Hi HN, I built this to solve a problem I kept hitting: AI agents can generate test scripts fast, but without enforcement they produce inconsistent output, skip patterns, and repeat the same mistakes.

The platform uses a 5-layer test architecture (Test → Role → Task → Page Object → BrowserInterface) based on the Screenplay pattern, enforced by a governance loop that runs inside the AI agent. Every 10 actions, the agent stops, re-reads its protocol, and checks its own work. If it fails, it records a lesson and never makes that mistake again.

The kernel follows a minimalistic design. no external dependencies. I'll post about the kernel in a separate post.

Stack: Python, Playwright MCP, Claude Code.

Happy to answer questions.