Open source
agentic-qa: the approach, in public
A sample project you can read, run and judge for yourself. It tests a demo shop we built, so it shows how the work is done, not results for a client.
What it is
agentic-qa is a set of Claude agents that handle the QA work on a project from start to finish. They plan from risk, write and run API and UI tests in Playwright + TypeScript, explore the app, verify and retest bugs, run regression and make a Go/No-Go call. Deterministic checks and human review sit on top of what they produce.
The app under test is ShopLite, a small shop with a REST API and a web interface that lives in the same repository. It has defects that can be switched on, to check whether the agents and the tests actually catch bugs.
Read this as a demo. Every number in the live report comes from the demo shop. None of it is a client result, and it is not presented as one.
What you can inspect
The same building blocks a client engagement uses, out in the open.
- Plan
A risk-based test plan, and acceptance criteria refined into testable ones.
- Automate
API tests for endpoints and journeys, and UI tests found by exploring the app.
- Explore
Charter-based ad hoc testing of the running app.
- Bugs
Verify, triage and retest bugs.
- Maintain
Impact-based regression, flaky-test hunting and test healing with a guard.
- Release
A smoke check, a test report and the Go/No-Go release gate.
The live report
On every push to the main branch, CI publishes a Go/No-Go assessment, a traceability table and a defect board to GitHub Pages. That is the same kind of report a client gets before each release.
Stack
- Claude Agent SDK, with skills and subagents
- Model Context Protocol (MCP) servers for knowledge and code search
- Playwright + TypeScript on Bun
- GitHub Actions
- MIT license
Want this on your product?
The service brings the same approach to your codebase: your repository, your CI, your release rhythm.
Or write to qa@releaseworthy.app.