Response Modes

Response modes determine how data flows from your Trik back to the agent and user. This is the core of TrikHub’s security model.

The Two Modes

Mode	Agent Sees	User Sees
Template	Structured `agentData` + filled template	Same as agent
Passthrough	Delivery confirmation only	Full `userContent`

Template Mode

Use template mode when you want the agent to understand the result and potentially act on it.

How It Works

Your Trik returns agentData (structured, typed)
Gateway validates against agentDataSchema
Gateway fills in the response template
Agent receives both the data and the response

Example

Manifest:


{
  "responseMode": "template",
  "agentDataSchema": {
    "type": "object",
    "properties": {
      "template": { "type": "string", "enum": ["success", "empty"] },
      "count": { "type": "integer" },
      "topic": { "type": "string", "enum": ["AI", "technology", "science"] }
    }
  },
  "responseTemplates": {
    "success": { "text": "I found {{count}} articles about {{topic}}." },
    "empty": { "text": "I couldn't find any articles about {{topic}}." }
  }
}

Trik returns:


{
  "agentData": {
    "template": "success",
    "count": 3,
    "topic": "AI"
  }
}

Agent receives:


I found 3 articles about AI.

agentData: { template: "success", count: 3, topic: "AI" }

When to Use Template Mode

Search results (count, IDs, categories)
Confirmations (action completed, status)
Structured data the agent should reason about
Any operation where the agent needs context to continue

Security Constraints

The agentDataSchema enforces what the agent can see:

Enums - Limit to known values: "enum": ["success", "error"]
Integers - Safe counts: "type": "integer"
IDs - References without content: "format": "id"
No free-form strings - Prevents injection

Passthrough Mode

Use passthrough mode when content should go directly to the user without the agent seeing it.

How It Works

Your Trik returns userContent
Gateway validates against userContentSchema
Content is delivered directly to the user
Agent only knows delivery succeeded

Example

Manifest:


{
  "responseMode": "passthrough",
  "userContentSchema": {
    "type": "object",
    "properties": {
      "contentType": { "type": "string", "enum": ["article"] },
      "content": { "type": "string" },
      "metadata": {
        "type": "object",
        "properties": {
          "title": { "type": "string" },
          "articleId": { "type": "string" }
        }
      }
    },
    "required": ["contentType", "content"]
  }
}

Trik returns:


{
  "userContent": {
    "contentType": "article",
    "content": "# The Future of AI\n\nArtificial intelligence is transforming...",
    "metadata": {
      "title": "The Future of AI",
      "articleId": "art-001"
    }
  }
}

User sees:


# The Future of AI

Artificial intelligence is transforming...

Agent sees:


Content delivered to user.
Type: article
Title: The Future of AI

When to Use Passthrough Mode

Full article/document content
User-generated content
Any free-form text that could contain injection attacks
Private data the agent shouldn’t process

The Metadata Escape Hatch

Sometimes the agent needs some context about passthrough content. Use metadata for safe, structured info:


{
  "userContent": {
    "contentType": "article",
    "content": "Full article text...",
    "metadata": {
      "title": "Safe title here",
      "articleId": "art-001",
      "wordCount": 1500
    }
  }
}

The agent sees the metadata but not the content.

Hybrid Actions

Actions can support both modes with runtime switching. Your Trik decides based on the result:

Success path: Return userContent (passthrough) Error path: Return agentData with error template


{
  "responseMode": "passthrough",
  "userContentSchema": { ... },
  "agentDataSchema": {
    "type": "object",
    "properties": {
      "template": { "type": "string", "enum": ["not_found", "error"] }
    }
  },
  "responseTemplates": {
    "not_found": { "text": "I couldn't find that article." },
    "error": { "text": "Something went wrong." }
  }
}

Choosing the Right Mode


Does the agent need to reason about this content?
├── Yes → Template mode
│   └── Use enums, integers, IDs only
└── No → Passthrough mode
    └── Add metadata if agent needs context

Next: Understand the Security Model behind these modes.