Code2Test: Intent-Driven Test Generation Agent

Hierarchical Test Generation • Multi-Language Support • Human-in-the-Loop Verification

Code2Test

Code2Test is an advanced AI agent that automatically generates, verifies, and maintains test suites for large-scale codebases. Unlike simple test generators that look at isolated functions, Code2Test understands the intent of your components, their role in the system architecture, and their dependencies.

It employs a hierarchical approach, extracting high-level intents from your codebase, generating corresponding tests, and running them in a verification loop to ensure correctness.

🚀 Key Features

• Intent-First Generation: Extracts "intent" from docstrings, naming conventions, and dependencies to generate meaningful tests, not just coverage padding.
• Hierarchical Understanding: Analyzes your dependency graph to understand how components interact, enabling integration test suggestions.
• Human-in-the-Loop: Interactive CLI allows you to review inferred intents, edit generated tests, and guide the agent before committing.
• Self-Healing Verification: If a generated test fails, the agent analyzes the error, determines if it's a test bug or a code bug, and attempts to fix the test automatically.
• Multi-Language Support:
  • 🐍 Python (pytest) - Fully Supported
  • ☕ Java (JUnit 5) - Beta
  • 🟨 JavaScript/TypeScript (Jest) - Beta
• CI/CD Ready: Includes --auto mode for headless execution in CI pipelines, with support for Github Actions and GitLab CI.
• Comprehensive Reporting: Generates HTML and JSON reports detailing test coverage, verification status, and confidence levels.

📦 Installation

# Clone the repository
git clone https://github.com/cirkitly/code2test.git
cd code2test

# Install dependencies
pip install .

# Verify installation
code2test --help

🛠️ Usage

1. Interactive Test Generation

The default mode. The agent scans your code, presents inferred intents, and asks for confirmation before generating tests.

# Generate tests for the current directory
code2test test .

# Generate tests for a specific module
code2test test src/core/auth.py

2. CI/CD / Automated Mode

Run without user interaction. Useful for CI pipelines or nightly builds.

# Run in auto mode with a confidence threshold
code2test test --auto --confidence 0.8 --exit-code src/

• --auto: Skip interactive prompts.
• --confidence 0.8: Only generate tests for intents with >80% confidence.
• --exit-code: Fail the build (exit 1) if any verification step fails.

3. Reporting

Generate detailed reports after a test run.

code2test test --auto --report html .

This creates reports/report.html with a visual summary of the run.

🏗️ Architecture

Code2Test operates in five phases:

1. Foundation: Parses the codebase to build a Dependency Graph.
2. Intent Extraction: Uses static analysis + LLMs to infer what a component should do.
3. Test Generation: Uses language-specific agents to write idiomatic test code (Pytest, Jest, JUnit).
4. Verification: Executes the generated tests immediately.
5. Diagnosis & Repair: If tests fail, a diagnosis agent analyzes the traceback to fix the test or flag the code.

🤝 Contributing

Contributions are welcome! Please check out the implementation plan to see what's next.

📄 License

MIT License