Blogs ⚡ Gagan Goswami

Claude Code Explore

What actually happens when you type a message into Claude Code? The agent loop, 50+ tools, multi-agent orchestration, and unreleased features — mapped straight from the source.

541+ FILES
148K+ LINES OF CODE
50+ TOOLS
60+ COMMANDS
START EXPLORING

The Agent Loop

From keypress to rendered response — every step mapped straight from the source.

1 / 11Input
1⌨

User Input

Input
src/main.tsx

User types a message into the REPL or pipes content via stdin. Commander.js parses all CLI flags. Session context prefetches run in parallel — git status, file context, MCP connections — so the first response is fast.

● ccexplore $
> What does the agent loop look like?█

Architecture Explorer

148K+ lines of TypeScript, visualised. Click blocks to explore what lives where.

tools & commands
core
ui
infra
support
personality

Click any block to explore

Tool System

Every tool Claude Code can invoke — 43 functions across 8 categories, with source.

FILE OPERATIONS6
EXECUTION3
SEARCH & FETCH4
AGENTS & TASKS11
PLANNING5
MCP4
SYSTEM11
EXPERIMENTAL8

Command Catalog

60+ slash commands sorted by workflow — LocalCommand, PromptCommand, and LocalJSXCommand.

SETUP & CONFIG12
DAILY WORKFLOW24
CODE REVIEW & GIT13
DEBUGGING & DIAGNOSTICS11
ADVANCED & EXPERIMENTAL14

Hidden Features

Feature-flagged, env-gated, or commented out. The stuff Anthropic hasn't shipped yet.

Features may require specific flags, env variables, or internal access.