Reverse-engineered internal system prompts from every Claude Code version.

What it is

Reverse-engineered collection of every internal Claude Code system prompt — 110+ prompts across 142 versions, with changelog and diffs. You see exactly how Claude Code works under the hood and can tune your CLAUDE.md around it.

Install / Setup

No install — it's a reference repo to read through.

Link: github.com/Piebald-AI/claude-code-system-prompts

Prompt version — when you want to learn something specific:

Look at https://github.com/Piebald-AI/claude-code-system-prompts and pull the 5 most important things I should know from the current system prompts to write a better CLAUDE.md. Specifically: which patterns/rules does Anthropic use that I should steal?

What you'll find

• Every system prompt from every Claude Code version
• Changelog with diffs between versions
• Tool definitions and internal logic

Use cases

• Tune your CLAUDE.md — See which rule patterns Anthropic uses internally and copy the approach.
• Debug why Claude did that — Behavior looks weird? Read it in the system prompt.
• Learn prompt engineering — Anthropic's own prompts are a free textbook.
• Track version diffs — New Claude Code version, behavior changed? The diff tells you why.

Pro tip

Don't just copy-paste. The internal prompts are built for the agent loop — your CLAUDE.md is additive on top. Pick individual patterns (how they describe tool usage, how they phrase a "refuse") and steal the style, not the content.

🚀 Installing skills is step one. Knowing which ones to use when — that's the real edge. In the community I show you my full skill setup and how I use it every day. → Join the Claude Code Mastery Community

Updated regularly — follow @vine.codes for more.