Understanding the Agentic Loop

The Core Loop

Claude Code operates on a loop often called ReAct (Reason + Act):

1. Observe: Look at user input and current environment (files, errors, git status)
2. Think: Formulate a plan - "To fix this bug, I need to read app.py first"
3. Act: Call a tool like read_file('app.py')
4. Loop: Receive tool output and go back to step 1

Example Flow

This loop continues until the task is complete or Claude needs more input from you.