How dontriskit/awesome-ai-system-prompts works
Project type and purpose
• This repository is not a runnable application but an open-source knowledge/reference library of “agentic AI system prompts.”
• Its purpose is to collect, categorize, and analyze real-world system prompts (in Markdown, TypeScript, Python, etc.) and surface best-practice patterns for builders.
File structure
• Top-level directories (Augment, Blackbox.ai, Bolt.new, ChatGPT, Claude-Code, etc.) each contain one system’s prompt examples and any helper scripts or schemas.
• Root files: a global README.md (the overarching guide), LICENSE, older readmes.
• No src/ or build/ directories—content is purely Markdown/text, with some .ts/.js extraction or prompt-definition scripts.
README for additional insights
• The README is the central driver, containing the guide outline (principles, case studies, best practices).
• Each case study in the README corresponds 1:1 with a directory.
Main components
a. Content Layer
• Global Guide (root README.md) that aggregates and analyzes all modules.
b. Build & Deployment Layer (even if ad-hoc)
• Static Site Generator (e.g., Docusaurus, Hugo, or GitHub Pages) that transforms Markdown into HTML/CSS.
• CI/CD Pipeline (e.g., GitHub Actions) that triggers on push, runs linting/spell-check, then builds and deploys the site.
c. Hosting & Delivery
• Web Host (GitHub Pages, Vercel, Netlify) serving the generated site to end users.
d. External Services (optional)
• Search Index (Algolia or Lunr) for full-text search across prompts.
• Analytics (Google Analytics) to track guide usage.
Relationships & interactions
• Authors commit content (Markdown, scripts) → Version Control (GitHub)
• On merge, CI/CD → invokes Static Site Generator → outputs a website artifact
• Hosting Service serves site to End Users (developers, AI builders)
• Analytics embedded in pages collect usage data for maintainers
Architectural patterns/design principles
• Separation of Concerns: content (Markdown) is decoupled from presentation (static site).
• Modular Content: each case study lives in its own directory—easy to add/remove examples.
• “Documentation-as-Code”: treat docs with the same pipelines, linting, versioning, and tests as code.
• Automated Publish: CI/CD ensures consistency and freshness.
Relevant technologies/frameworks
• Markdown, YAML (front-matter, if used)
• Static Site Generator (e.g., Docusaurus, Hugo, Jekyll)
• GitHub Actions for build & deploy
• Hosting: GitHub Pages / Vercel / Netlify
• Optional: Algolia/Lunr for search, Google Analytics
Be detailed but focused
• Prompt directories as leaf-nodes under “Content Modules.”
• Explicitly show the root README as the aggregator/analysis hub.
• This separation highlights how new modules (case studies) plug into the system and get published automatically.