How to Build a Coding Agent (Repository)

Overview

This repository accompanies Geoffrey Huntley's workshop on building coding agents from scratch. The code demonstrates that functional coding assistants require nothing exotic - just Go, the Anthropic API, and five core tools composed in an event loop.

Key Features

• Incremental build structure - Six progressive files, each adding one capability
• Minimal dependencies - Uses only Go standard library and Anthropic SDK
• Complete tool implementations - Read, list, bash, edit, and ripgrep-based search
• Event loop architecture - The "agent heartbeat" pattern that powers all coding assistants

Progressive Build Structure

The workshop builds capability layer by layer:

flowchart LR
    A[chat.go
Basic LLM] --> B[read.go
+File Read]
    B --> C[list_files.go
+Directory List]
    C --> D[bash_tool.go
+Shell Exec]
    D --> E[edit_tool.go
+File Edit]
    E --> F[code_search.go
+Ripgrep Search]

Code Snippets

Installation

# Clone and set up
git clone https://github.com/ghuntley/how-to-build-a-coding-agent
cd how-to-build-a-coding-agent

# Set your Anthropic API key
export ANTHROPIC_API_KEY=your-key-here

Running Each Stage

# Start with basic chat
go run chat.go

# Progress through stages
go run read.go
go run list_files.go
go run bash_tool.go
go run edit_tool.go
go run code_search_tool.go

Technical Details

Tools function as plugins defined by three components: name, input schema (generated from Go structs), and execution function. The agent loop follows a consistent pattern:

1. Accept user input
2. Send to Claude
3. Check for tool calls
4. Execute requested tools
5. Return results to context
6. Repeat until complete

Connections

• how-to-build-a-coding-agent - The workshop article that explains the concepts this code implements
• building-effective-agents - Anthropic's guide reinforcing the same "simplicity first" philosophy demonstrated here