Code Review agent: You can now initiate a code review for a specific scope by using the /code-review command in an agent mode session.
Browser agent: Use the /browser command in agent mode to access web pages and read online documentation.
Hook mechanism: The Agent execution flow now includes a hook mechanism that lets you insert custom logic at key stages of an agent's execution. It supports events such as UserPromptSubmit, PreToolUse, PostToolUse, PostToolUseFailure, and Stop.
Development container support: Define a standardized development environment with a .devcontainer/devcontainer.json configuration file to ensure a consistent experience for your team.
Windows terminal sandbox for Agent mode: Agent commands now run in a secure sandbox on Windows, effectively isolating potentially risky operations.
Extension Host process isolation: The built-in AI coding extensions now run in an isolated Extension Host process, improving runtime stability and the overall user experience.
