Most modern AI assistants’ behavior is defined less by model weights and more by the system prompts and tool definitions that steer them. This repository centralizes those textual artifacts across dozens of popular AI tools so you can inspect how real-world agents are prompted to act, call tools, and resolve ambiguous user intents.
What Sets It Apart
- Broad coverage: A large, file-first collection organized by tool (Cursor, Devin, Replit, v0, Perplexity, etc.), making side-by-side comparison of system prompts and tool JSON easy — useful when you want to compare e.g. browser/tool usage policies or file/terminal handling patterns.
- Community-driven and regularly updated: Contributors submit prompt variants, tool schemas, and small model configs rather than a single vendor's perspective, which reveals common patterns and anti-patterns across implementations.
- Research-friendly format: Everything is plain text/JSON/YAML; you can grep, diff, and run lightweight analyses on prompt patterns, tool interfaces, or guardrails without installing heavy dependencies.
Who It's For — Trade-offs and When to Use It
Great fit if you are a prompt engineer, researcher, or engineer designing an agent and need concrete, real-world examples of system prompts and tool definitions to model behavior or test safety heuristics. It’s also handy for onboarding engineers who want to learn common tool-calling conventions. Look elsewhere if you need production-ready libraries, maintained SDKs, or guaranteed legal clearance for proprietary content — this repo is an aggregation of found/curated prompt files and may carry licensing or compliance uncertainties. Expect noisy contributions, occasional duplicates, and the need to validate any snippet before production use.
