claude-code-source-analysis (xiaonancs)

The core agent loop, as the community summarizes it:

while (true) {
  response = call_model(messages)
  tool_results = execute_tools_if_any(response)
  messages.push(response, tool_results)
  if (!needs_follow_up(response, tool_results)) break
}

The complexity is not in the loop but in the harness around it: context management, permission control, security defense, memory persistence, multi-agent orchestration. (1,884 files, 513,216 lines; 40 tools, 86 slash commands, 90 compile-time feature flags.)

On caching: under Anthropic's prompt cache, a cache hit costs roughly 1/200th of a miss at 200K tokens - System Prompt cache design directly moves the cost of millions of daily API calls.

| 1 | CLAUDE.md instruction files | permanent | project/user/enterprise dirs | static instructions |
| 2 | Auto Memory (memdir) | cross-session | ~/.claude/projects/<slug>/memory/ | AI-extracted persistent knowledge |
| 3 | Session Memory | single session | .../session-memory/summary.md | structured notes for the current session |
| 4 | Agent Memory | cross-session | three scope directories | per-agent-type experience |
| 5 | Relevant Memories | injected per turn | in-memory attachment | Sonnet side-query recall |

Implementation: src/memdir/ (8 files), claudemd.ts (1,480 lines), extractMemories/ (771 lines), SessionMemory/ (821 lines), autoDream/ (465 lines), agentMemory.ts (178 lines).