Orbit Sentinel
đź’ˇ Inspiration
In complex, high-performance software architectures—such as low-level parallel runtimes or distributed financial engines—architectural drift is a silent killer. A single rogue #include or unapproved dependency can break strict compilation boundaries, introduce massive technical debt, and destroy system performance.
Manual code reviews caught these too late, and standard CI/CD pipelines only catch them after the code has been pushed to the remote server. We were inspired to build Orbit Sentinel: an autonomous, local-to-remote pre-push gatekeeper that acts as an deterministic immune system for codebases, intercepting structural drift before it ever leaves a developer's machine.
⚙️ How We Built It
Orbit Sentinel is architected as a highly optimized, local-to-remote agentic framework:
- The Core Orchestrator (
agent.py): A lightweight execution engine that hooks directly into the local Git lifecycle (pre-push/pre-commit). It extracts code deltas and file path modifications instantly. - Knowledge Graph Architecture (
tools.py): Uses structured rules to model codebase dependency boundaries and team ownership domains, treating software architecture as a deterministic graph. - GitLab AI Catalog Integration: Tied directly into the GitLab Duo Agent Platform using a fine-grained, multi-step Flow Engine Schema for managed, event-driven policy verification.
🛑 Challenges We Faced
- Asynchronous Integration: Bridging local git states smoothly with remote server tracking without blocking developer velocity required rigorous state stashing and atomic validation logic.
- Strict History Merging: Handling complex Git histories during automated deployments where local testing artifacts conflicted with cloud-scaffolded templates taught us deep lessons about forced index tracking, branch rebasing, and secure credential handling.
- Fine-Grained Authentication: Navigating enterprise-level security protocols required moving away from blunt authentication methods to modern, scoped, resource-constrained Fine-Grained access frameworks.
đź§ What We Learned
We gained deep insights into building event-driven AI primitives inside production environments. We learned how to transform standard developer workflows into deterministic security gates, mastering the balance between flexible multi-agent reasoning and deterministic script execution.
Log in or sign up for Devpost to join the conversation.