browser-test

Browser Test

Plan and execute browser-based UI testing in parallel using multiple agents.

Input

The user input is: $ARGUMENTS

• Spec/plan path: Use this to understand what was built and derive test scenarios
• Empty: Auto-detect the most recent spec from .claude/plans/specs/ or plan from .claude/plans/

If no spec or plan is found, ask the user what feature to test.

Step 1: Read Context

1. Read the spec/plan to understand what was built
2. Read relevant source files to understand the actual implementation (component locations, routes, UI structure)
3. Identify the app URL (default: http://localhost:3000)

Step 2: Create Test Plan

Analyze the feature and group scenarios by independence:

For each feature area, create test scenarios covering:

• Happy path: Core functionality works as expected
• Edge cases: Empty states, loading states, error handling
• Interactions: Click, hover, expand/collapse, open/close
• Visual verification: Layout, text content matches spec

Group into parallel batches — different pages or scenarios that can each start from a fresh page load.

Present the test plan:

Test Plan: [Feature Name]

Batch 1 (Agent 1): [Description]
  - Scenario 1.1: ...
  - Scenario 1.2: ...

Batch 2 (Agent 2): [Description]
  - Scenario 2.1: ...

Wait for user approval. Max 4-5 parallel agents.

Step 3: Dispatch Parallel Test Agents

Dispatch each batch as a separate agent (haiku model). Read the agent prompt template from the tester-prompt.md file in this skill's directory and provide it as context along with the batch's specific scenarios.

Launch all batches in parallel using the Agent tool.

Why haiku: Browser test agents perform simple, mechanical work — navigate to URL, click elements, verify text content. This doesn't require deep reasoning, and haiku's speed means faster test cycles.

Step 4: Collect Results

Compile results from all batches:

Test Results: [Feature Name]

PASSED: X/Y scenarios
FAILED: Z/Y scenarios

Batch 1: [status]
  - Scenario 1.1: PASS
  - Scenario 1.2: FAIL - [reason]

If tests failed, analyze whether the failure is a bug or a test environment issue. Report with specific details.