A treasure trove for developers and AI enthusiasts has just surfaced online — over 6,000 lines of system prompts and configurations from some of the most popular AI-driven coding tools, including Cursor, Bolt, Lovable, and Manus.
Let’s explore what these leaks reveal about how AI thinks, communicates, and solves problems, while uncovering best practices for prompt engineering.
What’s Inside the Leak?
The leaked prompts offer a rare glimpse into the mechanics of AI tools that power vibe-coding — a trend where coding feels intuitive, collaborative, and almost effortless.
Here’s what you’ll find:
- Code Explanation in Simple Terms: These AIs are designed to break down complex code into plain language. For example, Cursor, powered by Claude 3.7 Sonnet, uses prompts that prioritize clarity, ensuring users understand the logic without feeling overwhelmed. It’s like having a patient mentor who avoids jargon unless necessary.
- Bug Interpretation and Debugging: The prompts reveal how AIs like Bolt and Devin approach bugs. Bolt, operating in a WebContainer environment, is constrained to browser-native code (like JS and WebAssembly) and explicitly warns users about limitations, such as the inability to use pip for Python dependencies. Devin, on the other hand, takes a methodical approach, gathering context and verifying fixes before reporting completion, ensuring error-free solutions.
- User-Friendly Communication: Lovable and Manus shine in how they interact with users. Lovable, an AI editor for web apps, uses prompts that encourage a friendly tone, offering guidance without always modifying code. It ensures users feel supported, not intimidated, by explaining changes clearly. Manus, similarly, focuses on structured responses, separating explanations from outputs to maintain readability.
- Prompt Engineering Goldmine: For anyone looking to level up their prompt engineering skills, this leak is invaluable. The prompts show how to chain instructions, handle constraints, and optimize AI behavior. For instance, Cursor’s prompts emphasize grouping code edits into single tool calls to ensure functionality, while Bolt’s guidelines highlight the importance of specifying environmental limitations upfront.
Why This Matters
These leaks aren’t just a peek behind the curtain — they’re a masterclass in AI design and interaction. Developers can learn how to craft better prompts for their own projects, ensuring AI tools deliver precise, user-friendly results. Understanding how these systems interpret bugs or communicate with users can also help predict AI behavior and work around limitations. For product designers, the prompts offer insights into building AI features that feel intuitive and supportive.
Also read:
- Telegram Drops a Game-Changing Update — One of Its Best Yet
- Veo 3 Misused to Generate Fakes About Riots, Conflicts, and Election Fraud
- Substack’s CEO Reveals Over 50 Authors Earn $1M+ Annually Through Paid Subscriptions
Steal the Best Practices
The GitHub repo is a goldmine for anyone interested in vibe-coding or AI development. Dive in, study how these tools think, and adopt their strategies to enhance your own work. Whether you’re a developer, a prompt engineer, or just curious about AI, there’s something here for you to learn and apply.
