Why do different model architectures require different prompts?

Autoregressive models like Claude follow step-by-step instructions, whereas reasoning models like o1 perform better when you define the search space and constraints rather than the thinking process itself.

What is the Triangulation Constraint technique?

It is a strategy that forces a reasoning model to hold multiple conflicting perspectives simultaneously and synthesize a solution that addresses all of them, leading to more balanced results.

How do negative constraints improve AI performance?

By removing 'easy outs'—such as hedging or generic advice—negative constraints force the model to provide specific, decisive, and high-quality technical answers.

When should I use the Scenario Injection pattern?

Use it when you need to turn abstract requests into high-stakes simulations, which anchors reasoning models into specific domain logic and prevents superficial analysis.

Prompt Craft

Name: Prompt Craft
Author: yzavyas

byyzavyas

•

Data Science & ML

Optimizes AI prompts by applying architecture-specific techniques for autoregressive and reasoning-based models.

Prompt Craft is a specialized skill designed to help developers engineer high-performance prompts tailored to specific AI model architectures. It bridges the performance gap between traditional autoregressive models like Claude or GPT-4 and internal reasoning models like o1 or Gemini Deep Think. By providing structured patterns such as chain-of-thought, triangulation constraints, and scenario injection, this skill ensures that your AI interactions are optimized for the specific way each model processes information, leading to more accurate, rigorous, and domain-specific outputs.

Key Features

01Architecture-specific prompting strategies for Claude, GPT-4, and o1

024 GitHub stars

03Chain-of-thought and structured reasoning implementation

04Search space constraint patterns for reasoning models

05Scenario injection for high-stakes simulation and debugging

06Triangulation techniques to synthesize conflicting hypotheses

Use Cases

01Optimizing technical code generation through negative constraints

02Adapting existing prompts for next-generation reasoning models

03Designing robust system prompts for complex AI agents

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yzavyas/claude-1337 prompt-craft

For use in Claude.ai and ChatGPT

Download Skill