Microsoft Open Sources code-testing-generator: A Polyglot Unit-Test Agent Achieving 92.1% Task Completion vs 78.9% for Stock Copilot

Microsoft has open sourced code-testing-generator, a polyglot unit-test agent that writes tests and proves they work. It ships as a dotnet-test plugin in the MIT-licensed dotnet/skills repository. The agent addresses the gap where coding assistants don't specify test frameworks, file locations, or assertions. It first reads the repository, then plans, writes, runs, and checks tests. On Microsoft's internal 152-task benchmark, it completed 140 tasks (92.1%) versus 120 (78.9%) for stock GitHub Copilot, using the same model and prompts, which is 63% fewer failures. The gains are concentrated on vague prompts (79/89 resolved vs 59/89) and diff-targeted tasks (15/15 vs 0/15). It generated 2.3% fewer tests (6,963 vs 7,129) at nearly identical line coverage (72.4% vs 72.2%), and was 5.5% faster on average task time. The agent coordinates work through a Research-Plan-Implement pipeline, detects language and test framework, finds build and test commands, and never modifies production code. It runs five verification checks before reporting completion, including a lightweight mutation testing approach. The agent is an agent definition with skills, not a hosted service, so it runs inside existing coding agents and keeps code local.