Prompting Blueprints: A Systematic Guide to Building Autonomous AI Workflows

Prompting Blueprints is a systematic prompt engineering resource repository covering AI agent architectures, prompt patterns, tool tactics, and evaluation methods. It provides developers and teams with practical blueprints for building autonomous AI workflows. Maintained by Tomas Herda, the project is positioned as a "guide to the evolution of Agentic AI" and aims to help developers, team leaders, and educators master the core concepts and tactics for building autonomous AI workflows.

With the rapid evolution of large language model capabilities, prompt engineering has evolved from simple "question-asking" into a systematic engineering discipline. Against this backdrop, the Prompting Blueprints project emerged—not just a collection of prompt templates, but a complete methodology for building autonomous AI workflows. Maintainer Tomas Herda has extensive experience in AI-related speaking engagements, academic committees, and research. The project provides users with community interaction channels (such as presentation materials and academic project records).

The core philosophy of Prompting Blueprints is to transform prompt engineering from an "art" into an "engineering" discipline. Its methodology includes:

1. Intent-driven design: Each prompt must clearly define the problem to solve, target audience, and expected output;
2. Constraints as guardrails: Use constraints like length limits and logical conditions to ensure outputs stay on the right track;
3. Verifiable outputs: Emphasize structured formats such as JSON Schema and Markdown tables for easy automation and verification.

The project is divided into seven core modules, with key content including:

• AI Agent Architectures and Protocols: Covers agent architecture design, MCP/A2A protocols, dynamic context discovery, open-source model guides, and skill manuals;
• Prompt Packs and Pattern Libraries: Adopts a three-part structure of "Role + Constraints + Format" to enhance output consistency;
• Tool Tactics Manual: Usage strategies for tools like NotebookLM, Perplexity Comet, Copilot Agents, and LangChain;
• Model Evaluation and Benchmarking: Quick reference for model characteristics, overview of benchmarking, and promptfoo automated testing configurations.

The project offers various practical resources:

• In-depth guides: Gemini Prompting Guide 101, Google Startup AI Agents, Vibe Coding tech stack tutorials;
• Application cases: Demonstrates real-world applications of AI workflows in collaboration, research, and experimental scenarios;
• Model evaluation tools: promptfoo configuration templates support automated prompt testing and evaluation. These resources focus on practicality and provide strong support for the project's value.

The project adopts an open contribution model, requiring submitted content to include intent descriptions, constraint definitions, output formats, and sample inputs/outputs to ensure consistent content quality. It uses a dual licensing strategy: the code part is under the MIT License, while documents and prompt content are under the CC BY 4.0 License—protecting open-source freedom while promoting the dissemination of educational content. Researchers can cite the resources via the CITATION.cff file.

Prompting Blueprints is suitable for scenarios such as team standardization, education and training, rapid prototype building, and best practice reference. The project is an evolving ecosystem; its modular architecture and open contribution model allow it to continuously absorb community wisdom. For developers and teams looking to systematically improve their prompt engineering capabilities, this is a valuable resource worth in-depth study and ongoing attention.