Inspiration
New engineers and automated AI agents can inspect code, but they still lack the operational architecture map: which files, modules, projects, owners, pipelines, and risks are connected. That missing context slows onboarding and makes agents brittle.
What it does
Orbit Graph is a GitLab Duo Agent Skill and custom-agent starter that turns a natural-language architecture prompt into GitLab Orbit graph queries and Mermaid diagrams. A user can ask: "Map architecture around this import" or "What breaks if we change this service?" Orbit Graph plans the relevant Orbit query, shows the exact query envelope for auditability, runs it through Orbit or an Orbit-shaped fixture, and renders the result as a repository architecture diagram.
How I built it
The project includes a project-level skill at .agents/skills/orbit-graph/SKILL.md, an editable agent skill copy, an AI Catalog system prompt, a Python CLI, query planners for architecture/blast-radius/CI/security/ownership workflows, Mermaid rendering, tests, and a real public-repo demo using gitlab-org/cli CI utility imports.
GitLab Orbit usage
Orbit Graph is designed to call GitLab Duo Agent Platform graph tools such as get_graph_schema and query_graph, or glab orbit remote query locally. It targets graph entities including Project, File, Definition, ImportedSymbol, Pipeline, Vulnerability, MergeRequest, and User.
Challenges
Live Orbit access depends on an Orbit-enabled namespace. To keep the demo reproducible under hackathon time pressure, the project ships a real gitlab-org/cli fixture shaped like Orbit results while preserving the same generated query envelope.
What's next
Publish the agent to the GitLab AI Catalog, add larger multi-query graph clustering, export diagrams as SVG, and collect architecture prompt packs for real engineering teams.
Log in or sign up for Devpost to join the conversation.