E2E Agent

Create AI-powered end-to-end tests that plan, execute, and document themselves.

Write test specs in plain English. The AI agent plans comprehensive test cases, executes them in a real browser, captures evidence (screenshots, logs), and generates professional test reports—all autonomously.

Why E2E Agent?

• Natural Language Testing: Describe what to test, not how to test it
• Autonomous Execution: Agent handles test planning, execution, and reporting
• Real Browser Testing: Uses Chrome DevTools for authentic end-to-end validation
• Professional Reports: Industry-standard test cases and defect documentation
• Long-Running Capability: Stateless sessions with automatic progress checkpoints

Read the Anthropic blog post on building effective agent harnesses.

Installation

curl -fsSL https://raw.githubusercontent.com/tyyzqmf/e2e-agent/main/install.sh | bash

Or build from source.

Prerequisites:

• Node.js & Chrome/Chromium browser
• AWS Bedrock access OR Anthropic API key

Configure API credentials

AWS Bedrock (Recommended):

export CLAUDE_CODE_USE_BEDROCK=1
export AWS_REGION=us-west-2
aws configure

Anthropic API:

export ANTHROPIC_API_KEY='your-api-key-here'

See full configuration guide.

Quick Start

# 1. Check dependencies
e2e check

# 2. Start services (Web UI + Executor)
e2e start

Open http://localhost:5000 to:

1. Write test specs in plain English
2. Submit test job
3. Monitor autonomous execution
4. Download test reports

Or use the CLI:

e2e job submit quick-start.md    # Submit test job
e2e job status <job-id>          # Check progress
e2e stop                         # Stop services

See all CLI commands.

How It Works

Two-Agent Pattern

Test Spec (Plain English)
         ↓
┌────────────────────────┐
│  1. Test Planner       │  Generates test_cases.json
│     (Session 1)        │  with detailed test plans
└────────────────────────┘
         ↓
┌────────────────────────┐
│  2. Test Executor      │  Executes tests in Chrome
│     (Sessions 2+)      │  Captures evidence
│                        │  Generates reports
└────────────────────────┘

Stateless Sessions: Each agent run has a fresh context. Progress persists through test_cases.json (test status) and claude-progress.txt (session notes).

Autonomous Execution: The executor agent runs continuously, auto-continuing until all tests complete or a blocking issue occurs.

See architecture details.

What You Get

After test execution:

• Interactive HTML Report with test results and evidence
• Test Case Reports (TC-*.md) with step-by-step execution
• Defect Reports (DEFECT-*.md) with severity, reproduction steps, and evidence
• Screenshots at every major step
• API & Console Logs for debugging failures

Example reports →

Documentation

• Installation Guide: Detailed installation instructions
• Configuration Guide: API credentials and environment setup
• CLI Reference: Complete command reference
• Architecture: System design and implementation details
• Test Reports: Understanding test reports and evidence
• Troubleshooting: Common issues and solutions
• CLAUDE.md: Development guide for contributors
• Anthropic Blog: Design patterns for long-running agents

Tech Stack

• Runtime: Bun (TypeScript execution and web server)
• AI Agent: Claude Agent SDK
• Browser Automation: Chrome DevTools MCP
• Job Queue: SQLite

Contributing

Contributions welcome! This is a demonstration framework for building autonomous testing agents.

License

MIT - see LICENSE.md