Agent-first test automation for Playwright.
Agents that explore your app, write tests the way your team would, and audit their own work.
| Project | What it is |
|---|---|
| playgentic-agents | One plugin for Claude Code, GitHub Copilot and Codex that writes and heals Playwright tests in any repo, plain Playwright or playwright-bdd, greenfield or brownfield. |
| playgentic · npm | Scaffolds an opinionated Playwright-BDD suite: docstringed steps, page objects, a generated suite wiki, lint gates and CI. It wires up the agents in the same command. |
The two work separately or together. The plugin learns any suite's conventions. The scaffold gives it a suite with conventions worth learning.
New suite, agents included:
npx playgentic init my-suite --agents claude,copilot,codex # any subset, or "all"Existing repo, agents only:
/plugin marketplace add https://github.com/playgentic/playgentic-agents # Claude Code, then: /plugin install playgentic@playgentic
copilot plugin marketplace add https://github.com/playgentic/playgentic-agents # GitHub Copilot, then: copilot plugin install playgentic@playgentic
codex plugin marketplace add https://github.com/playgentic/playgentic-agents # Codex, then: codex plugin add playgentic@playgentic
Then ask your agent to run author-tests:
author-tests --gen cover applying and removing coupons at checkout
author-tests --heal the login scenario fails after the redesign
- Branch. Every run gets its own feature branch, and a worktree if the work is parallel.
- Learn. It reads your suite, your
CONVENTIONS.md, and your per-agent extensions. - Plan. It agrees numbered acceptance criteria with you before building anything.
- Explore → build. An explorer reads the live app. Builders write and run tests in your style.
- Audit. A read-only auditor checks that each test really verifies its criterion, not just that it passes.
- Hand off. It commits test files only, and never pushes without asking.
Your team's rules live in your repo (.playgentic/wiki/), not in the plugin, so updates never overwrite them.
- Your conventions win. The agents follow your code style and testing expectations over their own defaults.
- Honest coverage. No
.skip, no weakened assertions, no config tweaks to get green, unless you explicitly allow it. - Finding a bug is a success. Agents report real app defects instead of bending tests until they pass.
- Minimal by design. Markdown skills and agents, zero-dependency tooling, nothing to sync.