The Problem
Teams building AI‑driven coding assistants need a reproducible reference for how Claude Code structures its system prompts, how it coordinates sub‑agents, and how it enforces security classifications. Without a documented prompt catalog, reverse‑engineering efforts become ad‑hoc and error‑prone.
What This Does
The repository is a curated library of 30 markdown files that expose the exact prompt fragments Claude Code uses at runtime. The core prompt lives in prompts/01mainsystemprompt.md and is assembled from modular sections (e.g., prompts/04cyberriskinstruction.md, prompts/13toolprompts.md). Coordinator logic is described in prompts/05coordinatorsystemprompt.md, while specialized agents such as verification, exploration, and agent‑creation are defined in files 07‑10. Security handling is captured in prompts/11permissionexplainer.md and the two‑stage auto‑mode classifier in prompts/12yoloautomodeclassifier.md. The single top‑level README.md indexes the catalog and explains the high‑level architecture.
How To Use It
Clone the repo – no build step is required because the assets are plain markdown. git clone https://github.com/yourorg/claude-code-system-prompts.git cd claude-code-system-prompts Select the prompt set you need. For a full Claude Code session, concatenate the core prompt with any optional modules, for example: from pathlib import Path
core = Path("prompts/01mainsystemprompt.md").readtext() security = Path("prompts/04cyberriskinstruction.md").readtext() tools = Path("prompts/13toolprompts.md").readtext()
systemprompt = "\n".join([core, security, tools]) print(systemprompt)
Adjust the list to include coordinator, agent, or utility prompts as required by your workflow. Inject the assembled prompt into your Claude Code API call (or any compatible LLM endpoint). The repository does not provide a client wrapper, so you must integrate the string into your existing request code.
No additional configuration files, environment variables, or runtime dependencies are defined in the repo.
Real‑World Use
A development platform that auto‑generates code from natural language can import this catalog to keep its Claude Code prompt consistent across releases. For example, a CI step could run a script that rebuilds the system prompt from the markdown files whenever a new version of the prompt catalog is merged, guaranteeing that the LLM always receives the latest security and orchestration instructions.
pseudo‑code for CI integration def buildprompt(): parts = [ "prompts/01mainsystemprompt.md", "prompts/05coordinatorsystemprompt.md", "prompts/11permissionexplainer.md", ] return "\n".join(Path(p).readtext() for p in parts)
prompt = buildprompt() llmclient.send(systemprompt=prompt, usermessage=msg)
Code Health & Issues
Medium – No test suite – repository contains only markdown; there are no automated checks that the prompt fragments remain syntactically valid. Medium – No CI/CD – no .github/workflows or other pipeline files; changes are not automatically linted or validated. Medium – Missing LICENSE – usage and redistribution rights are undefined, which may limit commercial adoption. Low – Documentation completeness – README.md provides a good index, but individual prompt files lack explicit versioning or change history.
The Bottom Line
The repo offers a rare, well‑organized snapshot of Claude Code’s internal prompting strategy, valuable for research, audit, or building compatible agents. Its utility is limited to documentation; there is no executable code, test coverage, or licensing clarity, so organizations should treat it as a reference artifact rather than a drop‑in library. Suitable for teams that already have an LLM integration pipeline and need authoritative prompt fragments.