​

Veritas Integration

​

What is Veritas?

• Reusable prompt fragments for common tasks
• Versioned and validated prompts
• Domain-organized content
• Bidirectional integration with agents

Figure 1 — VERITAS Library Structure. The 255-prompt library is partitioned into three categories: task prompts (single-purpose), fragment prompts (composable snippets), and chain prompts (multi-step sequences). All prompts are organised by domain, purpose, and version.

​

Prompt Categories

{
  "id": "vrt-a1b2c3d4",
  "category": "task",
  "domain": "automation",
  "purpose": "n8n workflow design patterns",
  "version": "1.0.0",
  "content": "When designing n8n workflows, follow these patterns:\n1. Use webhook triggers for external events\n2. Include error handling nodes after each action\n3. Add retry logic with exponential backoff..."
}

{
  "id": "vrt-e5f6g7h8",
  "category": "fragment",
  "domain": "engineering",
  "purpose": "Zod validation boilerplate",
  "version": "1.0.0",
  "content": "Always validate inputs with Zod:\n- Define schema before handler\n- Use .parse() for throwing validation\n- Use .safeParse() when you need error details\n- Export schemas for reuse"
}

{
  "id": "vrt-chain001",
  "category": "chain",
  "domain": "prompts",
  "purpose": "Design-implement-test workflow",
  "version": "1.0.0",
  "steps": [
    {"step": 1, "action": "Analyze requirements", "output": "spec"},
    {"step": 2, "action": "Design solution", "input": "spec", "output": "design"},
    {"step": 3, "action": "Implement code", "input": "design", "output": "code"},
    {"step": 4, "action": "Write tests", "input": "code", "output": "tests"}
  ]
}

​

Agent Configuration

veritas:
  consumes:
    - id: "vrt-a1b2c3d4"
      purpose: "Workflow design patterns"
    - id: "vrt-e5f6g7h8"
      purpose: "Error handling best practices"
  produces:
    - category: task
      output_path: "veritas/agent-prompts/automation/"

​

Consuming Prompts

​

Producing Prompts

​

Prompt Flow

Figure 2 — VERITAS Prompt Lifecycle. Prompts flow from the VERITAS core to consuming agents, who produce draft prompts from novel patterns. Drafts undergo human review; approved drafts are promoted back into VERITAS core for reuse.

Trace ID	Requirement	Rationale	Test Case
SO1-VRT-003	Draft prompts shall include source-agent metadata and discovery context	Enables reviewers to assess relevance and accuracy	TC-SO1-VRT-003
SO1-VRT-004	Promoted prompts shall receive a new versioned VERITAS ID (vrt-xxxxxxxx)	Maintains catalogue integrity and supports rollback	TC-SO1-VRT-004

​

Prompt ID Format

vrt-xxxxxxxx

Where:
- vrt- : Veritas prefix
- xxxxxxxx : 8-character alphanumeric identifier

• vrt-a1b2c3d4 - Workflow patterns
• vrt-wh001xyz - Webhook security
• vrt-chain001 - Design chain

​

Prompts by Domain

​

Draft Prompt Schema

{
  "$schema": "veritas-draft-schema.json",
  "metadata": {
    "source_agent": "workflow-architect",
    "discovered_at": "2024-01-15T10:30:00Z",
    "context": "Learned while designing retry patterns"
  },
  "prompt": {
    "category": "fragment",
    "domain": "automation",
    "purpose": "Exponential backoff configuration",
    "content": "For retry logic, use exponential backoff:\n- Initial delay: 1 second\n- Multiplier: 2x\n- Max retries: 5\n- Max delay: 60 seconds"
  },
  "status": "draft"
}

​

Review Process

1. Agent produces draft → saved to veritas/agent-prompts/
2. Human reviews → checks accuracy, relevance, quality
3. Approval → promoted to Veritas core with new ID
4. Rejection → feedback provided, draft deleted or revised

​

Validation

./scripts/validate-agents.sh

# Checks:
# ✓ veritas.consumes defined
# ✓ Prompt IDs match format (vrt-xxxxxxxx)
# ✓ veritas.produces has valid category
# ✓ output_path exists

​

Related Documentation