# qa-runner

> Agent skills that test one flow or a whole web app in a real browser, from a ticket, a spec, a pull request or a request in chat. They draft the test cases for approval, run them, back every verdict with marked screenshots and a recording, and leave the sign-off to a human. Cases the human approves are saved as scripts that replay without an agent, at $0 in tokens. Works with Claude Code, Codex, Cursor, Gemini CLI, OpenCode, GitHub Copilot CLI and other agents that run shell commands and MCP tools.

Install for every detected agent:

```bash
npx skills add he1ox/AgenticTestRunner
```

Then register the browser once: ask the agent to "set up qa-runner" (the qa-setup skill), and ask it to "test the checkout flow" or "run the test cases for SHOP-42". Requirements: Node.js 20+, npx (for Playwright MCP), ffmpeg only for recordings. No npm dependencies.

## Skills

- [qa-setup](https://github.com/he1ox/AgenticTestRunner/blob/main/skills/qa-setup/SKILL.md): prepare a project once; check Node, npx and ffmpeg, register the browser, write .qa/config.json
- [write-test-plan](https://github.com/he1ox/AgenticTestRunner/blob/main/skills/write-test-plan/SKILL.md): turn a ticket, a spec or a request in chat into .qa/plans/<KEY>.plan.json and get the case list approved
- [run-test-plan](https://github.com/he1ox/AgenticTestRunner/blob/main/skills/run-test-plan/SKILL.md): run the plan with browser agents, open the viewer, retest what a reviewer rejects
- [replay-test-plan](https://github.com/he1ox/AgenticTestRunner/blob/main/skills/replay-test-plan/SKILL.md): replay saved Test plans without an agent, repair stale scripts

## Docs

- [README](https://github.com/he1ox/AgenticTestRunner#readme): install per agent, project setup, CLI, CI mode, export to Playwright
- [Harness notes](https://github.com/he1ox/AgenticTestRunner/tree/main/skills/run-test-plan/references/harness): per-agent details
- [Changelog](https://github.com/he1ox/AgenticTestRunner/blob/main/CHANGELOG.md)