Fix test runner: proper exit codes and visible output (#49)

## TLDR

Fixed critical bugs in test runner AND achieved 100% coverage in
dart_node_mcp by fixing the coverage tool to merge data across test
files.

## What Does This Do?

### 1. Coverage Merging Fix (dart_node_mcp: <70% → 100%)
**Problem**: dart_node_coverage was only capturing 42/42 lines from
server.dart, completely missing mcp_server.dart (111 lines) and
stdio_transport.dart (6 lines). Total coverage was <70%.

**Root Cause**: Every test file called
`writeCoverageFile('coverage/coverage.json')` in tearDownAll,
**overwriting** each other's coverage data. The last test to finish
would wipe out all previous coverage.

**Fix**: Modified `dart_node_coverage/lib/src/runtime.dart` to **merge
coverage data** instead of overwriting:
- Added `readFileSync` to read existing coverage before writing
- Implemented `_mergeData()` to combine line execution counts across
test runs
- Now accumulates coverage from all test files: 159/159 lines = **100%
coverage**

### 2. Test Runner Fixes
Fixed two critical bugs in `tools/test.sh`:
- Test failures now properly fail the build (exit code 1)  
- Test output is now visible in real-time while also being logged to
files

**Bug 1: Silent Failures**
Parallel test runner spawned subshells that always exited with code 0.
Fixed by adding explicit exit codes.

**Bug 2: Hidden Output**  
All output redirected to log files only. Fixed using `tee -a` to stream
to both terminal and files.

## How Do The Tests Prove The Change Works?

### Coverage Verification
```bash
cd packages/dart_node_mcp
dart run dart_node_coverage:coverage -- \
  --project-root . \
  --packages dart_node_mcp \
  --node-test-runner \
  --format lcov --output coverage/lcov.info
```

Results:
- server.dart: 42/42 lines (100%)
- mcp_server.dart: 111/111 lines (100%)  
- stdio_transport.dart: 6/6 lines (100%)
- **TOTAL: 159/159 lines = 100.0% coverage**

### Test Runner Verification  
- `./tools/test.sh --tier 1` - All tests pass, exit code 0
- `./tools/test.sh --tier 2` - Fails properly on dart_node_react (needs
Chrome), exit code 1 ✅
- `MIN_COVERAGE=99 ./tools/test.sh packages/reflux` - Coverage failure
detected, exit code 1 ✅
- All test output now visible in terminal while saved to `logs/*.log` ✅

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Author: MelbourneDeveloper

.claude/skills/setup-claude/SKILL.md

@@ -0,0 +1,130 @@
+---
+name: setup
+description: Install development tools (claude | playwright)
+disable-model-invocation: true
+allowed-tools: Bash
+---
+
+# Setup Development Tools
+
+Install Claude Code CLI with Too Many Cooks extension OR Playwright with Chromium.
+
+**Usage**: `/setup <tool>`
+- `claude` — Install Claude Code CLI and Too Many Cooks VSCode extension
+- `playwright` — Install Chromium and Playwright for website E2E testing
+
+---
+
+## Claude Setup
+
+Sets up Claude Code CLI and the Too Many Cooks VSCode extension for multi-agent coordination.
+
+### Step 1: Install Claude Code CLI
+
+```bash
+npm install -g @anthropic-ai/claude-code
+```
+
+Verify:
+```bash
+claude --version
+```
+
+### Step 2: Build the Too Many Cooks VSCode Extension
+
+The extension provides visual agent coordination (locks, messages, plans) via the MCP server.
+
+#### 2a. Build the MCP Server first
+
+```bash
+cd examples/too_many_cooks && dart pub get && npm install
+dart compile js -o examples/too_many_cooks/build/bin/server.js examples/too_many_cooks/bin/server.dart
+dart run tools/build/add_preamble.dart \
+  examples/too_many_cooks/build/bin/server.js \
+  examples/too_many_cooks/build/bin/server_node.js \
+  --shebang
+```
+
+#### 2b. Build and package the extension
+
+```bash
+cd examples/too_many_cooks_vscode_extension
+npm install
+npm run compile
+npx @vscode/vsce package
+```
+
+This creates a `.vsix` file in `examples/too_many_cooks_vscode_extension/`.
+
+#### 2c. Install the extension
+
+```bash
+code --install-extension examples/too_many_cooks_vscode_extension/*.vsix
+```
+
+Or use the full build script that does steps 2a-2b:
+```bash
+bash examples/too_many_cooks_vscode_extension/build.sh
+```
+
+### Step 3: Configure Claude Code for this project
+
+The project's `.claude/settings.local.json` already has the required permissions:
+- `Bash(dart pub get:*)` — dependency installation
+- `mcp__too-many-cooks__register` — MCP agent registration
+- `Bash(docker ps:*)` — Docker status checks
+
+Custom skills are in `.claude/skills/` — run `/help` to see them.
+
+### Multi-Agent Usage
+
+After setup, agents coordinate via the Too Many Cooks MCP server:
+- **Lock files** before editing, unlock after
+- **Check messages** regularly between agents
+- **Update plans** so other agents can see your intent
+- Keep your agent key — it's critical for authentication
+
+---
+
+## Playwright Setup
+
+Sets up Playwright with Chromium for running the website's E2E test suite.
+
+### Steps
+
+1. **Install website npm dependencies** (includes `@playwright/test`):
+   ```bash
+   cd website && npm ci
+   ```
+
+2. **Install Chromium browser and OS dependencies**:
+   ```bash
+   cd website && npx playwright install --with-deps chromium
+   ```
+   This installs the Chromium binary plus required system libraries (libgbm, libasound, etc.).
+
+3. **Verify installation**:
+   ```bash
+   cd website && npx playwright --version
+   ```
+
+### Notes
+
+- Only Chromium is needed — this project does not test against Firefox or WebKit.
+- On the Dev Container (Ubuntu 24.04), `--with-deps` installs the OS packages automatically.
+- On macOS, Playwright downloads its own Chromium binary — no Homebrew needed.
+- Browser cache lives at `~/.cache/ms-playwright/`. Delete this to force a clean reinstall.
+- The website tests are at `website/tests/` and configured in `website/playwright.config.js`.
+- Base URL: `http://localhost:8080` (Eleventy dev server).
+
+### After installing
+
+Run the website tests:
+```bash
+cd website && npm test
+```
+
+Or with UI mode:
+```bash
+cd website && npm run test:ui
+```

.claude/skills/setup-playwright/SKILL.md

@@ -113,13 +113,21 @@ jobs:
           done
 
       - name: Test Tier 1
-        run: ./tools/test.sh --ci --tier 1
+        run: ./tools/test.sh --tier 1
 
       - name: Test Tier 2
-        run: ./tools/test.sh --ci --tier 2
+        run: ./tools/test.sh --tier 2
 
       - name: Test Tier 3
-        run: ./tools/test.sh --ci --tier 3
+        run: ./tools/test.sh --tier 3
+
+      - name: Upload test logs
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: test-logs
+          path: logs/
+          retention-days: 7
 
   too-many-cooks-mcp:
     name: Too Many Cooks MCP Server