VS Code 1.118 adds chat evals, remote control

Microsoft is no longer treating AI features as bolt-ons; this release turns VS Code into a control plane for supervising agents across devices and across codebases. The most interesting change is not the flashy UX, but the move toward making prompts, skills, and agent behavior observable and lintable.

• –Remote control for Copilot CLI removes a real workflow bottleneck: long-running sessions no longer die just because you stepped away from your desk
• –Dedicated skill context is a practical quality win; isolating tool-heavy tasks should reduce context pollution and make chat responses more focused
• –The new evaluations extension is a strong signal that prompt engineering is becoming an engineering discipline, not just a prompt box
• –Semantic workspace search plus GitHub org search makes Copilot more useful in multi-repo environments where the right answer is often outside the current project
• –Enterprise controls and token-efficiency improvements show Microsoft is balancing agent capability with governance and cost, which matters for teams trying to deploy this at scale