Understand message flow in Claude Code

Link copied

Attach video time

Every message you send fires off an API call to Anthropic's large language models. Let's trace out exactly what happens when you send in a message.

Claude Code appends your message to the conversation history. This gives Claude the full context from all previous exchanges, so it always knows what you've been working on.

Claude Code injects its system instructions at the top of every request. These instructions define how Claude interacts with your terminal, which tools it can use, and how it should structure responses.

Right after the system prompt, your CLAUDE.md file gets included. This persistent context lives across all conversations, and helps Claude understand your project's architecture and conventions without you needing to repeat yourself.

Claude then evaluates which tools it needs for your request. Tools are specialized functions that let Claude read files, write code, execute bash commands, and interact with your development environment.

When Claude determines it needs a specific capability, it makes a tool call to access it.

This architecture is why Claude Code feels so responsive to your project's context. The combination of conversation history, persistent CLAUDE.md context and tool calls creates a development environment that can really adapt to any custom workflow.