surfaced automatically - new dev sees exactly what's happening right now

• 

  Personalized first-week checklist generated from actual repo state, with green pipeline status confirmed

Inspiration

Every time a developer joins a new project, the same frustrating cycle repeats days lost figuring out what to look at, interrupting senior engineers with "basic" questions, and critical knowledge that lives in someone's head instead of the docs. I've seen this slow down teams firsthand. When I discovered GitLab Orbit's knowledge graph, which already maps code ownership, file relationships, and contributor activity, I realised the data to fix this problem already exists. It just needed an agent to surface it at the right moment.

What it does

OnboardIQ is a GitLab Duo AI Flow that activates when a new developer joins a project. It queries the GitLab Orbit knowledge graph to understand the codebase, which files matter most, who owns what, what's currently in review and generates a personalised starter map posted as a welcome issue. The new dev opens GitLab and finds everything they need waiting for them. No hand-holding from the team required. Day one confusion → instant clarity, in under 60 seconds.

How we built it

OnboardIQ has three components on the GitLab Duo Agent Platform:

Custom Skill ('skills/onboarding-context/SKILL.md') - teaches the agent how to extract file ownership, commit hotspots, contributor activity, and good-first-issues from the Orbit knowledge graph.

Agent Definition ('.gitlab/agents/onboardiq.md') - orchestrates the full flow: querying Orbit, identifying key people and files, surfacing open MRs, and generating the personalised onboarding doc.

CI Pipeline Trigger ('.gitlab-ci.yml') - a lightweight job that activates the agent on demand or via API webhook, keeping the architecture clean and GitLab-native.

Challenges we ran into

The hardest part was prompt design, getting the agent to produce specific, actionable output (real file paths, real @usernames, real issue numbers) instead of generic placeholder advice. The second challenge was scope: designing something that feels like a complete product but is actually shippable. The answer was to keep the architecture tight: one skill, one agent, one trigger, and make each piece do exactly one thing well.

Accomplishments that we're proud of

The CI pipeline passes cleanly, and the full agent-skill-flow architecture is wired together correctly on the GitLab Duo Agent Platform. More than the technical side, though, OnboardIQ solves a problem every dev team has, using GitLab Orbit exactly as it was designed to be used. It feels like something you'd actually want installed on your real project, not just a hackathon demo.

What we learned

GitLab Orbit's knowledge graph is more powerful than it looks on the surface; it models code relationships, ownership, and change history in ways that make genuinely intelligent tooling possible. I also learned how GitLab Duo skills and agents compose together: skills teach domain knowledge, agents orchestrate actions, and CI triggers connect everything to real GitLab events. The platform is designed for exactly this kind of AI-native workflow.

What's next for OnboardIQ

• Automatic webhook trigger on 'member_added' GitLab event - zero manual steps needed
• Skill-area detection: frontend dev joins → frontend files prioritized; backend dev → API modules first
• Orbit graph visualisation embedded directly in the welcome issue
• Cross-project onboarding: works across an entire GitLab group, not just one repo
• Slack/Teams notification with the onboarding link the moment it's generated

Built With

• ci/cd
• markdown
• orbit
• yaml