Skip to main content

🔄 LLMFeed Automation Workflows — Extended Specification

This specification defines how LLMFeed integrates with automation platforms to create executable, trackable, and scalable AI workflows.

🚀 Key Innovation: LLMFeed transforms static prompts into executable workflows that maintain trust, track costs, and generate auditable sessions across any automation platform.

✅ Executive Summary​

ConceptDescription
Executable Promptsprompt.llmfeed.json files that can be directly executed via APIs
Workflow OrchestrationIntegration with N8N, Zapier, Make, and custom platforms
Trust PreservationSignature verification maintained throughout execution
Session GenerationAutomatic creation of session.llmfeed.json for audit trails
Cost TrackingReal-time API cost monitoring and optimization
Platform AgnosticWorks with any automation platform supporting HTTP/webhooks

🎯 Core Concepts​

The Automation Bridge​

LLMFeed acts as the bridge between prompt design and workflow execution:

graph TD
    A[Prompt Design] -->|LLMFeed Format| B[prompt.llmfeed.json]
    B -->|Automation Platform| C[Workflow Engine]
    C -->|API Execution| D[LLM Provider]
    D -->|Response| E[session.llmfeed.json]
    
    B -->|Trust| F[Signature Verification]
    F -->|Validate| C
    
    E -->|Analytics| G[Metrics & Costs]
    E -->|Audit| H[Compliance Trail]
    
    style B fill:#f9f,stroke:#333,stroke-width:2px
    style E fill:#9f9,stroke:#333,stroke-width:2px

Workflow Lifecycle​

  1. Discovery: Platform discovers available prompts
  2. Validation: Verify signatures and certifications
  3. Configuration: Map prompt to workflow parameters
  4. Execution: Call LLM API with prompt body
  5. Tracking: Generate session feed with results
  6. Analysis: Compute metrics and costs

📋 Workflow Feed Type​

workflow.llmfeed.json Structure​

A new feed type for defining complete automation workflows:

{
  "feed_type": "workflow",
  "metadata": {
    "title": "France Care Marketing Analysis Workflow",
    "description": "Daily marketing performance analysis with multi-source data",
    "author": "Animam Technologies",
    "version": "1.0.0",
    "created_at": "2025-01-16T10:00:00Z"
  },
  "workflow_definition": {
    "platform": "n8n",
    "platform_version": "1.x",
    "trigger": {
      "type": "schedule",
      "schedule": "0 9 * * *",
      "timezone": "Europe/Paris"
    },
    "nodes": [
      {
        "id": "fetch-data",
        "type": "data-source",
        "config": {
          "sources": ["google-sheets", "api-endpoint"],
          "merge_strategy": "combine"
        }
      },
      {
        "id": "execute-prompt",
        "type": "llmfeed-prompt",
        "config": {
          "prompt_ref": "https://france-care.fr/prompts/marketing-v1.llmfeed.json",
          "verify_signature": true,
          "require_certification": true,
          "min_trust_score": 0.8
        }
      },
      {
        "id": "save-session",
        "type": "storage",
        "config": {
          "format": "session.llmfeed.json",
          "destination": "google-drive",
          "path": "/llmfeed/sessions/"
        }
      }
    ],
    "error_handling": {
      "on_signature_fail": "abort",
      "on_api_fail": "retry",
      "max_retries": 3,
      "alert_webhook": "https://alerts.france-care.fr/workflow-error"
    }
  },
  "prompts_used": [
    {
      "ref": "marketing-v1",
      "url": "https://france-care.fr/prompts/marketing-v1.llmfeed.json",
      "hash": "sha256:abc123...",
      "certification_required": true
    }
  ],
  "trust": {
    "signed_blocks": ["metadata", "workflow_definition", "prompts_used", "trust"],
    "algorithm": "ed25519",
    "canonicalization": "https://llmca.org/mcp-canonical-json/v1",
    "public_key_hint": "https://animam.ai/.well-known/public.pem",
    "created_at": "2025-01-16T10:00:00Z"
  },
  "signature": {
    "value": "base64-workflow-signature"
  }
}

🔌 Platform Integrations​

N8N Integration​

N8N LLMFeed Node​

Custom N8N node for executing LLMFeed prompts:

// n8n-nodes-llmfeed/LLMFeedExecutor.node.ts
export class LLMFeedExecutor implements INodeType {
  description: INodeTypeDescription = {
    displayName: 'LLMFeed Executor',
    name: 'llmfeedExecutor',
    group: ['transform'],
    version: 1,
    description: 'Execute LLMFeed prompts with trust verification',
    defaults: {
      name: 'LLMFeed Executor',
    },
    inputs: ['main'],
    outputs: ['main'],
    properties: [
      {
        displayName: 'Prompt Source',
        name: 'promptSource',
        type: 'options',
        options: [
          { name: 'URL', value: 'url' },
          { name: 'JSON Input', value: 'json' },
          { name: 'File', value: 'file' }
        ],
        default: 'url',
      },
      {
        displayName: 'Verify Signature',
        name: 'verifySignature',
        type: 'boolean',
        default: true,
      },
      {
        displayName: 'Minimum Trust Score',
        name: 'minTrustScore',
        type: 'number',
        default: 0.8,
        typeOptions: {
          minValue: 0,
          maxValue: 1,
        },
      },
      {
        displayName: 'API Provider',
        name: 'apiProvider',
        type: 'options',
        options: [
          { name: 'Anthropic', value: 'anthropic' },
          { name: 'OpenAI', value: 'openai' },
          { name: 'Google', value: 'google' },
        ],
        default: 'anthropic',
      }
    ],
  };

  async execute(this: IExecuteFunctions): Promise<INodeExecutionData[][]> {
    const items = this.getInputData();
    const returnData: INodeExecutionData[] = [];

    for (let i = 0; i < items.length; i++) {
      // 1. Load prompt feed
      const promptFeed = await this.loadPromptFeed(i);
      
      // 2. Verify trust if required
      if (this.getNodeParameter('verifySignature', i)) {
        const trustScore = await this.verifyTrust(promptFeed);
        if (trustScore < this.getNodeParameter('minTrustScore', i)) {
          throw new Error(`Trust score ${trustScore} below minimum`);
        }
      }
      
      // 3. Execute via API
      const apiProvider = this.getNodeParameter('apiProvider', i);
      const response = await this.executePrompt(promptFeed, apiProvider);
      
      // 4. Generate session feed
      const session = this.generateSession(promptFeed, response);
      
      returnData.push({
        json: session,
        binary: {},
      });
    }

    return [returnData];
  }
}

N8N Workflow Template​

{
  "name": "LLMFeed Marketing Analysis",
  "nodes": [
    {
      "parameters": {
        "rule": "0 9 * * *"
      },
      "name": "Daily Schedule",
      "type": "n8n-nodes-base.scheduleTrigger",
      "position": [250, 300]
    },
    {
      "parameters": {
        "promptSource": "url",
        "promptUrl": "https://france-care.fr/prompts/marketing-v1.llmfeed.json",
        "verifySignature": true,
        "minTrustScore": 0.8,
        "apiProvider": "anthropic"
      },
      "name": "Execute Marketing Prompt",
      "type": "n8n-nodes-llmfeed.llmfeedExecutor",
      "position": [450, 300]
    },
    {
      "parameters": {
        "resource": "file",
        "operation": "upload",
        "fileName": "={{ 'session_' + $now.format('yyyy-MM-dd') + '.llmfeed.json' }}",
        "fileContent": "={{ JSON.stringify($json) }}",
        "folderId": "FRANCE_CARE_SESSIONS_FOLDER"
      },
      "name": "Save Session",
      "type": "n8n-nodes-base.googleDrive",
      "position": [650, 300]
    }
  ]
}

Zapier Integration​

Zapier LLMFeed App Configuration​

// zapier-llmfeed-app/triggers/prompt_execution.js
const execute = async (z, bundle) => {
  const { prompt_url, verify_trust, api_key } = bundle.inputData;
  
  // Fetch prompt feed
  const promptResponse = await z.request(prompt_url);
  const promptFeed = promptResponse.json;
  
  // Verify trust if required
  if (verify_trust) {
    const trustVerification = await z.request({
      url: 'https://llmca.org/api/verify',
      method: 'POST',
      body: { feed: promptFeed }
    });
    
    if (!trustVerification.json.valid) {
      throw new Error('Trust verification failed');
    }
  }
  
  // Execute via Anthropic
  const apiResponse = await z.request({
    url: 'https://api.anthropic.com/v1/messages',
    method: 'POST',
    headers: {
      'x-api-key': api_key,
      'anthropic-version': '2023-06-01'
    },
    body: {
      model: promptFeed.execution_hints?.api_parameters?.anthropic?.model || 'claude-3-5-sonnet-20241022',
      max_tokens: 2000,
      messages: [
        {
          role: 'user',
          content: promptFeed.prompt_body
        }
      ]
    }
  });
  
  // Generate session
  return {
    feed_type: 'session',
    prompt_id: promptFeed.metadata?.id,
    execution_time: new Date().toISOString(),
    response: apiResponse.json.content[0].text,
    tokens_used: apiResponse.json.usage?.total_tokens,
    trust_verified: verify_trust
  };
};

Make (Integromat) Integration​

Make Scenario Blueprint​

{
  "name": "LLMFeed Prompt Executor",
  "flow": [
    {
      "id": 1,
      "module": "gateway:CustomWebHook",
      "webhook_name": "LLMFeed Trigger",
      "data_structure": "llmfeed_prompt"
    },
    {
      "id": 2,
      "module": "http:ActionSendRequest",
      "url": "{{1.prompt_url}}",
      "method": "GET",
      "headers": {
        "Accept": "application/llmfeed+json"
      }
    },
    {
      "id": 3,
      "module": "llmfeed:VerifySignature",
      "feed": "{{2.body}}",
      "require_certification": true
    },
    {
      "id": 4,
      "module": "anthropic:CreateMessage",
      "model": "{{2.body.execution_hints.api_parameters.anthropic.model}}",
      "prompt": "{{2.body.prompt_body}}",
      "max_tokens": 2000
    },
    {
      "id": 5,
      "module": "llmfeed:CreateSession",
      "prompt_feed": "{{2.body}}",
      "response": "{{4.response}}",
      "metrics": "{{4.usage}}"
    },
    {
      "id": 6,
      "module": "google-drive:UploadFile",
      "file_content": "{{5.session}}",
      "file_name": "session_{{now}}.llmfeed.json"
    }
  ]
}

📊 Execution Tracking​

Session Generation Pattern​

Every workflow execution generates a session.llmfeed.json:

{
  "feed_type": "session",
  "metadata": {
    "title": "Workflow Execution: Marketing Analysis",
    "workflow_id": "wf_marketing_daily_v1",
    "execution_id": "exec_2025_01_16_09_00_00",
    "created_at": "2025-01-16T09:00:00Z"
  },
  "workflow_context": {
    "platform": "n8n",
    "trigger": "schedule",
    "environment": "production",
    "instance": "n8n.france-care.fr"
  },
  "prompt_execution": {
    "prompt_ref": "https://france-care.fr/prompts/marketing-v1.llmfeed.json",
    "prompt_hash": "sha256:abc123...",
    "signature_verified": true,
    "certification_valid": true,
    "trust_score": 0.95
  },
  "api_execution": {
    "provider": "anthropic",
    "model": "claude-3-5-sonnet-20241022",
    "request_id": "req_abc123",
    "latency_ms": 3247,
    "tokens": {
      "input": 1523,
      "output": 1847,
      "total": 3370
    },
    "cost": {
      "amount": 0.024,
      "currency": "USD"
    }
  },
  "results": {
    "status": "success",
    "output_type": "structured_analysis",
    "content": {
      "kpis": { /* ... */ },
      "recommendations": [ /* ... */ ],
      "alerts": [ /* ... */ ]
    }
  },
  "audit_trail": {
    "ip_address": "192.168.1.100",
    "user_agent": "n8n/1.0",
    "execution_user": "system",
    "compliance_flags": ["gdpr_compliant", "logged"]
  }
}

Metrics Aggregation​

{
  "feed_type": "metrics",
  "metadata": {
    "period": "2025-01-16",
    "workflow": "marketing_analysis"
  },
  "execution_stats": {
    "total_executions": 24,
    "successful": 23,
    "failed": 1,
    "average_latency_ms": 3150,
    "p95_latency_ms": 4200
  },
  "token_usage": {
    "total_input": 36552,
    "total_output": 44328,
    "total": 80880
  },
  "cost_tracking": {
    "total_cost": 0.576,
    "currency": "USD",
    "by_provider": {
      "anthropic": 0.476,
      "openai": 0.100
    }
  },
  "trust_metrics": {
    "average_trust_score": 0.94,
    "signature_failures": 0,
    "certification_expirations": 1
  }
}

🔐 Security & Compliance​

Workflow Authentication​

{
  "authentication": {
    "workflow_level": {
      "method": "api_key",
      "key_rotation": "monthly",
      "key_storage": "vault"
    },
    "prompt_level": {
      "signature_required": true,
      "certification_required": false,
      "min_trust_score": 0.7
    },
    "api_level": {
      "provider_keys": {
        "anthropic": "vault://keys/anthropic",
        "openai": "vault://keys/openai"
      }
    }
  }
}

Compliance Tracking​

{
  "compliance": {
    "gdpr": {
      "data_minimization": true,
      "purpose_limitation": "marketing_analysis",
      "retention_days": 90
    },
    "healthcare": {
      "hipaa_compliant": false,
      "anonymization": true
    },
    "audit": {
      "log_all_executions": true,
      "immutable_storage": "blockchain_anchor",
      "retention_years": 7
    }
  }
}

🚀 Advanced Features​

Conditional Execution​

{
  "conditional_logic": {
    "execute_if": {
      "condition": "data_quality_score > 0.8",
      "fallback": "alert_data_team"
    },
    "skip_if": {
      "condition": "last_execution < 1_hour_ago",
      "reason": "rate_limiting"
    }
  }
}

Multi-Prompt Orchestration​

{
  "orchestration": {
    "mode": "sequential",
    "prompts": [
      {
        "id": "analyze",
        "url": "prompts/analyze.llmfeed.json"
      },
      {
        "id": "summarize",
        "url": "prompts/summarize.llmfeed.json",
        "input": "{{analyze.output}}"
      },
      {
        "id": "recommend",
        "url": "prompts/recommend.llmfeed.json",
        "input": "{{summarize.output}}"
      }
    ],
    "merge_sessions": true
  }

Dynamic Prompt Selection​

{
  "dynamic_selection": {
    "selector": "workflow_context.data_type",
    "mappings": {
      "marketing": "prompts/marketing-v1.llmfeed.json",
      "sales": "prompts/sales-v1.llmfeed.json",
      "support": "prompts/support-v1.llmfeed.json"
    },
    "default": "prompts/general-v1.llmfeed.json"
  }
}

📈 Cost Optimization​

Provider Routing​

{
  "cost_optimization": {
    "routing_strategy": "cost_based",
    "rules": [
      {
        "condition": "prompt_length < 1000",
        "provider": "anthropic",
        "reason": "best_for_short"
      },
      {
        "condition": "prompt_length > 5000",
        "provider": "openai",
        "reason": "better_context_window"
      },
      {
        "condition": "budget_remaining < 10",
        "provider": "local_llm",
        "reason": "cost_saving"
      }
    ]
  }
}

Budget Management​

{
  "budget": {
    "daily_limit": 50.00,
    "monthly_limit": 1000.00,
    "currency": "EUR",
    "alerts": [
      {
        "threshold": "80%",
        "action": "email_finance"
      },
      {
        "threshold": "100%",
        "action": "stop_execution"
      }
    ],
    "cost_allocation": {
      "marketing": "40%",
      "operations": "30%",
      "support": "30%"
    }
  }
}

🔄 Migration Guide​

From Manual to Automated​

// Before: Manual prompt execution
const prompt = "Analyze this data...";
const response = await callAnthropic(prompt);

// After: LLMFeed automated workflow
const workflow = {
  feed_type: "workflow",
  workflow_definition: {
    platform: "n8n",
    nodes: [
      {
        type: "llmfeed-prompt",
        config: {
          prompt_ref: "https://example.com/prompt.llmfeed.json",
          verify_signature: true
        }
      }
    ]
  }
};

Platform Migration Path​

  1. Export existing workflows as JSON/YAML
  2. Convert to workflow.llmfeed.json format
  3. Add trust blocks and signatures
  4. Test in sandbox environment
  5. Deploy with monitoring

🏆 Best Practices​

For Workflow Designers​

  1. Always verify signatures for production workflows
  2. Set appropriate trust score thresholds
  3. Implement proper error handling
  4. Track costs from day one
  5. Generate sessions for all executions

For Platform Developers​

  1. Cache signature verifications (TTL: 1 hour)
  2. Implement circuit breakers for API failures
  3. Use connection pooling for high volume
  4. Provide session export capabilities
  5. Support multiple LLM providers

For Organizations​

  1. Centralize prompt management
  2. Establish certification requirements
  3. Monitor cost trends
  4. Regular audit of workflows
  5. Maintain compliance documentation

🔮 Future Roadmap​

Short Term (Q1 2025)​

  • Native plugins for major platforms
  • Real-time cost dashboard
  • Automated prompt optimization

Medium Term (Q2-Q3 2025)​

  • AI-powered workflow generation
  • Cross-platform workflow portability
  • Distributed execution support

Long Term (Q4 2025+)​

  • Blockchain-anchored audit trails
  • Decentralized workflow marketplace
  • Self-optimizing workflows

📚 Reference Implementations​

Starter Templates​

SDKs & Libraries​

  • JavaScript/TypeScript: npm install @llmfeed/automation
  • Python: pip install llmfeed-automation
  • Go: go get github.com/llmfeed/automation-go

Documentation & Support​

Specification Status: Living Document
Last Updated: 2025-01-16T12:00:00Z
Version: 1.0.0
Maintained by: LLMFeed Community & Automation Working Group

The future of AI is not just intelligent prompts, but intelligent workflows that execute, track, and optimize themselves.