Adding E2E tests with an agent

This suite is built to be extended by a coding agent. You describe the feature you want covered; the agent runs a proven loop and leaves you with a working, locally-verified Playwright test.

Two ways to start

Type the command: /comet:add-e2e-test — then describe the flow you want covered.
Just ask in plain English: tell Claude Code "add an e2e test for the experiments comparison page" (or "…for the feature I just built", or "…for this branch"). It routes to the same procedure.

Both land in the writing-e2e-tests skill, which carries the full procedure and the suite's conventions.

What the agent does

A five-step loop — analyze the feature + frontend code → explore the live UI with the Playwright MCP → write the Page Object Model + spec → run it locally until green — with two lightweight checkpoints where it confirms direction with you (scope, and the live-UI discovery findings). The step-by-step is in the writing-e2e-tests skill.

Prerequisites

Opik running locally (./opik.sh, frontend at http://localhost:5173). That's the default target.
The Playwright MCP servers ship in the repo's .mcp.json — no setup needed.

Going deeper

The procedure, conventions, and the map of where files land (tests/, pom/, fixtures/, the SDK bridge) live in the writing-e2e-tests skill and its conventions.md. The live-UI selector hunt is the playwright-pom-discovery skill. See README.md for the environment-variable surface and quick start.