AegisFlow AI

• 

  GitLab CI/CD pipeline showing successful Python tests and secret detection, with manual human approval required.

• 

  GitLab Duo Release Guardian AI session generating an explainable security assessment with a REVIEW REQUIRED recommendation.

Inspiration

Writing code is only one part of software development. Before releasing software, development teams must verify test results, review security risks, and determine whether a release is safe.

These activities can be time-consuming, and important evidence may be missing or difficult to interpret.

AegisFlow AI was created to explore how AI-powered automation can make release decisions safer, clearer, and more efficient while keeping humans in control.

What it does

AegisFlow AI is an intelligent DevSecOps workflow built around the GitLab Duo Agent Platform.

The project combines automated testing, secret detection, release risk assessment, and human oversight.

Its key capabilities include:

• Running automated Python tests through GitLab CI/CD.
• Checking repository content for potential exposed secrets.
• Applying release risk assessment logic.
• Using the AegisFlow Release Guardian workflow to analyze available project evidence.
• Generating an explainable release recommendation.
• Keeping release approval under human supervision.

The workflow supports three decision categories: APPROVED, REVIEW REQUIRED, and BLOCKED.

How we built it

AegisFlow AI was developed using Python, GitLab CI/CD, and the GitLab Duo Agent Platform.

We implemented a Python-based risk assessment engine, automated tests, security checks, and a CI/CD pipeline with a manual approval stage.

We also configured AI agents and a custom workflow called AegisFlow Release Guardian to review release-readiness evidence and explain potential risks.

The project is hosted in a GitLab repository containing the source code, pipeline configuration, and documentation.

Challenges we faced

One of the main challenges was connecting deterministic security and testing checks with AI-assisted release assessment.

Another challenge was designing a workflow that provides useful automation without treating incomplete evidence as proof that a release is safe.

We addressed this by emphasizing explainable recommendations and retaining human oversight of release decisions.

Accomplishments

• Implemented a working GitLab CI/CD pipeline.
• Created automated Python tests and secret detection checks.
• Integrated a custom GitLab Duo Agent Platform workflow.
• Demonstrated a completed AI analysis session.
• Generated a REVIEW REQUIRED recommendation when release-readiness evidence was incomplete.
• Preserved human control over the final release decision.

What we learned

This project demonstrated that AI can support DevSecOps by interpreting evidence and explaining risks, while automated checks provide repeatable technical validation.

We also learned the importance of distinguishing between a successful test run and complete release readiness.

What's next

Future improvements could include broader security scanning, richer testing evidence, automated collection of pipeline results, and more detailed release risk policies.

Our long-term goal is to make AegisFlow AI a reusable assistant for teams that need safer, more transparent software releases.

Built With

• cd
• ci
• duo
• gitlab
• pytest
• python
• yaml