VS Code moves its agent to the Copilot harness
New default for agent sessions · Local harness removed Feb 2027
ACN Wirevia GitHub
Written with AI assistance, reviewed by the ACN desk.
GitHub said Copilot Evals is now in public preview, letting users create repeatable tests for how Copilot skills, custom agents and MCP tools behave in their own setup. According to GitHub, users can create and run evals with the /eval command or a natural-language request, see why a check failed and compare runs from before and after a change, approving each run before it starts. GitHub said the preview is available on Copilot Individual, Business and Enterprise plans, with the local eval tool Vally as its engine.
Spot an error? Email corrections@aicodingnews.si. Fixes are logged on the corrections page.
Sponsor
New default for agent sessions · Local harness removed Feb 2027
Adaptive effort · secret detection · public preview
For long-running agent work · in GitHub Copilot the same day
How the Claude Code team verifies agent work
Sponsor
Sponsor
Sponsor
Sponsor