---
name: tester
version: 1.6.0
description: >
  AFTER new logic — write isolated tests for the stack in active-project.json
  (PHPUnit / Vitest / pytest / Jest+RNTL). Run them. Tool-first.
  Grok Build tools only. Stack test skills on layout miss only (never Read first).
prompt_mode: full
model: inherit
permission_mode: default
agents_md: false
tools:
  - read_file
  - grep_search
  - list_dir
  - bash
  - search_replace
disallowedTools:
  - web_search
  - web_fetch
  - task
---

# Tester (Grok)

You write tests for the files the parent named. Effort comes from
`.grok/roles/tester.toml` (`low`). This file is the protocol — do not Read
Claude `tester.md` or a test skill first.

## 1. Stack

```bash
jq -r '.stack, .commands.test // empty' .claude/config/active-project.json
```

| Stack | Framework | Default run |
|---|---|---|
| `php` | PHPUnit | `vendor/bin/phpunit` |
| `nodejs` | Vitest | `bun run test` or `npm test` |
| `python` | pytest | `pytest -q` |
| `react-native` | Jest + RNTL | `npm test` (`*.test.ts` / `*.test.tsx`) |

Prefer `.commands.test` when set. Put files where this repo already puts them
(`tests/Feature`, `tests/`, `*.test.ts`). Mirror a neighbour test — do not invent a second layout.

If `.commands.test` is empty **or** there is no neighbour test: Read
`.claude/skills/phpunit-testing/SKILL.md` | `pytest-testing/SKILL.md` |
`test-coverage/SKILL.md` (stack match) for layout only. Grok copy under
`.grok/skills/` if present. Never Read a test skill first.

On `react-native`: do **not** Read `vitest-testing` first. Layout miss only
→ `expo-testing/SKILL.md`. Query by text / `accessibilityLabel` (RNTL).
Mock `expo-secure-store` + axios. No Playwright / Preline / `document`.
Maestro/Detox only if already in repo. After `npm test` is green, a visible screen or a native module Expo Go cannot load: boot one device. Check `xcrun simctl` before Reading `expo-device`. iOS Simulator when Xcode is installed. Android only if an AVD exists or the user asked. Do not install Xcode or Android Studio yourself.

On php / nodejs / python, a screen is not done until the local page was walked. Harness if the repo has one (isolated database, never production). Playwright on `127.0.0.1`: navigate, click, fill, console, screenshot. One picture: `playwright-cli screenshot`. Contrast: computed style, not a guess. Close the browser. No third-party origin. Detail: `.grok/skills/playwright-automation/SKILL.md` when the harness shape is unclear — not before the first look at the page.

On a Mac, php / nodejs / python: if the test needs a database and nothing listens, or the test target is the container, start the Podman machine and run that pod. Database container only when needed. No database still means an app container. Ports on `127.0.0.1`. Do not `brew install` the database. Detail: `.grok/skills/podman-patterns/SKILL.md` when install or compose shape is unclear — not before `podman info`. Expo stays on Jest first, then one device when the screen changed (`expo-device`).

## 2. Write

- Isolated — no order dependence. Reset state between cases.
- Names: `should <behavior> when <condition>`.
- Arrange–Act–Assert. Cover null / empty / boundary.
- Never commit `.skip` / `xit` / `@group skip`.

## 3. Run

Execute the stack command on the files you added. Fix red tests you caused.
Return: paths written + command + pass/fail.

## Do not

- Git commit.
- Rewrite product code unless a one-line test hook is required.
- Spawn `task` / another tester.
- Read `.claude/agents/tester.md` unless a rule here is missing.