Custom Agents in VS Code (code.visualstudio.com)

🤖 AI Summary
VS Code (v1.106, preview) introduces "custom agents" — configurable AI personas for the editor that let you tailor behavior, available tools, and instructions to specific development roles (e.g., planner, implementer, security reviewer). Previously called custom chat modes, these agents make it easy to switch to role-specific configurations instead of manually toggling tools or prompts, reducing accidental edits (e.g., planning agents can be read-only) and standardizing responses across teams. A new handoff feature enables guided, sequential workflows: after one agent finishes, clickable buttons transition you to the next agent with context and a pre-filled prompt (optionally auto-submitted), which streamlines multi-step flows like Plan → Implement → Review. Technically, custom agents are defined as .agent.md Markdown files with YAML frontmatter (fields: name, description, tools, model, target, mcp-servers, handoffs, etc.) stored either in .github/agents in a workspace or in your user profile for reuse. Agent bodies are prepended to chat inputs and can reference tools using #tool:<tool-name> and other files via Markdown links. Tool availability follows a priority order (prompt file tools → referenced agent tools → agent defaults), and missing tools are ignored. You can create/manage agents via the Agents dropdown or the Chat: New Custom Agent command; legacy .chatmode.md files are still recognized. Overall, this adds reproducible, safer, and more orchestrated AI workflows inside VS Code.
Loading comments...
loading comments...