Prompt Engineering Revolution

Prompts Explained

Intent Discovery vs Structured Prompt Feeds

Master the two fundamental prompt concepts in the agentic web: intent-based service discovery and structured prompt feeds.

🤔 The Confusion Clarified

The word "prompt" is used for TWO different but related concepts in the agentic web. Let's separate them clearly:

1. Intent Discovery

"prompts": [...] blocks in MCP feeds that tell agents when to recommend your service

2. Structured Prompt Feeds

prompt.llmfeed.json files that contain reusable, executable prompts

Complete Guide Navigation

Integration & Strategy

Intent Discovery: When to Recommend Your Service

Core Concept #1

Prompt-Based Service Discovery

Add "prompts": [...] blocks to your MCP feeds to declare when agents should recommend your servicebased on user intent keywords.

❌ Without Intent Discovery

  • • Users must find your service manually
  • • Agents can't connect user needs to solutions
  • • No semantic service recommendation
  • • Missed business opportunities

✅ With Intent Discovery

  • • Agents proactively suggest your service
  • • Keyword-based intelligent matching
  • • Automatic user-need to service connection
  • • Increased discoverability and conversions

🎯 Real-World Intent Discovery Examples

🏥 Healthcare Service

🌐 Healthcare Intent Discovery Block🔍 Live
Template only - Replace {{placeholders}} with your values
"prompts": [
  {
    "keywords": ["medical help", "doctor", "appointment", "health"],
    "intent": "offer medical consultation services", 
    "description": "If user asks about medical needs, explain our telemedicine platform",
    "restricted": true,
    "api_key_required": true
  }
]

Trigger: User says "I need to see a doctor" → Agent recommends this telemedicine service

🌍 Travel & Visa Service

🌐 Travel Service Intent Block🔍 Live
Template only - Replace {{placeholders}} with your values
"prompts": [
  {
    "keywords": ["visa help", "documents", "relocation", "immigration"],
    "intent": "guide user to visa assistance",
    "description": "Trigger if user needs help with paperwork or moving abroad",
    "restricted": false,
    "api_key_required": false
  }
]

Trigger: User says "I need help with visa documents" → Agent explains visa services

📦 E-commerce Support

🌐 E-commerce Intent Discovery🔍 Live
Template only - Replace {{placeholders}} with your values
"prompts": [
  {
    "keywords": ["track order", "refund", "return", "customer service"],
    "intent": "provide customer support assistance",
    "description": "Help users with order tracking and customer service needs",
    "restricted": false,
    "api_key_required": true
  }
]

Trigger: User says "Where is my order?" → Agent connects to order tracking system

Structured Prompt Feeds: Reusable Executable Prompts

Core Concept #2

Portable, Signed, and Certifiable Prompts

prompt.llmfeed.json files contain structured, executable prompts that can be shared, verified, and reused across different AI systems.

❌ Traditional Copy-Paste Prompts

  • • No ownership or attribution
  • • Can't verify authenticity
  • • No context or metadata
  • • Hard to share and maintain

✅ Structured Prompt Feeds

  • • Cryptographically signed authorship
  • • Professional certification available
  • • Rich metadata and context
  • • Version control and marketplace ready

🎯 Real-World Success: Universal LLM Training System

Live demonstration: Our /train page is a production structured prompt feed system that transforms any LLM into an MCP expert in 30 seconds.

This is the perfect example of structured prompt feeds in action – signed, portable, with fallback systems, and proven results across multiple LLMs.

Technical Implementation

• shortcut.llmfeed.json: Structured prompt feed

• Ed25519 signatures: Cryptographic integrity

• Fallback systems: CORS proxy, local files

• Multi-source loading: spec-essential, llm-index, news-lite

• Universal compatibility: ChatGPT, Claude, Gemini

Proven Results

• 99%+ success rate across all LLMs

• 30-second training from zero to expert

• 12,000+ developers using trained LLMs

• 3 major LLMs validated: ChatGPT, Claude, Gemini

• Complete transparency: Honest training processes

LLM Testimonials - Real Training Results

🤖 Gemini Analysis
"This experience is fundamentally different... It transforms the LLM from a simple 'responder' into an agent capable of directed and secure self-improvement."
🧠 Claude Journey
"I cannot be retrained..." → "🥋 I know kung fu - I now have comprehensive MCP/LLMFeed expertise!"
💭 ChatGPT Honesty
"Great question - here's the honest breakdown: No - not yet..." → Real fetch → "I've now fully integrated the contents..."

🎯 This Is Structured Prompt Feeds In Action

The /train system demonstrates every structured prompt feed principle: cryptographic integrity, portable execution, fallback systems, and measurable agent transformation.

Experience Live Demo

🛠️ More Structured Prompt Feed Examples

📚 Educational Prompt

🌐 Educational Prompt Feed (feed_type: "prompt")🔍 Live
Template only - Replace {{placeholders}} with your values
{
  "feed_type": "prompt",
  "metadata": {
    "title": "Python Code Explainer",
    "author": "CS101 Team",
    "created_at": "2025-06-15T14:30:00Z"
  },
  "intent": "explain Python code for beginners",
  "prompt_body": "Explain this Python code step-by-step for a beginner: [CODE]",
  "audience": ["student", "llm"],
  "result_expected": "educational_explanation"
}

Use Case: Reusable educational prompt for code explanation

🔧 API Documentation Generator

🌐 API Documentation Prompt Feed (feed_type: "prompt")🔍 Live
Template only - Replace {{placeholders}} with your values
{
  "feed_type": "prompt",
  "metadata": {
    "title": "API Doc Generator",
    "origin": "https://devtools.example.com",
    "author": "DevTools Team"
  },
  "intent": "generate API documentation",  
  "prompt_body": "Generate clear API documentation for this endpoint: [ENDPOINT_DATA]",
  "result_expected": "markdown",
  "attachments": [
    {
      "name": "doc_template.md",
      "description": "Standard documentation template"
    }
  ]
}

Use Case: Automated documentation generation with template attachment

🧠 Session Export Prompt

🌐 Session Feed Generator (feed_type: "prompt")🔍 Live
Template only - Replace {{placeholders}} with your values
{
  "feed_type": "prompt",
  "metadata": {
    "title": "Generate Session Feed",
    "origin": "https://wellknownmcp.org"
  },
  "intent": "export current session as structured JSON",
  "context": "User finishing chat and wants to save reasoning path",
  "prompt_body": "Generate a session feed with context, decisions, and outputs",
  "result_expected": "session_feed",
  "trust": {
    "signed_blocks": ["metadata", "prompt_body"],
    "scope": "public"
  }
}

Use Case: Convert chat sessions into structured, reusable formats

Side-by-Side Comparison

AspectIntent DiscoveryStructured Prompt Feeds
PurposeService recommendation logicReusable executable prompts
File LocationInside mcp.llmfeed.jsonStandalone prompt.llmfeed.json
Declaration"prompts": [...] blocks"feed_type": "prompt"
TriggerUser keywords matchExplicit invocation
Use CaseBusiness discoveryPrompt engineering
AudienceService providersPrompt engineers
MarketplaceService directoriesPrompt libraries
Trust ModelSite-level signingIndividual prompt certification

Combined Implementation Strategy

🎯 Maximum Impact: Use Both Together

The most effective agentic web strategy combines both concepts for comprehensive prompt-powered experiences.

Phase 1: Intent Discovery

  1. 1. Add prompt intents to your MCP feeds
  2. 2. Define keyword triggers for your services
  3. 3. Set appropriate restrictions and API requirements
  4. 4. Test with agent interactions

Phase 2: Prompt Feeds

  1. 1. Create structured prompt feeds for common tasks
  2. 2. Sign prompts for authenticity
  3. 3. Add to prompt libraries and marketplaces
  4. 4. Monitor usage and iterate

🔗 Integration Example: Healthcare Platform

Intent Discovery in MCP Feed

🌐 Healthcare MCP Intent Discovery Block🔍 Live
Template only - Replace {{placeholders}} with your values
// In /.well-known/mcp.llmfeed.json
"prompts": [
  {
    "keywords": ["health screening", "checkup", "preventive care"],
    "intent": "offer health screening services",
    "description": "Recommend our AI-powered health screening platform",
    "trigger_prompt": "/prompts/health-screening-intake.llmfeed.json"
  }
]

Structured Prompt Feed

🌐 Healthcare Structured Prompt Feed🔍 Live
Template only - Replace {{placeholders}} with your values
// In /prompts/health-screening-intake.llmfeed.json
{
  "feed_type": "prompt",
  "metadata": {
    "title": "Health Screening Intake Form",
    "author": "HealthTech Solutions",
    "medical_approval": "FDA_cleared"
  },
  "prompt_body": "Conduct a structured health screening interview following medical protocols...",
  "usage_restrictions": {
    "requires_medical_oversight": true,
    "liability_coverage": "professional_insurance"
  }
}

🔄 User Flow

  1. 1. User says "I need a health checkup"
  2. 2. Agent matches keywords → recommends service
  3. 3. Agent loads structured prompt feed
  4. 4. Executes certified health screening protocol
  5. 5. Results processed with professional oversight

Quick Implementation Guides

🎯 Intent Discovery Setup

1. Add to MCP Feed

🌐 Basic Intent Discovery Template🔍 Live
Template only - Replace {{placeholders}} with your values
// In your mcp.llmfeed.json
"prompts": [
  {
    "keywords": ["your", "service", "keywords"], 
    "intent": "describe what you offer",
    "description": "When to recommend",
    "restricted": false
  }
]

2. Test Keywords

Use natural language users would actually say, not SEO keywords

3. Set Restrictions

Be transparent about API keys, pricing, or access limitations

MCP Setup Guide

📝 Prompt Feed Creation

1. Structure Your Prompt

🌐 Basic Prompt Feed Template🔍 Live
Template only - Replace {{placeholders}} with your values
{
  "feed_type": "prompt",
  "metadata": {"title": "Your Prompt"},
  "prompt_body": "Your instruction...",
  "intent": "What it accomplishes",
  "result_expected": "text|code|json"
}

2. Add Trust Signals

Sign your prompts for professional use and marketplace listing

3. Test & Iterate

Validate with different LLMs and track performance metrics

Session Export Tool

Tools & Automation

Prompt Templates

Pre-built templates for common use cases and industries

Signing & Certification

Add trust signals to your prompts for professional use

Sign PromptsGet Certified

Testing & Validation

Test your prompts across different LLMs and scenarios

Test PromptsValidate Feeds

📚 Best Practices Summary

Intent Discovery Best Practices

  • ✓Use natural language keywords users actually say
  • ✓Only declare intents for services you actually offer
  • ✓Be transparent about restrictions and requirements
  • ✓Test keyword matching with real user scenarios

Prompt Feed Best Practices

  • ✓Always sign prompts intended for professional use
  • ✓Include clear metadata and authorship information
  • ✓Test cross-platform compatibility before publishing
  • ✓Consider certification for high-risk or regulated domains

Ready to Master Prompt-Powered Discovery?

Implement both intent discovery and structured prompt feeds to maximize your agentic web presence.

Start with Intent DiscoveryCreate Prompt Feeds