ClaudeCode Code Walkthrough: From Request to Tool Execution

This chapter gives you a practical map of how ClaudeCode handles a request, so you can read the source code with clear priorities.

1. Main Flow

Think of one request in four steps:

1. Parse user input and current context
2. Build the system prompt
3. Run model inference and decide whether to call tools
4. Persist session state and produce the final response

2. Core Modules to Read First

• query.ts: main request loop entry
• QueryEngine: session state and message orchestration
• Tool modules: tool definitions, permission checks, execution pipeline
• Hook modules: extension points to intercept lifecycle events

3. Recommended Reading Strategy

• Pass 1: focus on call graph and control flow
• Pass 2: focus on state transitions and data structures
• Pass 3: focus on edge cases (timeouts, permission denial, tool errors)

4. Three Questions to Keep in Mind

• Where does ClaudeCode decide between "continue dialogue" and "call tool"?
• How are tool failures returned to the model for the next reasoning step?
• How is context compacted to avoid token explosion?

Next step: read query.ts first, then tool orchestration and context compaction modules.