About the project Codex Sentinel is an autonomous DevOps agent designed to monitor, diagnose, and repair broken CI/CD pipelines instantly. By connecting directly to GitHub Webhook infrastructure, the platform intercepts workflow failures in real time, extracts the broken log context, utilizes advanced Large Language Models to isolate the root cause, and opens a fully formed self-healing Pull Request to fix the build without requiring human developer intervention.
Inspiration Software engineering teams waste thousands of developer hours hunting down trivial CI/CD breaks, broken dependencies, and syntax typos in deployment pipelines. We wanted to build a zero-touch agent that acts as a 24/7 on-call engineer, diagnosing and patching infrastructure code the exact moment it breaks.
What it does Codex Sentinel integrates directly into GitHub Webhooks. When a continuous integration run fails, it captures the error state, passes it to an intelligent diagnostic agent, checks out the repository locally into an isolated secure sandbox, builds a structured patch, and submits a ready-to-merge Pull Request to fix the build immediately.
How we built it Backend Infrastructure: Python, FastAPI, and Uvicorn for asynchronous event ingestion and background task offloading.
AI Core: OpenAI API integration utilizing state-of-the-art context window prompting to read log files and write structured code updates.
Automation Engineering: System subprocess management coupled with Git authentication token layers to automate repository sandboxing, branch operations, and remote synchronization.
Challenges we ran into Handling environment isolation on Windows systems while maintaining secure, real-time file access within dynamic sandbox spaces proved complex. We also navigated strict token scope permissions and signature authorization overrides to test payload streams seamlessly in a local development environment.
Accomplishments that we're proud of We successfully achieved an end-to-end autonomous healing loop. A developer can push broken code, watch the CI fail, and receive an active, fully functional corrective Pull Request in under 60 seconds without ever touching their terminal.
What we learned We learned how to design resilient background task threads in asynchronous Python applications and how to effectively structuralize raw text log data so LLMs can process deep contextual stacks cleanly.
What's next for Codex Sentinel Expanding sandbox operations into hyper-secure Docker environments, adding support for multi-step validation loops where the agent compiles and tests its own code inside the sandbox before opening the Pull Request, and integrating Slack notifications.
Log in or sign up for Devpost to join the conversation.