The latest articles on DEV Community by Juan Bermudez Roldan (@juan_bermudezroldan_4f96). https://dev.to/juan_bermudezroldan_4f96 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3860700%2Ffc6284c2-a640-42e4-b349-19f7a1e90918.jpg https://dev.to/juan_bermudezroldan_4f96 en Juan Bermudez Roldan Sat, 04 Apr 2026 09:02:04 +0000 https://dev.to/juan_bermudezroldan_4f96/the-11-steps-that-run-every-time-you-press-enter-in-claude-code-ked https://dev.to/juan_bermudezroldan_4f96/the-11-steps-that-run-every-time-you-press-enter-in-claude-code-ked <p>Every time you press Enter in Claude Code, an 11-step loop kicks off. Not a simple request-response — a self-directed cycle that keeps running until the task is done.</p> <p>I spent a few weeks tracing through the source code and turned what I found into a free course. Here's the condensed version of that loop.</p> <h2> The 11 Steps </h2> <p><strong>Step 1 — Input.</strong> You type a prompt. The terminal captures the keypress via a custom React renderer (yes, React in a terminal).</p> <p><strong>Step 2 — Message.</strong> Your text becomes a UserMessage object appended to the conversation history.</p> <p><strong>Step 3 — Memory.</strong> CLAUDE.md files from 4 levels (enterprise → user → project → subdirectory) are loaded and concatenated into the system prompt.</p> <p><strong>Step 4 — System Prompt.</strong> Tool definitions, environment info, and memory are merged into the full system prompt.</p> <p><strong>Step 5 — API Call.</strong> Messages + system prompt + tool definitions stream to the Anthropic API. Tokens come back in real time.</p> <p><strong>Step 6 — Parse.</strong> The response is split into content blocks: text, thinking, and tool_use.</p> <p><strong>Step 7 — Decision.</strong> Does the response contain tool_use blocks? If yes → execute tools. If only text → done, return to user.</p> <p><strong>Step 8 — Execute.</strong> Each tool is found in the registry, validated against its JSON Schema, permission-checked (allow/ask/deny), then executed. Read-only tools run in parallel; write tools run one at a time.</p> <p><strong>Step 9 — Result.</strong> Tool outputs are wrapped as UserMessage objects (the API requires strict user/assistant alternation).</p> <p><strong>Step 10 — Loop.</strong> Updated history goes back to Claude. The loop continues until no more tool calls.</p> <p><strong>Step 11 — Render.</strong> Final text streams to the terminal. The screen buffer diffs and redraws only changed cells.</p> <h2> What surprised me </h2> <ul> <li>The agent loop itself is simple. The complexity lives in error recovery and retry logic (5 different retry strategies for different error types).</li> <li>There are 50+ tools, and read-only tools run concurrently while write tools get exclusive access.</li> <li>The source has ~8 unreleased features: a terminal companion pet (Buddy), persistent memory mode (Kairos), multi-agent teams, and more.</li> <li>Context compaction automatically summarizes old messages when the window fills up — there's a reactive safety net that catches prompt-too-long errors.</li> </ul> <h2> The full course </h2> <p>I built a free interactive course that goes deep on each of these systems:</p> <ul> <li>12 lessons covering all 8 architecture layers</li> <li>Animated video walkthroughs with narration</li> <li>Python code examples (not TypeScript — patterns you can reuse)</li> <li>Hands-on exercises and quizzes</li> <li>An interactive stepper that lets you click through all 11 steps</li> </ul> <p><strong><a href="https://code.bluggie.com" rel="noopener noreferrer">code.bluggie.com</a></strong></p> <p>Everything is free, no paywall. I'd love to hear what parts of Claude Code you'd want explained that I haven't covered yet.</p> <p><em>Not affiliated with Anthropic. Built by reading the public source code.</em></p> ai python architecture programming