Wheelder Human Agent — Autonomous Browser Testing Platform

1. What Is Wheelder Human Agent?

Wheelder Human Agent is an autonomous browser-based testing agent that does what a human QA tester does — but automatically, across multiple browsers, viewports, and test scenarios simultaneously.

It opens a real browser (Chromium, Firefox, or WebKit), navigates to any web page, detects forms, fills them with intelligently generated test data, clicks buttons, validates responses, captures screenshots on failure, and generates comprehensive QA reports — all without any human intervention.

The agent supports:

• Smart Test Mode — AI-powered auto-detection that works on ANY page automatically
• Comprehensive Testing — Full UI/UX, security, validation & boundary testing
• Security Testing — SQL injection, XSS, CSRF & input sanitization payloads
• Custom YAML Tests — Define your own test steps declaratively
• Multi-Browser Matrix — Chrome, Firefox, Safari simultaneously
• Responsive Testing — Desktop, Tablet, Mobile viewports
• Intelligent Variant Generation — Automatically creates 30-100+ test cases from a single spec
• Evidence Capture — Screenshots, logs, and HTML reports for every test run

2. Live Screenshots

The following screenshots were captured from the live running application on localhost:8001:

Dashboard — Target URL & Test Types

The main interface allows selecting a target URL and choosing from 5 test types: Smart Test (AI-powered), Comprehensive, Quick, Security, and Custom YAML.

Dashboard — Browser Matrix, Viewports & Options

Multi-browser selection (Chrome, Firefox, Safari), responsive viewport testing (Desktop, Tablet, Mobile), and advanced options including visible mode, HTML reports, standard tests, custom YAML, and AI-powered variants.

Dashboard — YAML Configuration & Action Buttons

Custom YAML test configuration with preset templates (Signup, Login, Generic Form, Requirements) and the execution controls (Run QA Test, Stop Tests, Clear Cache).

3. Complete Feature Breakdown

3.1 — 20+ Action Types

The agent supports a comprehensive set of browser actions that mirror everything a human tester would do:

3.2 — Two-Phase Testing Strategy

3.3 — Session Reuse for Efficiency

When intelligence is enabled, the agent keeps the browser window open between variant tests, creating a fresh page context for each test while reusing the browser instance. This dramatically reduces startup time for comprehensive testing sessions.

3.4 — Template Variable System

Test specifications use {{test_data.email}} syntax for dynamic value injection. The intelligence layer generates variants by modifying these values, enabling a single test spec to produce dozens of different test scenarios.

3.5 — Web Dashboard

A FastAPI-powered web interface provides:

• Interactive test configuration form
• One-click test execution with live progress
• Built-in QA report card with copy-paste summaries
• Screenshot gallery integrated into reports
• Preset YAML templates for common test types (Signup, Login, Generic Form)

4. Architecture & Technology Stack

Core Technology

Module Architecture

src/qa_agent/
├── orchestrator/        # Central execution controller
│   └── orchestrator.py  # Two-phase testing, browser management
├── runner/
│   ├── step_executor.py # Executes individual test steps
│   └── results.py       # Result data structures
├── intelligence/
│   ├── baseline_generator.py  # Deterministic variant generation
│   ├── expander.py            # Combines baseline + AI variants
│   └── ollama_provider.py     # AI integration
├── spec/
│   ├── schema.py        # Test specs, steps, actions
│   ├── loader.py        # YAML/JSON parsing
│   └── validator.py     # Spec validation
├── viewport/            # Device preset configurations
├── evidence/            # Screenshot/log capture
├── reports/             # JSON & HTML report generation
├── browser_matrix/      # Cross-browser orchestration
├── web/                 # FastAPI dashboard
│   ├── app.py
│   └── templates/
└── cli.py               # CLI entry point

5. Intelligence Layer — Smart Test Generation

The intelligence layer automatically generates comprehensive test variants from minimal input. Given a simple test with an email and password, it produces:

Baseline Variants (30+ per field)

• Email: empty, missing @, double @, no TLD, whitespace, unicode domains, plus addressing, 64+ char local part
• Password: empty, 1 char, only lowercase, only numbers, leading/trailing spaces, Chinese characters, emoji, backslash, null bytes, common weak passwords (password, 123456)
• Generic Fields: XSS payloads, SQL injection, template injection, path traversal

Security Variants (45+ payloads)

• SQL Injection: ' OR 1=1--, '; DROP TABLE users;--, 1 UNION SELECT * FROM users
• XSS: <script>alert(1)</script>, <img onerror=alert(1)>, javascript:alert(1)
• Path Traversal: ../../etc/passwd, ..\\..\\windows\\system32
• Command Injection: ; ls -la, $(cat /etc/passwd)
• NoSQL Injection: {"$gt": ""}, {"$ne": null}

Deterministic Reproducibility

6. Git Commit Timeline — Verifiable Evidence

Every commit is publicly visible on GitHub with author dates set by Git at commit time:

How to Verify

git clone https://github.com/Wheelder/wheelder-human-agent.git
cd wheelder-human-agent
git log --format="%H %ai %s" --all

# Original source repository:
# https://github.com/abbaays/human-agent

GitHub Commit History — Public Screenshots

The following screenshots show the public commit history on GitHub, proving all original development commits are from December 15-26, 2025:

Live commit history: github.com/Wheelder/wheelder-human-agent/commits/main

7. Public Timeline Comparison — Agentic Systems

The following table presents Wheelder Human Agent's development dates alongside the public announcement dates of similar full-stack agentic systems by other companies. All dates are based on publicly available information.

8. Roadmap — Continuing Innovation

Wheelder Human Agent is an active, evolving project. The foundation built in December 2025 continues to expand:

• Full AI Integration: Deep integration with LLMs (Ollama, Groq) for truly intelligent page understanding — not just pattern matching but semantic comprehension of what's on screen
• Visual Regression Testing: Pixel-level comparison of screenshots across test runs to detect UI changes
• Natural Language Test Specs: Write tests in plain English ("go to the signup page, fill in the form, and verify the success message") and have the agent execute them
• Multi-Page Workflow Testing: Complex user journeys spanning multiple pages with state preservation
• API + Browser Hybrid Testing: Combine API calls with browser automation for end-to-end testing
• Self-Healing Selectors: When a CSS selector breaks due to a UI change, automatically find the correct element using AI
• Continuous Monitoring: Run tests on a schedule and alert on failures — autonomous QA as a service

This project continues to open new doors for innovation in the autonomous agent space.

9. Public Repository

Additional Evidence

• Innovation Timeline: wheelder.com/timeline
• Deep Research Evidence: wheelder.com/releases?id=3
• GitHub Organization: github.com/Wheelder