Browser QA Agent

The Browser QA agent (AG-BROWSER-QA) provides agentic browser automation for exploratory UI testing. It executes YAML-defined test scenarios against running web applications using Playwright CLI, captures screenshot evidence at each step, and reports results with probabilistic pass rates.

When to Use

Use this agent when:

• You need to validate UI behavior in a running application
• You want automated screenshot evidence for visual verification
• You need to run multi-step user workflow validation
• You want accessibility checks via Playwright's accessibility tree
• You need to verify design tokens in a live application
• You want exploratory testing beyond deterministic unit tests

How It Works

1. Loads test scenario - Reads YAML test spec with steps, selectors, and assertions
2. Verifies prerequisites - Checks target URL accessibility and Playwright installation
3. Executes steps - Navigates, clicks, fills forms, and asserts conditions
4. Captures evidence - Takes screenshots at key steps for visual verification
5. Handles failures - Classifies errors (timeout, assertion, agent-error) and retries
6. Generates report - Produces results.json with pass rates and evidence paths
7. Updates status - Marks story as agentic-validated if pass rate meets threshold

Key Concepts

Pass Rate Threshold

Browser tests are probabilistic, not deterministic. Non-determinism is expected and not a bug.

Error Classification

YAML Test Spec Format

Copytest_id: AGENTIC-001
story_id: US-0050
name: User Login Flow
url: http://localhost:3000/login
timeout: 60s
max_retries: 2
pass_rate_threshold: 0.80
 
steps:
  - name: Navigate to login page
    action: navigate
    url: /login
    wait_for: "[data-testid='login-form']"
    screenshot: true
 
  - name: Fill credentials
    action: fill
    fields:
      - selector: "[data-testid='email-input']"
        value: "test@example.com"
 
  - name: Submit form
    action: click
    selector: "[data-testid='login-button']"
    screenshot: true
 
  - name: Verify dashboard
    action: assert
    assertion: "User sees dashboard"
    wait_for: "[data-testid='dashboard']"
    screenshot: true

Evidence Storage

Evidence is stored at: .agileflow/ui-review/runs/{timestamp}/{test_id}/

Each run produces:

• Screenshot files for each step (step-1-navigate.png, step-3-submit.png)
• Failure screenshots with _FAILED suffix
• results.json with full pass rate data and timing

Tools Available

This agent has access to: Read, Write, Edit, Bash, Glob, Grep

Differences from Related Agents

Example Usage

CopyTask(
  description: "Run browser QA on login flow",
  prompt: "Execute the YAML test scenario at .agileflow/ui-review/specs/login-flow.yaml against http://localhost:3000. Capture screenshot evidence and report pass rate.",
  subagent_type: "agileflow-browser-qa"
)

Related Agents

• testing - Deterministic test strategy and implementation
• qa - Test planning and release readiness
• ui-validator - Static UI code validation
• ci - CI/CD pipeline integration