create-agent

Scaffold or develop an agent: $ARGUMENTS

Before Starting

1. Search OpenAI docs for the latest Agents SDK patterns:
   • Use mcp__openai-docs__search_openai_docs with query "Agents SDK agent definition" to confirm current Agent constructor and run() patterns
   • Use mcp__openai-docs__search_openai_docs with query "Agents SDK tools" for tool definition patterns
2. Read existing agent definitions to understand project conventions:
   • {{AGENTS_DIR}}/chat-coaches/index.ts — simple agent (tenant-driven prompt, single model)
   • Read 1-2 files in the ai-lab-agent repo at {{PROJECT_ROOT}}/{{AGENTS_DIR}}/ai-lab/index.ts for the function-based config pattern (instructions/tools/modelSettings as functions receiving RunContext)
   • {{PROJECT_ROOT}}/{{AGENTS_DIR}}/writing-assistant/index.ts — multi-mode agent with per-mode temperature/token configs
   • {{PROJECT_ROOT}}/{{AGENTS_DIR}}/seo-expert/index.ts — tool-heavy agent with ordered tool execution
3. Read the shared infrastructure:
   • {{AGENTS_DIR}}/shared/enhanced-agent-bridge.ts — bridge class that wraps agent execution
   • {{AGENTS_DIR}}/shared/types.ts — ChatKitRequest, ChatKitEvent, WritingContext types

Part 1: Agent Scaffolding