AI Coding Prompt Library

The AI Coding Prompt Library (awesome-ai-system-prompts) is an open-source GitHub repository maintained by dontriskit that aggregates the actual system prompts powering many of today's leading AI coding assistants and chat tools. Rather than a generic prompt collection, the repository specifically curates the leaked, reverse-engineered, or officially published system prompts of products like ChatGPT, Claude, Claude Code, Perplexity, Manus, Lovable, v0 by Vercel, Grok, Same.new, Windsurf, Notion AI, and MetaAI, giving developers and prompt engineers a behind-the-scenes look at how production-grade AI agents are instructed.

The primary audience is AI agent builders, prompt engineers, indie hackers, and software developers who want to understand the structural patterns that make commercial AI assistants behave reliably. By studying these system prompts side by side, users can learn how top products handle tool-calling instructions, refusal policies, code formatting conventions, multi-step reasoning chains, error recovery, and persona definition. This kind of comparative reference material is otherwise difficult to assemble because each vendor publishes its prompts in different places, at different times, or not at all.

The repository is organized as a flat collection of folders, each named after the product whose prompt it contains, with the raw prompt text stored as plain markdown or text files. This makes it trivial to clone the repo, grep through it, or feed selected prompts into your own LLM workflow as scaffolding. Because the content is plain text under an open-source license, contributors can submit pull requests when new prompts leak or are officially shared, keeping the collection reasonably current with the fast-moving AI tooling ecosystem.

For coding-focused use, the most valuable entries are the system prompts from Claude Code, Cursor, Windsurf, v0, Lovable, and Same.new — tools that have built sophisticated software-engineering agent loops. Reading these prompts reveals the exact phrasing used to enforce constraints like 'never modify files you haven't read,' how tool schemas are described to the model, how the agent is told to plan vs. execute, and how output formatting (diffs, code fences, file paths) is standardized. Engineers building their own coding agents on top of the Anthropic, OpenAI, or open-source model APIs can adapt these patterns rather than reinventing them from scratch.

The project is entirely free and community-maintained. There is no SaaS layer, no account, no API, and no telemetry — it is simply a GitHub repository you can star, fork, or clone. The trade-off is that there is no guarantee of accuracy, completeness, or freshness for any specific prompt, since vendors can change their system prompts at any time without notice. Users should treat the contents as study material and reference patterns, not as authoritative documentation.