What Inspired This Project

Accessibility is one of those requirements developers know matters but often struggle to implement correctly. The cognitive load of remembering WCAG guidelines while coding—or even while instructing AI coding assistants—makes it easy to ship regressions.

But this isn't just a nice-to-have: the European Accessibility Act (EAA) mandates accessibility compliance starting June 2025, making it a legal requirement for digital products. Manual accessibility reviews don't scale to every merge request, creating a compliance gap.

I wanted to build automation that catches these issues where they happen—at the MR level—before code ships to production. Something that works like a helpful reviewer who never forgets WCAG criteria.

What I Learned

This project was my deep dive into the GitLab Duo Agent Platform. I learned:

  • Flow orchestration patterns — How to chain agents sequentially and aggregate results
  • Tool usage — Working with list_merge_request_diffs, read_file, and create_merge_request_note
  • Prompt engineering for code analysis — Balancing detection accuracy with false positive prevention using confidence levels
  • Agent design philosophy — Splitting by task type (static analyzer → manual checks → reporter) rather than by domain area produces cleaner, more maintainable flows
  • GitLab-native constraints — Building useful automation without external dependencies or CI/CD integration

The platform's ability to read diffs and post contextual comments directly on MRs makes it perfect for this kind of pre-merge validation.

Built With

  • claude-(anthropic)
  • gitlab
  • yaml
  • yaml-for-flow-configuration
Share this project:

Updates