Orbit CI Detective

Inspiration

Every developer knows this moment: CI turns red. You stare at logs full of Unauthorized, ImportError, exit code 1 — with no context about why it broke or how to fix it. GitLab Duo Chat didn't have Orbit context. We built Orbit CI Detective to fix that.

What it does

Orbit CI Detective is a GitLab Duo Skill built on the GitLab Duo Agent Platform that uses GitLab Orbit's knowledge graph to diagnose CI/CD failures with full project context.

Instead of just reading logs, the skill:

• Understands the dependency graph, job relationships, and code context via Orbit
• Identifies the root cause (not just the symptom)
• Produces a structured Solved Solution Report with: Issue, Root Cause, Fix Applied, Validation steps, Result, and Remaining Risk
• Flags Unknowns honestly (e.g., which auth method is used, whether a cert rotation occurred)

Two real CI failure types diagnosed:

1. deploy-production failing with kubectl apply Unauthorized — Kubernetes credentials expired; fix: regenerate kubeconfig and update CI/CD variable
2. pytest failing with ImportError: cannot import name 'BaseModel' from 'pydantic' — pydantic v2 installed where v1 API was expected; fix: pin pydantic>=1.10,<2.0

How We Built It

• Skill interface: SKILL.md defines the orbit-ci-detective skill in the GitLab Duo Agent Platform skill format
• Knowledge graph: GitLab Orbit provides structured, queryable context about the repository — dependencies, pipeline stages, job configs, code relationships
• GitLab Duo Chat: The skill is invoked via Duo Chat, receiving raw CI log snippets and returning structured diagnostic reports
• Repository: MIT-licensed project at https://gitlab.com/Aryan-del360/orbit-dev-unblock

Challenges We Ran Into

• Learning the GitLab Duo Agent Platform skill format and SKILL.md spec from scratch during the hackathon
• Crafting a skill prompt that produces consistent, structured output across many different CI failure types
• Making the output genuinely actionable — not just describing the problem but providing exact fix commands

Accomplishments That We're Proud Of

• Built a skill that correctly diagnosed two distinct real-world CI failure patterns
• The structured Solved Solution Report format gives developers everything they need in one response
• The skill is honest about unknowns — it doesn't hallucinate fixes for things it can't verify
• Fully MIT-licensed, open source, and ready for the AI Catalog

What We Learned

• GitLab Orbit's knowledge graph is a powerful but underutilized resource for CI/CD debugging
• Structured output formats (like the Solved Solution Report) dramatically improve developer trust in AI responses
• The GitLab Duo Agent Platform skill interface is elegant and easy to extend

What's Next for Orbit CI Detective

• Publish to the AI Catalog so any GitLab project can invoke the skill
• Auto-trigger diagnosis on pipeline failure via webhook + Orbit event
• Expand to cover more failure patterns: Docker build errors, test flakiness, security scan failures
• Add confidence scores to each diagnosis
• Integrate fix suggestions directly as MR drafts

Built With

• gitlab-ci/cd
• gitlab-duo-agent-platform
• gitlab-orbit
• kubernetes
• python
• skill.md