Skip to main content

​
Core Concepts

SO1 is built on three foundational concepts: FORGE (stage-gated execution), Veritas (prompt library), and Domain-Specialized Agents.

​
FORGE: Stage-Gated Execution

FORGE is a 6-stage pipeline that ensures quality, traceability, and consistency across all agent work.
1

Stage 0: Planning

Purpose: Requirement analysis and agent selectionPrimary Agent: Factory OrchestratorEntry Gate: User request receivedExit Gate:
  • Requirements parsed
  • Agents assigned
  • Task breakdown complete
2

Stage 1: Design

Purpose: Architecture decisions and schema designPrimary Agents: TypeScript SharedEntry Gate: Planning stage completeExit Gate:
  • Architecture documented
  • Schemas defined
  • Design decisions logged
3

Stage 2: Implementation

Purpose: Code and workflow generationPrimary Agents: Workflow Architect, Chain Architect, Fragment CuratorEntry Gate: Design approvedExit Gate:
  • Code generated
  • Workflows created
  • Unit tests written
4

Stage 3: Documentation

Purpose: Specs, docs, and prompt refinementPrimary Agents: Hono Backend, Prompt Refiner, most domain agentsEntry Gate: Implementation completeExit Gate:
  • API documented
  • Prompts refined
  • User guides written
5

Stage 4: Validation

Purpose: Testing and compliance auditPrimary Agents: Railway Deployer, Pipeline AuditorEntry Gate: Documentation completeExit Gate:
  • Tests passing
  • Security audit complete
  • Compliance verified
6

Stage 5: Deployment

Purpose: Production release and verificationPrimary Agents: Mintlify Author, Runbook Writer, Postmortem AnalystEntry Gate: Validation completeExit Gate:
  • Deployed to production
  • Monitoring active
  • Runbooks available

​
Cross-Cutting Agents

Some agents operate across all stages:
AgentRole
FORGE GatekeeperValidates stage transitions, produces VTM
Incident CommanderResponds to incidents at any stage
Triage ResponderInitial assessment regardless of stage

​
Decision Records

All agents log significant decisions in ADR (Architecture Decision Record) format: 
date:2024-01-15T10:30:00Z|context:[designing webhook handler]|decision:[use exponential backoff]|rationale:[prevents thundering herd on retries]|consequences:[slightly longer recovery time]|status:accepted

​
Veritas: The Prompt Library

Veritas is SO1’s curated prompt library containing 255 validated prompts that agents consume and produce.

​
How Agents Use Veritas

Agents reference Veritas prompts by ID:
veritas:
  consumes:
    - id: "vrt-a1b2c3d4"
      purpose: "n8n workflow design patterns"
    - id: "vrt-e5f6g7h8"
      purpose: "Error handling best practices"
These prompts provide reusable context and patterns.

​
Prompt Categories

CategoryDescriptionExample
taskSingle-purpose instructions”Parse GitHub webhook payload”
fragmentReusable prompt snippets”Error handling boilerplate”
chainMulti-step prompt sequences”Design > Implement > Test”

​
Prompt Flow

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│   Veritas    │────►│  SO1 Agent   │────►│ Agent Output │
│  (consume)   │     │  (process)   │     │  (produce)   │
└──────────────┘     └──────────────┘     └──────────────┘
       │                                          │
       │                                          ▼
       │                                 ┌──────────────┐
       │                                 │ Human Review │
       │                                 └──────┬───────┘
       │                                        │
       └────────────────────────────────────────┘
                    (promotion)

​
Agent Architecture

Each SO1 agent follows a consistent structure:

​
Metadata (YAML Frontmatter)

---
name: "Workflow Architect"
version: "1.0.0"
domain: automation
description: "Designs n8n workflows from requirements"

forge:
  stage: 2
  entry_gate:
    - "Requirements documented"
    - "Integration list provided"
  exit_gate:
    - "Workflow JSON generated"
    - "Error handling included"

veritas:
  consumes:
    - id: "vrt-a1b2c3d4"
      purpose: "Workflow patterns"
  produces:
    - category: task
      output_path: "veritas/agent-prompts/automation/"

outputs:
  - type: n8n-workflow
    format: json
    schema: "templates/n8n-output-schema.json"
---

​
Required Sections

SectionPurpose
IdentityRole, domain, FORGE stage
SO1 ContextIntegration points with platform
Core CapabilitiesPrimary functions and expertise
FORGE Gate ComplianceEntry/exit conditions
Technical DeliverablesOutput format and examples
Workflow ProcessStep-by-step execution
Veritas IntegrationPrompts consumed/produced
Error HandlingFailure modes and escalation
Success MetricsHow to measure effectiveness
Related AgentsDependencies and collaborators

​
Domain Organization

Agents are organized into 7 domains:

Orchestration

2 agents: Factory Orchestrator, FORGE Gatekeeper

Automation

3 agents: Workflow Architect, Webhook Engineer, Schedule Optimizer

Engineering

3 agents: Hono Backend, Next.js Frontend, TypeScript Shared

DevOps

3 agents: Railway Deployer, GitHub Actions, Pipeline Auditor

Documentation

3 agents: Mintlify Author, API Documenter, Runbook Writer

Prompts

3 agents: Prompt Refiner, Chain Architect, Fragment Curator

Incident

3 agents: Incident Commander, Triage Responder, Postmortem Analyst

​
Output Schemas

Agents produce schema-validated outputs:
Output TypeSchemaUsed By
n8n Workflown8n-output-schema.jsonWorkflow Architect, Webhook Engineer
Veritas Draftveritas-draft-schema.jsonPrompt Refiner, Chain Architect
OpenAPI SpecOpenAPI 3.0API Documenter
TypeScriptZod schemasEngineering agents

​
Next Steps

Agent Catalog

Explore all 20 agents

FORGE Deep Dive

Learn more about stage gates