dev-test

YOU MUST WRITE E2E TESTS FOR USER-FACING FEATURES. This is not negotiable.

When your changes affect what users see or interact with, you MUST:

1. Write an E2E test that simulates user behavior
2. Run it and verify it PASSES (not just unit tests)
3. Document: "E2E: [test name] passes with [evidence]"
4. Include screenshot/snapshot for visual changes

Unit tests prove components work. E2E tests prove YOUR feature works for users.

E2E Facts

• Manual testing leaves no evidence and misses regressions — "it worked when I tried it" becomes an unverified claim the moment the code changes again. Capture the manual check as an automated test.
• Backend-only changes still need E2E when they affect user-visible output; "this is just backend" classifies by where the code lives, not by what the user sees.
• "The UI is unchanged" is an assumption until a visual snapshot proves it; asserting it without one is an unverified claim presented as fact.

Fake E2E Detection - STOP

If your "E2E test" does any of these, it's NOT E2E:

The test: If removing the actual implementation still passes your "E2E test", it's fake.

Example of fake E2E that caught nothing:

# FAKE E2E - only checks logs
def test_icon_theme_change():
    run_command("set-theme papirus")
    logs = read_logs()
    assert "papirus" in logs  # ❌ FAKE - only proves code ran
    # BUG: 89% of icons weren't changed, test still passed!

Real E2E that would have caught the bug:

# REAL E2E - verifies actual output
def test_icon_theme_change():
    run_command("set-theme papirus")
    screenshot = capture_desktop()
    assert visual_diff(screenshot, "expected_papirus.png") < threshold  # ✅ REAL
    # This would have shown 89% of icons were wrong

YOU MUST USE CHROME MCP FOR API/CONSOLE DEBUGGING. NO EXCEPTIONS. YOU MUST USE PLAYWRIGHT MCP FOR CI/CD TESTING. NO EXCEPTIONS.

Quick Decision Table

Capability Comparison

Browser MCP Facts

• Playwright MCP cannot read console messages or network requests — debugging those requires Chrome MCP (read_console_messages, read_network_requests). Chrome MCP cannot run headless — CI/CD requires Playwright MCP. Choosing by familiarity instead of by these constraints produces a test that cannot observe what it claims to verify.
• An API response inferred from code is not the real response — real data differs. Asserting behavior from inference instead of read_network_requests is an unverified claim presented as fact.

Platform Detection

Detect the operating system and display server to select the appropriate testing tool:

# Detect platform for desktop automation
case "$(uname -s)" in
    Darwin) echo "macOS - use dev-test-hammerspoon" ;;
    Linux)
        if [ "$XDG_SESSION_TYPE" = "wayland" ]; then
            echo "Linux/Wayland - use dev-test-linux (ydotool)"
        else
            echo "Linux/X11 - use dev-test-linux (xdotool)"
        fi
        ;;
esac

Desktop Automation Decision Tree

┌─────────────────────────────────────────────────────────────────┐
│                   DESKTOP AUTOMATION REQUIRED?                   │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
                    ┌─────────────────┐
                    │   Platform?      │
                    └─────────────────┘
                    /        |         \
                 macOS    Linux      Windows
                   │         │           │
                   ▼         ▼           ▼
        ┌──────────────┐ ┌─────────┐ ┌─────────┐
        │ HAMMERSPOON  │ │ LINUX   │ │ NOT     │
        │ (dev-test-   │ │ (dev-   │ │ SUPPORTED│
        │ hammerspoon) │ │ test-   │ └─────────┘
        └──────────────┘ │ linux)  │
                         └─────────┘
                              │
                    ┌─────────┴─────────┐
                    │   Display Server?  │
                    └───────────────────┘
                         /         \
                    Wayland        X11
                       │            │
                       ▼            ▼
                 ┌──────────┐ ┌──────────┐
                 │ ydotool  │ │ xdotool  │
                 └──────────┘ └──────────┘

Sub-Skills Reference

# Find test directory
ls -d tests/ test/ spec/ __tests__/ 2>/dev/null

# Find test framework
cat package.json 2>/dev/null | grep -E "(test|jest)"
cat pyproject.toml 2>/dev/null | grep -i pytest
cat Cargo.toml 2>/dev/null | grep -i "\[dev-dependencies\]"
cat meson.build 2>/dev/null | grep -i test

# Run with test inputs
./app --test-mode input.txt > output.txt

# Compare to expected
diff expected.txt output.txt

# Check exit code
./app --validate file && echo "PASS" || echo "FAIL"

## Test Run: [Description]

**Tool:** [Chrome MCP / Playwright / Hammerspoon / ydotool / pytest / etc.]

**Command:**
```bash
pytest tests/ -v

tests/test_feature.py::test_basic PASSED
tests/test_feature.py::test_edge_case PASSED
tests/test_feature.py::test_error FAILED

1 failed, 2 passed

## Integration

For TDD protocol (RED-GREEN-REFACTOR), see:

Read `${CLAUDE_SKILL_DIR}/../../skills/dev-tdd/SKILL.md` and follow its instructions.

This skill is invoked by Task agents during `dev-implement` phase.