Microsoft has open-sourced code-testing-generator, a polyglot agent that not only writes unit tests but also verifies they execute successfully. Available in the dotnet-test plugin within the MIT-licensed dotnet/skills repository, this agent addresses a common gap in AI-assisted development: generic prompts like “generate unit tests” leave framework choice, file location, and assertion style ambiguous. code-testing-generator resolves these uncertainties by analyzing the repository structure before generating any code. It then plans, writes, runs, and validates the tests it produces.
On Microsoft’s internal 152-task benchmark, the agent completed 140 tasks—compared to 120 for stock GitHub Copilot—using identical models and prompts. This translates to a 92.1% task completion rate versus 78.9%, highlighting a significant performance improvement in test generation.
Deployability
The agent is delivered as a definition alongside skills rather than a hosted service, ensuring it integrates directly into your existing coding agent while keeping all code local. This design prioritizes privacy and control, making it suitable for enterprise environments with strict data-handling requirements.
via MarkTechPost