xuxh21/ui-regression-recorder-skill

This project is a skill for AI coding agents (specifically OpenAI Codex) that changes how you record and reuse automated browser tests. Most browser testing tools let you record clicks once and replay them, but that one recording is hard to maintain and every new test starts from scratch. This project takes a different approach: it treats each recorded flow as a source of reusable building blocks. When you give the skill a completed recording, it breaks that recording into smaller named capabilities, such as "open the list page", "create a record", or "verify duplicate rejection". Those capabilities get saved to a registry file called flows.json. The next time you need a new test, the skill plans it from the business goal first, matches each required step to an existing capability, and only records new footage when a step is genuinely missing. The repository includes a template for the capability registry (flows.template.json) and a starter helper library for common Playwright browser actions (playwright-common-flows.template.ts). There is also a full operating guide (SKILL.md) that describes how to initialize, record, clean, compose, rerun, and debug test workflows using this system. The project is written in TypeScript and targets teams already using the Playwright testing library. Installation is done through a companion skill installer if you already have one configured in Codex, or by cloning the repository manually and copying it into the Codex skills folder. The current stable release is v0.1.4. Once installed, the skill runs tests in a visible browser window by default so a human can watch what happens, and if a test fails it backtracks through earlier steps to find where a selector or postcondition went wrong rather than patching blindly. The core value here is that the first recording is not thrown away or treated as a fixed script. It becomes a library of known actions that new tests can combine in different ways, reducing how often you need to record from the beginning.