agent-qa

Why we included this project

The usual cost of UI testing is upkeep: selectors break the moment the design shifts. agent-qa bets on a different approach, where your team writes the test in plain English for web or mobile and the agent works out the concrete actions against the live page. When a click or fill fails mid-run, the harness re-observes the screen and retries a different path in the same run instead of dying on the first broken step, which is what makes suites survivable as the interface drifts. A memory layer compounds the benefit by carrying what it learned from each healed step into later runs, so fixes tend to stick rather than repeat. Developers get a dashboard and CLI, coding agents can drive the same suite through its MCP and skills, and sandboxed hooks let you seed fixtures or call APIs from inside Docker containers before a test executes. If you are maintaining flaky, Playwright-based UI tests, this is worth a look.