AutoHarness: Agentic CI/CD Architect

Inspiration

Developers spend roughly 30% of their time maintaining CI/CD pipelines and debugging broken builds rather than writing new features. We wanted to build a tool that acts as a "Senior DevOps Engineer" inside the repository—an autonomous agent that doesn't just notify you of a failure, but actually fixes it.

What it does

AutoHarness is an autonomous agentic CI/CD architect. Powered by Codex and GPT-5.6, it monitors your repository in real-time to:

Self-Heal Pipelines: Automatically identify the root cause of a failed build and open a PR with the corrected code or configuration.
Auto-Generate Test Harnesses: Write comprehensive unit and integration tests based on new feature logic.
Document Architectural Debt: Map out the system architecture and flag areas where code complexity exceeds safety thresholds.

How we built it

We integrated the GitHub Actions API with OpenAI’s GPT-5.6 for high-level reasoning and Codex for precise, syntax-perfect code transformations. The agent uses a "Reflect-and-Execute" loop: analyzing logs, reflecting on the architecture, and then using Codex to implement the necessary fixes or tests.

Challenges we ran into

The primary challenge was managing "context awareness" for massive codebases. We solved this by implementing a vector-based retrieval system (RAG) that allows GPT-5.6 to focus only on the relevant modules while Codex handles the code generation.

Accomplishments that we're proud of

We successfully demonstrated a workflow where the agent detected a dependency conflict in a CI pipeline, researched the compatible versions, updated the package.json and Dockerfile, and passed the build—all without human intervention.

What we learned

We discovered that Codex performs significantly better when given "architectural context" rather than just raw code snippets. Agentic reasoning patterns are the future of developer tooling.