Skip to content

agent_specifications.md

Latest commit

 

History

History
165 lines (127 loc) · 5.59 KB

agent_specifications.md

File metadata and controls

165 lines (127 loc) · 5.59 KB

Agent Specifications

This document provides technical specifications for the AI GTM Stakeholder Engine's 11-agent pipeline, focusing on key agents, execution flow, and integration patterns.

System Overview

The system consists of a meta-orchestration layer (Agents 0a/0b) followed by a 9-agent content generation pipeline, concluding with cloud database integration (Agent 10). The pipeline executes in a single unified flow without manual intervention points.

Phase 0: System Configuration & Orchestration

Agent 0a: Project Configurator

Technical Purpose: Automated project workspace creation and industry-specific system prompt generation

Implementation Details:

  • Script: generate_simple.py with interactive configuration interface
  • Output: Complete project workspace with tailored prompts across all 11 agents
  • Project Structure: Generates {industry}_{product}_{timestamp} naming convention
  • Prompt Engineering: Dynamic system prompt customization based on industry context

Input Requirements:

{
  "industry": "string (e.g., 'SaaS', 'E-commerce', 'Healthcare')",
  "product_name": "string",
  "target_audience": "string",
  "business_model": "string"
}

Technical Dependencies:

  • Python file system operations
  • JSON configuration parsing
  • Template string interpolation for prompt generation

Agent 0b: Pipeline Orchestrator

Technical Purpose: Automated unified pipeline execution with state tracking

Implementation Architecture:

  • Unified Script: run_unified_pipeline.py - Single continuous execution
  • Execution Sequence: 1→2→5→7→8→9→3→4→10 (no manual stops)
  • State Management: JSON-based execution tracking with checkpoint persistence
  • Dependency Management: Automatic file copying between agents based on requirements

Pipeline Execution Flow:

Agent 1 (Message House) → Agents 2, 5, 9
Agent 2 (Brand Personas) → Agents 4, 5, 7, 8
Agent 3 (Customer Analysis) → Agents 4, 5, 7, 8, 9
Agent 5 (Keywords) → Agents 7, 8, 9
All Agents (1-9) → Agent 0b copies all outputs → Agent 10 (RAG Upload ALL)

State Tracking Structure:

{
  "project_name": "saas_projectmgmt_20250124",
  "pipeline_type": "unified",
  "completed_agents": [1, 2, 5, 7, 8, 9],
  "failed_agents": [],
  "auto_approval_enabled": true
}

Key Features: Cross-agent file management, dependency resolution, error recovery

User Input Process

12 Strategic Questions

Purpose: User provides strategic foundation through structured questionnaire

Input Requirements:

  • Template: Available at ../examples/12_questions_template.md
  • Format: Markdown file with question-answer pairs
  • Placement: Copy completed file to message_house_agent/1_input/{project}/

Strategic Questions Cover:

  • Product positioning and value proposition
  • Target audience and customer segments
  • Competitive landscape and differentiation
  • Business model and strategic objectives

Core Agent Pipeline

Agent 1: Message House (Strategic Foundation)

Purpose: Processes 12 strategic questions into core messaging framework Input: User's completed strategic questionnaire Output: Strategic messaging document Dependencies: None (pipeline starting point)

Agents 2-9: Content Generation Pipeline

High-Level Flow:

  • Agent 2: Brand personas from strategic foundation
  • Agent 3: Customer analysis (optional customer data)
  • Agent 4: Gap analysis between brand and customer perspectives
  • Agent 5: Keywords expansion and vocabulary building
  • Agent 7: Testimonial generation
  • Agent 8: Social media content creation
  • Agent 9: Website copy generation

Execution Pattern: Each agent processes inputs from dependencies, generates outputs for downstream agents

Agent 10: RAG System & Cloud Integration

Technical Architecture

Purpose: Processes all pipeline outputs and uploads to vector database for global team access

Technology Stack:

  • Framework: LangChain (TypeScript/Node.js)
  • Embeddings: OpenAI text-embedding-ada-002
  • Vector Database: Compatible with pgvector extension
  • Bridge: Python script calls Node.js RAG system

Processing Workflow

Agent 0b copies ALL pipeline outputs → Agent 10 collection → Chunking → Embedding → Vector Database → Frontend Access

Data Processing:

  1. Collection: Reads all .md files from all agents (1-9) via rag_system_agent/1_input/{project}/
  2. Chunking: Recursive character text splitter (1200 chars, 300 overlap)
  3. Embedding: OpenAI embeddings generation
  4. Storage: Vector database with project-specific metadata

Database Configuration

Vector Database: Compatible with pgvector extension Recommended: Supabase (built-in pgvector, real-time, free tier) Alternatives: PostgreSQL + pgvector, Pinecone, Weaviate

Configuration:

DATABASE_URL=postgresql://[user]:[password]@[host]:[port]/[database]
OPENAI_API_KEY=your_openai_api_key
PROJECT_ID=saas_projectmgmt_20250124

Technical Integration

API Configuration

Shared Configuration (config.json):

{
  "api_key": "your_anthropic_api_key",
  "current_project": "saas_projectmgmt_20250124",
  "model": "claude-3-5-sonnet-20241022",
  "project_isolation": true
}

Pipeline Execution

Command: python run_unified_pipeline.py Result: Complete pipeline execution from user input to cloud database upload

For complete data storage architecture, see data_store_specification.md