Piebald-AI/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.

What this repo is

Piebald‑AI/claude-code-system-prompts is a curated collection of the exact system‑prompt strings that power Claude Code, Anthropic’s code‑focused LLM assistant. Claude Code does not use a single prompt; its behaviour is driven by hundreds of separate prompt fragments for different built‑in tools, sub‑agents, slash‑commands and utilities. This repository extracts those fragments from the official Claude Code npm package (or binary) and stores them as individual markdown files together with the token count for each fragment.

The repo is kept up‑to‑date automatically – a new release is published within minutes of every Claude Code version. A changelog tracks how the prompts evolve across more than 300 releases.

Why it matters

  • Transparency – developers can see exactly what instructions are being fed to the model, which is useful for auditing, debugging, or understanding model behaviour.
  • Customization – the companion tool tweakcc lets you replace any of these prompt fragments with your own markdown, then patches your local Claude Code installation accordingly.
  • Reference – the list serves as a quick lookup for the many specialised agents (e.g., /code‑review, Explore, Plan, Security‑review, etc.) that Claude Code ships with.

What you’ll find in the repo

Category Example prompts Token count (approx.)
Sub‑agents Explore, Plan (enhanced) 862 – 1066
Creation assistants CLAUDE.md creation, Status line setup 631 – 4815
Slash‑command agents /batch, /code‑review (many modes), /simplify, /schedule, /security‑review 80 – 4815
Utility agents Conversation summarisation (short/lean/long), background‑job handling, memory consolidation, artifact comment handling, etc. 63 – 6237

Each prompt lives in system-prompts/ as a markdown file (e.g., agent-prompt-code-review-part-3-extra-high-and-maximum-effort-modes.md). The header of the file shows the raw prompt text and the token count shown in the README.

How to use it

  1. Browse – open any markdown file to read the raw system prompt that Claude Code will send to the model.
  2. Patch – if you want to change a fragment, install the companion repo tweakcc. It lets you edit a prompt locally and then patches your Claude Code installation (npm or binary) with the new text, handling conflicts when Anthropic updates the same fragment.
  3. Stay informed – star the repo; each new Claude Code version triggers a GitHub release, and you’ll receive a notification.

Who maintains it

The repository is published by the team behind Piebald, an “agentic AI developer experience” platform (see https://piebald.ai). The same team builds the tweakcc tool and provides updates via Discord and X.

Quick take‑away

If you work with Claude Code—whether you’re building extensions, auditing its behaviour, or simply want to understand the many specialised agents it ships with—this repo gives you a single source of truth for every system‑prompt string that drives the model.

Related

  • Project
    wesammustafa/Claude-Code-Everything-You-Need-to-Know

    A practical, up‑to‑date guide for Anthropic’s Claude Code CLI, covering installation, prompt engineering, custom slash‑commands (Skills), automatic Hooks, sub‑agents, JavaScript workflows, and the Model‑Context‑Protocol for external tool integration. Includes ready‑to‑copy `.claude/` examples for teams building AI‑augmented development pipelines.

  • Project
    YeonwooSung/my-claude-code-mods

    A set of Node‑based plugins (mods) for Anthropic’s Claude Code that add profiling, context analysis, timeline snapshots, loop detection, pinned notes, and model‑effort visualisation. Each mod is installable via Claude’s marketplace and includes unit tests and TypeScript type‑checking.

  • Project
    feiskyer/claude-code-settings

    A set of Claude Code plugins (skills & sub‑agents) that add brainstorming, deep research, GitHub issue/PR automation, image generation, translation, and more, with ready‑made config files for many LLM providers.

  • Project
    codeaashu/claude-code

    A leaked source archive of Anthropic's Claude Code CLI, a terminal-based AI coding assistant that can edit files, run commands, and manage git workflows.

  • Project
    michaelshimeles/skills

    A curated set of Claude Code agent skills that automate common development tasks—visual diff generation, service‑layer structuring, evidence‑driven UI testing, iterative Greptile review loops, work‑tree isolation, and prose clean‑up—plus a ready‑to‑drop workflow tying them together.