nora-weisser

AI Assisted Testing Demo — Testing skill for Claude Code

Testing community

Companion repo for "Just Enough Context: Teaching Claude to test with Playwright".

How to install AI Assisted Testing Demo

This entry records only its repository, not the path inside it, so there is no exact command to give. Open nora-weisser/ai-assisted-testing-demo and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What AI Assisted Testing Demo does

Companion repo for "Just Enough Context: Teaching Claude to test with Playwright". Three context layers (AGENTS.md, skills, CLI) for creating tests and fixing flaky ones, with a concert-ticket app and five flaky-test exercises.

Alternatives in Testing

Sentient — Test prompt to showcase claude code hook blocking a dangerous rm -rf command 3.4k ★
Flaky Fix — 不安定なテスト（flaky test）の原因を分析し、テストの安定化を試みます 974 ★
Grid Tester — Use the @grid-tester agent for test strategy, TDD, or flaky test diagnosis: $ARGUMENTS 523 ★

README

Just Enough Context: Teaching Claude to test with Playwright.

Companion repo for the talk.

📖 [Command reference](docs/commands.md) · 🧪 [Exercises](docs/exercises.md)

If you're here straight after the session, the two-minute version is below. Everything you watched is reproducible from this repo.

First presented at the [BrowserStack Meetup Amsterdam: AI's Impact on QA](https://luma.com/browse-mtmg), August 2026, with a todo app. That version is tagged [`v1-meetup`](https://github.com/nora-weisser/ai-assisted-testing-demo/tree/v1-meetup).

The idea

Most context problems are filing problems. Each layer answers exactly one question — which is why none of them has to be long.

Does every task need it? -> AGENTS.md: loaded every session, keep it short.
Do some tasks need it? -> A Skill: loaded only when the task matches.
Only true right now? -> The CLI: fetched on demand, stays on disk.

The prompt is just the task. Everything else the agent needs should already be in one of the three layers.

The failure mode isn't too little context or too much. It's the right fact in the wrong place.

Two tasks, one app

**Front Row** is a small concert-ticket app that behaves like a real one: every API call takes 50–350 ms, search is debounced, favourites are optimistic, seat holds expire, payments are confirmed in the background, and seats are shared between customers.

Create a test case. Automate TC-S06 from the test plan. The trap: the favourite button flips before the server has saved it, so the obvious test is flaky.
Fix a flaky test. Five tests in tests/specs/flaky/ pass sometimes. Find the cause in the traces, fix it without retries or sleeps, prove it over 20 runs.

The same three layers do both jobs: the facts in `AGENTS.md` stop the agent writing a flaky test, and the same facts help it find the cause when one exists.