Claude Code source code leak, some say "shit mountain," others say "really good."

After looking at 1902 files, objectively speaking:

It’s not a shit mountain, but "a prototype that smart people don’t have time to clean up."

The good side:

• The architecture is clearly layered, coordinator/agent/memdir each perform their duties

• Multiple Agents use tmux + email files, simple but effective

• Memory extraction uses independent sub-Agents for isolation, the idea is correct

The rough side:

• USER_TYPE === 'ant' check scattered in 154 places not cleaned up

• Feature flags used as configuration, historical flags accumulated

• The same problem occurred once in 2025 and again in 2026

───

512K lines of TypeScript can achieve this level, the team's engineering capability is not an issue.

The problem lies in the rhythm: the feature iterations are too fast, and there was no "cleanup" process after release.

───

For those who write code: it’s worth studying the source code, but it’s not a learning benchmark—it demonstrates "how to quickly create good things," not "how to maintain good things."

Anthropic may need its own Claude Code to help with its code review.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。