The Core Pattern
Every coding agent — Claude Code included — is this one loop:
while True:
response = model(messages, tools)
if response.stop_reason != "tool_use":
return response.text
results = execute(response.tool_calls)
messages.append(results)The model calls tools until it is done. Memory, tasks, teams, MCP — every one of the 20 lessons is a refinement around this loop.
The Build, Stage by Stage
Key Concepts
Agency comes from the model
An agent product = model + harness. The intelligence is trained into the model; your job is the plumbing around it, not the reasoning.
One loop is the whole engine
A while-True loop that calls tools until the model stops. Every mechanism in the course hangs off this single loop.
Add capability, never rewrite the loop
Each lesson registers a new mechanism — tools, memory, tasks, teams — around the same loop instead of changing it.
Source + English + runnable code
Every chapter ships a narrative README, English/Japanese translations, and a standalone code.py you can actually run.
Related Projects
Learn Claude Code — full 20-lesson course
Browse the complete curriculum, stage by stage.
Everything Claude Code (ECC)
A searchable index of a production agent harness — agents, skills, commands, rules.
shareAI-lab/learn-claude-code
The source repository — narrative chapters, translations, and runnable code.
