9lives is an open-source testing framework for coding agents. It runs your existing Playwright tests from a native CLI, and each result comes with the assertions and execution receipt behind it. That way you, or your agent, can check what actually passed. In our Demo Bank example, we deliberately broke a sign-in button locator. 9l repaired it offline, with no model calls, and reran the spec in isolation. It saved the passing candidate separately for review and left the original test untouched. You can reproduce it from the repo with npm run demo:bank. Beyond healing, 9l assess flags problems such as fixed waits and missing assertions. Agents can drive the runner directly through MCP (claude mcp add 9lives -- 9l mcp), and an optional QA skill helps them write and check Playwright tests. New in 0.6.0: experimental Android support through Appium and UiAutomator2. So far it's tested on API 35 emulators and one Android 17 phone. It's Apache-2.0, runs on Linux, macOS and Windows, and needs no QualityMax account. Healing and Android are still experimental. A passing rerun can miss a wrong click when the assertions are weak. Feedback from early users has already led us to work on better target evidence and checks for competing repairs. Try it on one failing spec from your suite. I'd love to hear whether reviewing the repair saves you time, and what you'd need to see before trusting it. Forks, issues and PRs are welcome.

9lives

Open-source e2e testing for coding agents

Open-source e2e testing for coding agents

Forum Threads

Fixing the locators has never been easier!

Forum Threads

Fixing the locators has never been easier!