When using Claude Code, we can only see its surface functions, but it is difficult to understand the working mechanism behind it, especially its carefully designed system prompts. Coincidentally, I discovered the claude-code-system-prompts project on GitHub, which makes all the system prompts for Claude Code public. More than 40 prompt word files have been compiled, including prompt words for the main system, prompt words for sub-agents, tool descriptions, slash commands, etc., and also include token count statistics for each prompt word. GitHub: https://t.co/LsxwU3SoHv In addition to the prompt words themselves, a changelog spanning 52 versions is provided, showing the evolution of Claude Code from v2.0.14 to v2.0.71. If you want to gain a deeper understanding of how Claude Code works, or if you want to customize it based on the official hints, this project is worth saving and studying.
