Inspiration
A failed CI pipeline often sends developers through lengthy logs, recent code changes, and repeated test runs. We want to shorten that investigation while keeping developers in control of code changes.
What it does
Pipeline Medic is being developed as a GitLab Duo agent that:
- Examines failed CI jobs and relevant code changes.
- Explains the likely root cause with references to log evidence.
- Proposes a focused fix and a verification plan.
- Runs relevant tests and reports the results.
- Requests human review before changes are merged.
How we plan to build it
We will use GitLab Duo Agent Platform to coordinate diagnosis, proposed changes, and verification in a GitLab project. CI/CD pipelines will provide observable evidence of whether each proposed fix works.
The initial scope is reproducible Python test failures. We will then evaluate whether the workflow can handle additional failure types.
Challenges to address
The agent must distinguish code defects from infrastructure failures, avoid changing tests merely to make a pipeline pass, and clearly report uncertainty when the evidence is insufficient.
Learning and evaluation goals
We will measure diagnosis accuracy, successful repair rate, and the number of unnecessary changes across controlled failure cases. We will document unsuccessful attempts alongside successful repairs.
Current status
Registration is complete. Implementation and evaluation are pending. This description is a planning draft and does not claim completed functionality.
Log in or sign up for Devpost to join the conversation.