qa-exploring-tester（qa 探索测试员）

ExploreAI Tester

An autonomous quality engineering framework that combines AI-driven exploratory testing with deterministic scripted validation to maximize defect detection across the full application stack.

How It Works

This framework operates through 8 specialized subagents coordinated by an orchestrator, each backed by dedicated skills. Instead of brittle selector-based or pixel-matching approaches, it uses semantic UI understanding (visual AI) to interpret applications the way a human tester would.

Architecture Overview

User Request
     │
     ▼
┌─────────────────────────┐
│   Orchestrator (you)     │
│   Routes to agents       │
└─────┬───────────────────┘
      │
      ├──► E2E Suite
      │    ├─ qa-e2e-exploratory-agent      (explores UI autonomously)
      │    └─ qa-e2e-deterministic-runner   (scripted critical paths)
      │
      ├──► Integration Suite
      │    ├─ qa-api-contract-fuzzer        (behavioral API fuzzing)
      │    └─ qa-multi-service-validator    (cross-service transactions)
      │
      ├──► Visual Suite
      │    ├─ qa-semantic-visual-regression (AI-powered UI diff)
      │    └─ qa-cross-browser-explorer     (browser/viewport matrix)
      │
      └──► Core Infrastructure
           ├─ qa-test-oracle-synthesizer    (generates expected behavior)
           └─ qa-benchmark-evaluator        (measures effectiveness)

Quick Start

When a user asks you to test something, follow this decision tree:

1. Identify what they want tested — UI? API? Visual fidelity? All of the above?
2. Gather inputs — URL, API specs, design references, browser matrix, credentials
3. Select and spawn agents — based on the testing scope (see Agent Selection below)
4. Collect and synthesize results — merge bug reports, generate summary

Agent Selection Guide

Required Inputs

Before spawning agents, collect these from the user (not all are required for every scenario):

• Target URL / staging environment — where to point the agents
• API specs (OpenAPI/GraphQL schemas) — for integration testing
• Design references (Figma URLs or screenshots) — for visual regression
• Browser/viewport matrix — which browsers and screen sizes to cover
• User journeys — critical paths to validate (login → checkout → confirmation)
• Credentials / test data — how to authenticate and what data to use
• Prior bug reports — context for the test oracle to avoid known issues

Spawning Agents

Read the subagent definition from the agents/ directory (e.g., agents/<agent-name>.md relative to the QA core root) before spawning each subagent. Pass the subagent definition as instructions along with the relevant inputs.

Parallel spawning is preferred — agents are designed to run independently. Spawn all relevant agents in the same turn to maximize throughput.

Example spawn instruction for a subagent:

You are the qa-e2e-exploratory-agent. Read and follow the definition in agents/qa-e2e-exploratory-agent.md.

Target: https://staging.example.com
Credentials: test@example.com / TestPass123
Focus areas: checkout flow, user settings, search
Save outputs to: <workspace>/exploration-results/

Use the skills in skills/qa-exploring-application-ui/ and skills/qa-generating-bug-reports/.

Skills Reference

Each skill has its own SKILL.md with detailed instructions. Read the relevant skill before executing its capabilities:

Output Structure

All agents write results to a shared workspace:

<workspace>/
├── exploration-results/      ← qa-e2e-exploratory-agent
├── e2e-results/              ← qa-e2e-deterministic-runner
├── integration-results/      ← qa-api-contract-fuzzer
├── service-results/          ← qa-multi-service-transaction-validator
├── visual-results/           ← qa-semantic-visual-regression
├── browser-results/          ← qa-cross-browser-explorer
├── oracle-baselines/         ← qa-test-oracle-synthesizer
├── benchmarks/               ← qa-benchmark-evaluator
└── summary-report.md         ← orchestrator (you generate this)

After all subagents complete, synthesize their findings into summary-report.md using the report generation script located in this skill's scripts/ directory:

# Replace <path-to-qa-exploring-tester> with the actual path to this skill directory
python <path-to-qa-exploring-tester>/scripts/generate_summary.py <workspace>/

Implementation Priority

If building incrementally, follow this order:

Tier 1 — Foundation (high impact, fewer dependencies):

1. qa-semantic-visual-regression — immediate value for flaky visual test replacement
2. qa-api-contract-fuzzer — leverages existing OpenAPI specs for security + robustness
3. qa-e2e-deterministic-runner — CI/CD regression gate

Tier 2 — Intelligence Layer (requires Tier 1 baselines): 4. qa-e2e-exploratory-agent — needs VLM integration (Claude Vision) 5. qa-test-oracle-synthesizer — uses Tier 1 outputs as validation baselines 6. qa-multi-service-transaction-validator — requires service mesh/contract definitions 7. qa-cross-browser-explorer — extends visual regression across browser matrix 8. qa-benchmark-evaluator — measures everything else's effectiveness