Prompt Engineering Skills Library: Best Practices for Claude/GPT/Gemini Multi-Model Prompting

Project Overview: A One-Stop Multi-Model Prompt Skills Collection

Anyone building AI applications knows that the same prompt can produce drastically different results when fed to Claude versus GPT. Writing effective prompts tailored to different models has become an essential skill that developers simply can't avoid.

The open-source GitHub project treylom/prompt-engineering-skills addresses this exact pain point — it provides a comprehensive collection of prompt engineering skills covering mainstream AI models, including model-specific patterns and battle-tested best practices.

The project currently has 78 stars and 18 forks. While not massive in scale, it stands out for its broad coverage and targeted approach, making it a practical reference worth bookmarking for any prompt engineer.

Which Models and Platforms Are Supported?

Claude Series

The project provides dedicated prompt patterns for Claude Code as well as Claude Opus 4.7/4.6. Claude is renowned for its powerful XML-structured comprehension capabilities, and the project includes specific patterns and techniques that fully leverage this feature, helping developers achieve more precise interactions on the Claude platform.

Claude's exceptional XML tag parsing ability stems from Anthropic's special optimization for structured inputs during training. XML (Extensible Markup Language) is a markup language for data transmission and storage with strict hierarchical nesting and tag closure rules. When developers wrap different parts of their prompts in XML tags, Claude can precisely identify the semantic role and hierarchical relationships of each section, resulting in higher instruction-following accuracy when handling complex multi-step tasks. This capability makes Claude particularly outstanding in enterprise applications that require strict output format control.

ChatGPT GPTs Series

For OpenAI's product line, the project makes a key distinction: GPT-5.5 uses an "outcome-first" pattern, while GPT-5.4 follows the traditional XML pattern.

This distinction reflects a significant shift in prompt strategy driven by OpenAI's model architecture evolution — newer versions prefer developers to directly describe desired outcomes rather than guiding model behavior through structured markup.

GPTs is a custom AI assistant creation platform launched by OpenAI in late 2023, allowing users to build domain-specific AI applications through system prompts, knowledge base uploads, and API calls — no coding required. Google's Gemini Gems is a similar competing feature launched in 2024, enabling users to create personalized AI assistants based on the Gemini model. The core competitive advantage of both platforms lies in the quality of system prompts — a well-crafted system prompt can transform a general-purpose model into a highly specialized domain assistant, and the prompt writing approach needs to be strictly adapted to each platform's model characteristics.

Additionally, the project covers prompt techniques for the gpt-image-2 image generation model.

Google Gemini Ecosystem

The project covers Gemini 3.1, the Veo 3.1 video generation model, and prompt patterns for Gemini Gems custom AI assistants. It also includes content related to "Nano Banana 2," which likely represents specific optimization strategies for Google's lightweight models.

Veo is a video generation AI model developed by Google DeepMind that can generate high-quality video clips from text descriptions. Unlike image generation, video generation requires the model to understand temporal coherence — including object motion trajectories, lighting changes, camera movements, and other cinematic language elements. Prompt engineering for video generation models is more complex than for text or image generation. Developers need to precisely describe temporal sequences, camera techniques, scene transitions, and other dynamic elements in their prompts while controlling style consistency and physical plausibility.

Core Value: Why Model-Specific Prompt Patterns Matter

Different AI models have significant differences in training data, architecture design, and alignment strategies. A prompt that works beautifully on Claude often delivers disappointing results when copied directly to GPT or Gemini.

Alignment refers to the technical process of making AI model behavior conform to human intentions and values. Different companies' alignment strategies directly impact optimal prompt writing approaches. Anthropic's Constitutional AI method makes Claude inclined to follow explicit rules and structured instructions; OpenAI's InstructGPT approach trains models to understand implicit intentions through extensive human preference data; Google's alignment strategy places greater emphasis on safety in multimodal scenarios. These underlying differences mean that expressing the same request — say, "please help me write an article" — requires different contextual framing and constraint expressions on different models to achieve optimal results.

The project's core value manifests on three levels:

1. Targeted Optimization: Provides battle-tested best practice patterns for each model
2. Version Awareness: Distinguishes strategy differences between versions of the same model (e.g., GPT-5.5 and 5.4 require completely different prompt approaches)
3. Cross-Platform Reference: Helps developers understand the "thinking style" differences across models and avoid common pitfalls

From XML to Outcome-First: A Prompt Paradigm Shift in Progress

There's a particularly noteworthy trend in this project: the shift from structured XML prompts to "outcome-first" prompts.

In the GPT-5.4 era, developers were accustomed to using XML tags to organize the structure of complex prompts — for example, using tags like <role>, <task>, and <constraints> to categorize instructions. With GPT-5.5, OpenAI now encourages developers to directly describe expected outputs and behavioral constraints without relying on structured markup.

The emergence of the outcome-first pattern is closely related to improvements in large language model reasoning capabilities. Earlier models required developers to "hand-hold" the reasoning path through structured markup, essentially because the model's autonomous planning ability was limited. As model scale has increased and RLHF (Reinforcement Learning from Human Feedback) technology has matured, the new generation of models possesses stronger intent understanding and autonomous task decomposition capabilities. Developers only need to clearly describe the desired end result and constraints, and the model can plan its own execution path. This is similar to the transition from "imperative programming" to "declarative programming" — you tell the system "what you want" rather than "how to do it."

This shift cuts both ways: on one hand, it lowers the technical barrier to prompt writing; on the other, it requires developers to have clearer requirement articulation skills — you need to truly think through what you want, rather than relying on templates to piece together structure.

Who Should Use This Project?

This project is particularly suitable for:

• Multi-platform AI application developers: Teams that need to interface with Claude, GPT, Gemini, and other models simultaneously can quickly look up prompt differences across models
• Prompt engineers: Those looking to systematically improve cross-model prompting skills rather than only mastering a single platform
• GPTs/Gems builders: Developers and creators building custom AI assistants on OpenAI or Google platforms
• AI content creators: Creative professionals who need to effectively use image generation (gpt-image-2) and video generation (Veo 3.1) models

Conclusion: The Trend Toward Model-Specific Prompt Engineering

As AI models iterate rapidly, prompt engineering is evolving from a "one-size-fits-all" approach to a "tailored-to-each-model" discipline. The prompt-engineering-skills project represents the practical implementation of this trend — rather than pursuing a universal prompt template, it acknowledges and embraces the differences between models.

Prompt Engineering has matured from an emerging concept in 2023 into a well-established professional field. Based on industry observations, enterprise demand for prompt engineers is concentrated in three areas: product-facing system prompt design that directly impacts AI product user experience; efficiency-focused workflow automation prompt optimization; and security-focused prompt injection defense and red team testing. As model capabilities rapidly iterate, the core challenge for prompt engineers is no longer "how to make models understand instructions" but rather "how to efficiently transfer best practices across models" and "how to keep up with strategy changes brought by model version updates."

For developers who need to maintain competitiveness in multi-model environments, understanding and mastering these model-specific prompt patterns is becoming an increasingly critical professional skill. Rather than spending time searching for "universal prompts," it's better to thoroughly internalize the best practices for the models you use most frequently.