Write practical Agent configurations for the multi-agent app. Use when the user asks for Agent names, descriptions, system prompts, tool tags, output formats, or model choices.
Diagnose failed canvas nodes in the multi-agent app. Use when a node, tool call, model call, file read/write, Python service, or workflow run fails and the user wants a practical explanation and next step.
Advise model/provider choices for the multi-agent app. Use when the user asks which LLM to use for long documents, images, reasoning, low cost, speed, or fallback routing.
Generate and review custom Python tools for the multi-agent app. Use when built-in tools do not cover a task and the user wants a new tool, or when failure diagnosis suggests a missing tool/library.
Plan canvas workflows for the multi-agent desktop app. Use when the user wants to design, explain, or improve a canvas made of Agent nodes, tool nodes, gate nodes, and data flow connections.