Building Coding Agents

Build your own Claude Code-like AI coding agents from scratch.

01 — CORE ARCHITECTURE

Coding Agent Architecture

Focus: Core concepts and agent initialization

Agent class structure

Model provider setup (Bedrock, OpenRouter, Anthropic, Google)

Agent context and state management

Model mapping and max tokens configuration

02 — TOOL CREATION

Building Your First Tool

Focus: Tool creation from scratch

Tool structure using @openai/agents SDK

Zod schema for parameter validation

Tool execution with RunContext<AgentContext>

Error handling patterns

Event streaming with AgentMux

Progress reporting to frontend

03 — BEHAVIOR CONTROL

System Prompts & Agent Behavior

Focus: Defining agent capabilities and behavior

System prompt engineering

Tool descriptions and usage guidelines

Agent modes: 'agent' vs 'chat' mode

Context injection (todos, file structure, design system)

Conditional tool availability based on project state

04 — OPTIMIZATION

Advanced Features

Focus: Multi-agent systems and optimization

Model routing logic

State management (todos, file locks, caching)

Credit tracking and usage optimization

File operation sequencing with FileLockManager

Multi-model orchestration (routing between Sonnet/Gemini/etc.)