Piebald-AI/claude-code-system-prompts

What it solves

This project provides a comprehensive, up-to-date collection of the system prompts used by Claude Code. Because Claude Code uses hundreds of different prompt strings—including conditional prompts, tool descriptions, and specialized agent prompts—rather than a single system prompt, they are difficult to track and analyze. This repository extracts these strings from the compiled source code to make them transparent and accessible.

How it works

The maintainers use a script to extract system prompts directly from the latest npm version of Claude Code's minified JavaScript files. This ensures the extracted prompts are identical to those used in the actual software. The repository is updated rapidly following each new Claude Code release, and a detailed changelog tracks modifications across hundreds of versions.

Who it’s for

• AI Researchers and Prompt Engineers: Those looking to study how a complex agentic coding tool structures its instructions.
• Claude Code Users: People who want to understand the internal logic governing the tool's behavior.
• Customizers: Users who want to modify these prompts in their own installation (often using the companion tool tweakcc).

Highlights

• Extensive Coverage: Contains over 500 extracted prompt strings, including those for sub-agents like Explore and Plan.
• Detailed Breakdown: Includes token counts for each prompt to help users understand the context window impact.
• Version Tracking: Maintains a CHANGELOG.md covering 274 versions since v2.0.14.
• Specialized Prompts: Includes prompts for specific slash commands (e.g., /code-review, /security-review, /simplify) and utility functions like session title generation.