Inspiration
As a systems engineer, I’ve spent countless hours manually debugging concurrency issues like goroutine leaks or channel deadlocks. These "silent killers" of performance are often missed by standard health checks. I wanted to build a "Digital Teammate" that doesn't just chat, but actually understands the forensic runtime pathologies of a Go application.
What it does
godaktari is a specialized GitLab Duo agent that functions as a "Doctor" for Go source code. It reasons across Seven Pathology Classes:
- Goroutine Lifecycle Violations
- Channel Pathologies
- Sync Primitive Misuse
- Race Conditions
- Context Misuse
- Backpressure and Flow Control
- Atomic and Memory Order Violations
It doesn't just find bugs; it produces a precise diagnosis, applies a surgical fix, verifies the build passes, and opens a Merge Request with Go memory model references.
How we built it
The agent is written in Go and utilizes the GitLab Duo SDK. A unique aspect of this project is the development environment: the entire agent was built, compiled, and tested in a terminal-native environment using Termux on Android, utilizing Neovim and the Go toolchain.
Challenges we ran into
The biggest hurdle was mapping abstract SRE concepts into a reliable agentic workflow. Training the agent to differentiate between a healthy "waiting" goroutine and a "leaked" one required fine-tuning how the Duo SDK interacts with system-level diagnostic data.
Accomplishments that we're proud of
I successfully created a specialized agent that solves a real-world, high-stakes problem for backend developers. I am especially proud of maintaining a high-performance terminal workflow throughout the build.
What we learned
Building this taught me the power of the GitLab Duo Agent Platform in moving beyond simple chatbots to creating functional agents that can handle complex, low-level system forensics.
Log in or sign up for Devpost to join the conversation.