The Problem

This repository functions as a curated catalog of system prompts extracted from commercial AI coding tools. Teams or individuals seeking to understand, reverse-engineer, or benchmark the underlying instructions governing AI assistant behavior will find organized collections here. There is no software product, runtime, or deliverable produced by this repo - it is purely a reference collection.

What This Does

The repository groups system prompts and companion tool definitions by vendor (e.g., Anthropic/, Cursor/, Windsurf/, VSCode Agent/). Each vendor directory typically contains a Prompt.txt file (the core system instruction) and a Tools.json file (defining available functions or capabilities). File types are homogeneous within categories: prompts are .txt or .yaml, tool definitions are .json. Notable examples include Anthropic/Claude Code/Prompt.txt, Cursor Prompts/Agent Prompt 2.0.txt, and VSCode Agent/gpt-5.txt. The collection spans 30+ vendors and includes both closed‑source tool prompts and genuinely open‑sourced equivalents under Open Source prompts/. No source code that executes; the artifacts are static instruction files.

How It Is Wired

There is no execution graph, entry point, or runtime wiring in this repository. The contents are entirely static documentation—prompt text and JSON schemas intended to be consumed by external systems (LLM clients, agent frameworks, integrated development environments). The repository itself is a flat collection of files organized by vendor name; no module, function, or script ties them together. Any "call graph" would be defined by the consumer of these prompts, not by code within this repo. The only cross‑file dependency is the informal grouping under vendor directories, which the README does not formalize beyond listing.

How To Use It

There are no build, install, or runtime commands associated with this collection. The files are ready to be copied or referenced by whatever system consumes system prompts. If you are integrating a prompt into a framework, the typical pattern is to point the framework's configuration at the Prompt.txt and Tools.json for the desired vendor. No environment variables, keys, or secret material are required to read the files— the repo contains no committed secrets. Example layout:

# To reference the Cursor Agent Prompt 2.0:
Cursor Prompts/Agent Prompt 2.0.txt

# To reference the Claude Code tools definition:
Anthropic/Claude Code/Tools.json

No package.json, pyproject.toml, Dockerfile, or Makefile exists in the repo, so there is no formal setup process, dependency resolution, or container build pipeline.

Real-World Use

A practitioner benchmarking LLMs against vendor‑specific instructions can pull the relevant Prompt.txt to establish a baseline prompt for their model. For example, comparing model behavior under Anthropic/Claude Code/Prompt.txt versus Cursor Prompts/Agent Prompt 2.0.txt isolates differences in how each tool frames reasoning, tool use, and code generation expectations. The companion Tools.json files expose which functions each system declares as available, which can inform tool‑calling schema design or evaluation suites. This repo is useful as a reference dataset, not as a functional component.

Code Health & Issues

The static analysis produced the following measured findings:

  • tests present: yes - 1 test file exists
  • CI: no - no .github/ workflow or CI configuration detected
  • Dockerfile: no - no container build definition
  • lockfile: no - no package-lock.json, yarn.lock, or similar
  • licence: yes - LICENSE.md is present
  • committed secrets: none found

Beyond the measured block, the structure itself reveals the repo's nature: it is a static prompt collection with no source code, no testable units, and no automation. The single test file and absent CI pipeline are consistent with a reference/curated repo rather than a deliverable project.

The Bottom Line

This is a well‑organized catalog of system prompts from major AI coding tools, useful as a reference or benchmarking dataset. It ships with a license, no secrets, and a single test file, but offers no runtime code, no CI/CD, and no installation pathway. Teams looking to study prompt patterns, compare vendor instructions, or build evaluation suites will find immediate value; anyone seeking a functional agent framework, deployable tooling, or automated quality gates will need to look elsewhere.