Microsoft released an open source AI agent that automates unit test creation and validation across .NET, Python, Go, TypeScript, Java, and Rust.
On 31 July 2026, Microsoft released an open source AI agent named code-testing-generator to automate unit test creation across languages like .NET, Python, Go, TypeScript, Java, and Rust. Hosted in the dotnet-test plugin within GitHub’s dotnet/skills repository, it works via the GitHub Copilot CLI and Visual Studio Code, via preview plugin support.
The agent strictly generates unit tests, isolating code and mocking external dependencies while explicitly excluding integration, E2E, browser, and performance tests. It operates via a Research-Plan-Implement (RPI) pipeline, scanning repositories for languages, frameworks, and build commands before selecting one of three execution paths based on scope:
-
Direct: Reads code, writes tests, and validates results immediately.
-
Single-Pass: Researches and plans once before executing.
-
Iterative: Continuously repeats the cycle to fulfill large requests or test coverage goals.
Before finalizing tests, the agent runs validation checks: evaluating lightweight code mutations, auditing assertions, ensuring scenario coverage, and confirming tests build and pass using native workspace test commands.
In internal evaluations across 152 tasks, Microsoft’s code-testing-generator achieved a 92.1% completion rate compared to 78.9% for stock GitHub Copilot, generating 63% fewer failures. The performance gap was widest on vague prompts (88.8% vs. 66.3% success), while both tools performed equally well (96.8%) when provided with detailed instructions.
