LLM Wiki

Appearance

Built 26/04/16 13:13commit cc1e88d

Claude Code System Prompts Repository ​

中文 | English

Summary ​

This source captures Piebald-AI/claude-code-system-prompts as a continuously updated extraction vault for Claude Code's internal prompt surfaces. Its main value is not just the prompt text itself, but the fact that it turns Claude Code's hidden compiled prompt inventory into a versioned, inspectable operating surface with token counts, version diffs, and prompt-family boundaries.

Source ​

Key Contributions ​

  • Makes explicit that Claude Code's effective system behavior is assembled from many prompt fragments rather than one monolithic hidden string.
  • Breaks the product surface into inspectable prompt families: builtin tool descriptions, subagent prompts, slash-command prompts, memory flows, security monitors, utility prompts, and reminders.
  • Tracks token counts and prompt churn across Claude Code releases, which turns prompt changes into an observable operational signal instead of opaque vendor drift.
  • Shows that reverse-extracted prompt inventories can become a durable operator reference, especially when official docs only describe features at a higher abstraction layer.
  • Connects prompt inspection with prompt patching through the adjacent tweakcc tool, suggesting a broader workflow around auditing and selectively modifying local Claude Code behavior.

Practical Implications ​

  • Claude Code should be understood as a layered control plane whose behavior is distributed across many internal prompt assets, not as a single hidden system prompt.
  • Versioned prompt extraction repositories are becoming part of the agent-operations ecosystem because they let teams audit behavioral drift between releases.
  • Builtin tools, slash commands, memory routines, and safety monitors all deserve treatment as first-class runtime surfaces rather than invisible implementation details.
  • Repo-native catalogs of hidden prompt assets strengthen the broader agent-first pattern: make operationally important internals legible, inspectable, and diffable.
  • This kind of prompt inventory becomes more useful when paired with sources that explain the assembly logic, conditional inclusion rules, and runtime context layers around those fragments.