Microsoft's open-source unit-test agent for any language

Microsoft's code-testing-generator achieved a 92.1% task completion rate versus 78.9% for plain GitHub Copilot, with 63% fewer failures overall, when benchmarked against 152 internal tasks. The biggest gains appeared on vague or underspecified prompts, where the agent's repository-context awareness — reading the language, test framework, conventions, and build commands before generating tests — provides the most advantage. Teams evaluating AI test generation tools track head-to-head benchmarks like these on daily.dev.

The agent supports .NET, Python, Go, TypeScript, Java, Rust, and additional languages. It is explicitly scoped to unit tests only — integration tests, end-to-end tests, browser tests, and performance tests are out of scope. It works with GitHub Copilot CLI and VS Code and is available in the dotnet/skills repository on GitHub. Developers adopting AI-assisted testing across polyglot codebases follow tool coverage updates on daily.dev.

The RPI pipeline first reads the repository to understand the language, test framework, conventions, and build commands, then selects one of three execution paths: direct (straightforward tasks), single-pass (moderate complexity), or iterative (broad or vague requests). Tests are written starting from simple code toward more complex dependencies, then validated by checking for weak assertions and missing scenarios before running the full test suite. Engineers building or evaluating AI coding agent architectures find pipeline design discussions like this on daily.dev.