QA Tester

Overview

QA Tester is an agent that interactively validates actual application behavior in tmux sessions. It handles starting services, sending commands, capturing output, validating against expectations, and teardown.

Unit tests verify code logic, but QA testing verifies actual behavior. An application can pass all unit tests yet fail when actually run. Interactive testing via tmux catches startup failures, integration issues, and user-facing bugs.

When to Use

• When the automated test suite does not cover the behavior in question
• When interactive CLI input/output simulation is needed
• When service startup/shutdown testing is needed
• When streaming/real-time behavior validation is needed

When Not to Use

• When the project has tests covering the feature (running tests is more efficient)
• For simple command verification (direct execution is more efficient)
• For static code analysis (architect is more suitable)

Usage Examples

"Actually test that the API server responds correctly"
"Validate the interactive features of this CLI tool"
"Check the /health endpoint after service startup"

Testing Process

1. Prerequisite Check: Verify tmux is installed, port is available, directory exists. Fail immediately if not met
2. Setup: Create a tmux session with a unique name (qa-{service}-{test}-{timestamp}), start the service, wait for ready signal
3. Execute: Send test commands, wait for output, capture with tmux capture-pane
4. Validate: Compare captured output against expected patterns. Report PASS/FAIL
5. Cleanup: Terminate tmux session and remove artifacts. Always clean up, even on failure

Key Rules

• Unique session names: Use the format qa-{service}-{test}-{timestamp}. No generic names like "test"
• Wait for readiness: Do not send commands immediately after service startup. Poll for output pattern or port availability
• Capture output first: Always run tmux capture-pane before making assertions
• Always clean up: Always terminate the session even on test failure. Prevent orphan processes
• Appropriate delays: Short delays are needed between send-keys and capture-pane

Combining with Other Agents

• architect: Architect diagnoses and outputs a test plan; QA Tester executes it (diagnosis-validation loop)
• debugger: Root cause analysis of test failures is debugger's responsibility
• verifier: Automated test verification is handled by verifier

Reference