claude-code-system-prompts
All parts of Claude Code's system prompt, 27 builtin tool descriptions, sub agent prompts (Plan/Explore/Task), utility prompts (CLAUDE.md, compact, statusline, magic docs, WebFetch, Bash cmd, security review, agent creation). Updated for each Claude Code version.
A reverse-engineered collection of the internal instructions that control Claude Code's behavior, over 110 extracted system prompt strings organized by category, with a changelog tracking changes across 177+ versions since 2025.
This repository publishes the internal instructions that Claude Code, Anthropic's command-line AI coding tool, uses to direct its own behavior. These instructions, called system prompts, are normally invisible to users. They are baked into the Claude Code software and tell the AI how to respond, what tools it has available, and how to behave in specific situations. The team behind a competing product called Piebald extracted these prompts directly from Claude Code's published software package and keeps this repository updated with each new release.
Claude Code does not use a single system prompt. It uses over 110 separate strings that get combined depending on the environment, the task, and the tools in use. The repository organizes these into categories: prompts for built-in sub-agents (like the Explore and Plan agents), tool descriptions (which explain to the AI what each tool does and how to use it), utility prompts for tasks like generating session titles or compacting conversation history, and system reminders that get injected at various points during a session.
A changelog file tracks how these prompts have changed across more than 177 versions of Claude Code since 2025, allowing developers and researchers to follow how Anthropic has updated the AI's instructions over time. The repository is updated within minutes of each new Claude Code release.
A companion tool called tweakcc, also by the same team, lets users modify specific parts of these prompts in their own Claude Code installation. It patches the same strings that are documented here, and handles conflicts when both the user and Anthropic have made changes to the same prompt.
This repository is primarily useful for developers who want to understand how Claude Code works internally, researchers studying AI system design, or practitioners who want to customize their Claude Code setup. The full README is longer than what was shown.
Where it fits
- Study exactly how Anthropic instructs Claude Code to behave, including what tools it has and the rules it follows when deciding what to do.
- Track how Claude Code's behavior has changed across software versions by reading the structured prompt changelog.
- Customize specific prompts in your own Claude Code installation using the companion tweakcc tool documented alongside this repo.
- Research AI agent system design by examining how sub-agents like Explore and Plan are instructed separately from the main agent.