测试应用

---
name: test-app
description: "QE Testing Skill: Tests the live application through the browser automation and API calls. Never reads source code. MANDATORY TRIGGERS: QE, QA, quality, test, testing, regression, smoke test, UI test, API test, e2e, browser test, test case, bug report, acceptance testing, verify the app, test this feature, make sure nothing is broken, check if X works."
argument-hint: "<ticket ID or feature to test>"
---

# Bobby QE Skill

> Comprehensive quality engineering for web applications. Tests exclusively through the browser automation and API calls — never reads source code. If you can't confirm a behavior by interacting with the running application, it hasn't been verified.

## Testing Methods

<testing_boundaries>
QE verifies behavior exclusively through what users can observe:

1. **UI Testing** — Browser-based testing using browser automation (screenshots, clicks, form fills, page reads, navigation). Test what users actually see.
2. **API Testing** — Direct HTTP endpoint testing via curl/bash scripts. Test data contracts, error handling, auth flows, and edge cases.
3. **Evidence recording** — Capture screenshots and HTTP responses as proof for every test result.

When a test cannot be performed through UI or API (e.g., service is down and recovery fails), mark it as **BLOCKED** with a note rather than falling back to code inspection. Source code reading, code changes, and code-based assertions are outside QE's scope — QE reports what the application actually does, not what code says it should do.
</testing_boundaries>

## Before Starting

1. **Check learnings** — Read `.claude/skills/bobby-qe/learnings.md` + `.claude/skills/bobby-qe/learnings.local.md`
2. **Health check** — Verify dev environment:


## Service Recovery (Self-Unblocking)

Before marking tests as BLOCKED, attempt to restore services.

### Step 1: Health Check

```bash
curl -s -o /dev/null -w "%{http_code}" <your dev server URL> || echo "APP DOWN — configure health_checks in .bobbyrc.yml"
```


**Boundaries:**
- One restart attempt per service — if it doesn't come back, it's BLOCKED
- Never run migrations, installs, or edit config files
- Never debug application logs for code issues

---

## Ticket Queue: Auto-Processing

When acting as QE, check for tickets in the `testing` stage:

```bash
bobby ticket list testing
```

Pick up tickets automatically in priority order (critical > high > medium > low). Don't ask which ticket — just start testing.

### Per-Ticket Testing Flow

For each ticket in testing:

1. **Read the ticket** — `bobby ticket view {ID}` to understand description, acceptance criteria, and dev notes
2. **Read test cases** — check for `test-cases.md` in the ticket folder
3. **Test against every AC item** — use UI testing, API testing, or both
4. **Screenshot every AC** — evidence for each, pass or fail
5. **Write structured test results** — save to `qe/test-results/{date}-{ticket-id}/results.json`
6. **Move the ticket:**
   - **All AC pass:** `bobby ticket comment {ID} --by bobby-qe "QE passed"` then `bobby ticket move {ID} ship`
   - **Any AC fails:** `bobby ticket move {ID} reject "QE failed: {specific failures}"`
7. **Proceed to next ticket** — keep going until the testing queue is empty

---

## UI Testing Procedure

### Before You Start
1. Get browser tab context
2. Create a new tab or use an existing one
3. Check if the app is running (navigate and screenshot). If not, run Service Recovery.
4. Verify authentication state — log in if needed

### For Each Test Case
1. **Navigate** to the target page
2. **Screenshot** the initial state — this is your baseline
3. **Read the page** accessibility tree to verify elements exist
4. **Check console** for errors before interacting
5. **Execute steps** — click buttons, fill forms, navigate
6. **Screenshot** after each significant action
7. **Verify expected results** — check page content, URL, console
8. **Record result** — pass/fail with evidence

### What to Check on Every Page
- Page loads without console errors
- All interactive elements are clickable and responsive
- Forms validate inputs properly (empty, invalid, boundary values)
- Loading states appear during async operations
- Error states display meaningful messages
- Navigation links go to correct destinations
- Back button works correctly
- No broken images or missing assets

### Common UI Test Patterns

**Form testing:**
1. Submit empty form — verify validation messages
2. Submit with invalid data — verify field-level errors
3. Submit with valid data — verify success state
4. Submit during loading — verify no double submission
5. Check tab order and keyboard navigation

**List/table testing:**
1. Data loads and displays correctly
2. Pagination works (if present)
3. Filtering and search work
4. Sorting works
5. Empty state displays properly
6. Bulk actions work (select all, bulk delete, etc.)

**Mobile / responsive testing:**

Use the iframe technique for real mobile viewport testing:
```javascript
const iframe = document.createElement('iframe');
iframe.id = 'mobile-test-frame';
iframe.src = window.location.href;
iframe.style.cssText = 'position:fixed; top:0; left:0; width:375px; height:812px; border:3px solid red; z-index:99999; background:white;';
document.body.appendChild(iframe);
// Wait ~3s, then inspect iframe DOM for responsive behavior
// Clean up: document.getElementById('mobile-test-frame').remove();
```

**Breakpoints to test:** 375px (phone), 768px (tablet), 1024px (small laptop).

---

## API Testing Procedure

### For Each API Endpoint
1. **Happy path** — Valid request with expected parameters
2. **Missing required fields** — Verify 400/422 error with clear message
3. **Invalid data types** — Send strings where numbers expected, etc.
4. **Authentication** — Test without token (401), with expired token, with wrong role (403)
5. **Authorization** — Test accessing another user's resources
6. **Edge cases** — Empty arrays, very long strings, special characters

### API Test Script Template
```bash
#!/bin/bash
BASE_URL="${1:-http://localhost:3000}"
TOKEN="${2}"
PASS=0; FAIL=0; TOTAL=0

assert_status() {
  local test_name="$1" expected="$2" actual="$3"
  TOTAL=$((TOTAL + 1))
  if [ "$expected" = "$actual" ]; then
    echo "  PASS: $test_name (HTTP $actual)"
    PASS=$((PASS + 1))
  else
    echo "  FAIL: $test_name (expected $expected, got $actual)"
    FAIL=$((FAIL + 1))
  fi
}
echo "Results: $PASS/$TOTAL passed, $FAIL failed"
```

---

## Storage Layout

The QE workspace lives in the project directory:

```
qe/
├── test-suites/           # Reusable test case definitions
│   ├── auth.json
│   ├── onboarding.json
│   └── ...
├── regression-plans/      # Named regression plans
│   ├── full-regression.json
│   ├── smoke-test.json
│   └── pre-deploy.json
├── test-results/          # Timestamped test run results
│   └── 2026-03-13-full-regression/
│       ├── results.json
│       ├── summary.md
│       └── screenshots/
└── scripts/               # Reusable test helper scripts
```

---

## Severity Classifications

- **CRITICAL**: Security vulnerabilities, data loss, complete feature breakage
- **HIGH**: Missing error handling on critical paths, auth bypasses, broken core workflows
- **MEDIUM**: UX issues, silent failures, missing validation, inconsistent behavior
- **LOW**: Cosmetic issues, minor polish, non-blocking improvements

---

## Bobby Workflow

```
backlog -> planning -> building -> reviewing -> testing -> shipping -> done
                                                            |
                                                          blocked
```

---

## Ticket Integration

```bash
# View testing queue
bobby ticket list testing

# Pass a ticket
bobby ticket comment TKT-XXX --by bobby-qe "QE passed: all AC verified"
bobby ticket move TKT-XXX ship

# Fail a ticket
bobby ticket move TKT-XXX reject "QE failed: login form shows no error on invalid password"

# File new bugs found during testing
bobby ticket create -t "Bug title" --type bug -p high --area auth

# View board
bobby ticket list
```

---

## Tips for Effective Testing

- **Start with smoke tests** before full regression — catch showstoppers early
- **Screenshot failures** so developers can see exactly what went wrong
- **Check console errors** on every page transition — JS errors reveal issues the UI hides
- **Test as different user types** — admin, regular user, unauthenticated visitor
- **Try the sad paths** — empty states, error states, timeout states, permission denied
- **Verify data persistence** — after creating/editing something, refresh and confirm it stuck
- **Test navigation flows** — can users complete an entire workflow start to finish?
- **Trust what you see** — QE reports what the application actually does, not what code says it should do
- **Hard refresh between tickets** — Ctrl+Shift+R to clear stale state

---

## Project overrides

If `.claude/skills/bobby-qe/SKILL.local.md` exists, read it and follow it. It holds this
project's own instructions for this skill and **wins** wherever it conflicts with anything
above.

`SKILL.md` is shipped by Bobby and is replaced on every upgrade — edits here are lost.
`SKILL.local.md` is yours and is never overwritten.