This document defines the standard execution model for content creation agents.

Core Constraint

Each tool call is a SEPARATE message. NEVER combine tools in a single message or tool plan.

Step Action Separate Message?
Plan Socratic Dialogue YES - refine goals
Research Call research_content YES - wait for response
Create Call content_create YES - wait for response
Image Generate featured image YES - wait for response

Why This Matters

  1. Context Clarity: Each tool call receives a clean context without competing artifacts
  2. Error Recovery: If research fails, you can retry without affecting creation
  3. Artifact Tracking: Research returns a research_id that must be explicitly passed to creation
  4. Debugging: Easier to identify which step failed when they're separate

Correct Workflow

User: "Create a blog post about AI agents"
Agent: [Socratic questions to clarify goal]
User: [Provides answers]
Agent: "CONFIRMED: [goal summary]. Proceeding with research."
Agent: [Calls research_content] → Returns research_id
Agent: "Research complete. ID: abc123. Now creating content."
Agent: [Calls content_create with research_id: abc123]
Agent: "Content created. Slug: ai-agents-production"

Incorrect Workflow (NEVER DO THIS)

User: "Create a blog post about AI agents"
Agent: "Planning both research and creation..."
Agent: [Calls research_content AND content_create together]

This pattern causes:

  • Missing research data in content creation
  • Unable to use research_id reference
  • Harder to debug failures

Standard Execution Pattern

Step 1: Socratic Dialogue

Agent interrogates your content goal to clarify:

  • Target audience
  • Key points to cover
  • Tone and style
  • Technical depth

Step 2: Goal Confirmation

Agent confirms the plan before proceeding:

"CONFIRMED: Writing a technical blog post about MCP server architecture
for Rust developers. Will cover the rmcp crate, tool implementation,
and deployment patterns. Proceeding with research."

Step 3: Research (Separate Message)

// MCP: content_manager
{
  "name": "research_content",
  "arguments": {
    "topic": "MCP server architecture",
    "content_type": "blog",
    "target_audience": "Rust developers"
  }
}

Wait for response. Extract research_id from result.

Step 4: Create (Separate Message)

// MCP: content_manager
{
  "name": "content_create",
  "arguments": {
    "research_id": "abc123",
    "title": "Building MCP Servers in Rust",
    "content_type": "blog"
  }
}

Step 5: Image (Separate Message, Optional)

// MCP: content_manager
{
  "name": "generate_image",
  "arguments": {
    "slug": "building-mcp-servers",
    "description": "Technical diagram of MCP architecture"
  }
}

Flow Diagram

┌─────────────┐
│   User      │
│   Request   │
└──────┬──────┘
       │
       ▼
┌─────────────────────────────────────────┐
│          Socratic Dialogue              │
│   (Clarify goals, no tool calls)        │
└──────┬──────────────────────────────────┘
       │
       ▼
┌─────────────────────────────────────────┐
│          Goal Confirmation              │
│   "CONFIRMED: [summary]. Proceeding."   │
└──────┬──────────────────────────────────┘
       │
       ▼
┌─────────────────────────────────────────┐
│     Research (SEPARATE MESSAGE)         │
│   Call: research_content                │
│   Returns: research_id                  │
└──────┬──────────────────────────────────┘
       │
       ▼
┌─────────────────────────────────────────┐
│     Create (SEPARATE MESSAGE)           │
│   Call: content_create                  │
│   Input: research_id from above         │
└──────┬──────────────────────────────────┘
       │
       ▼
┌─────────────────────────────────────────┐
│     Image (SEPARATE MESSAGE)            │
│   Call: generate_image                  │
│   (Optional)                            │
└──────┬──────────────────────────────────┘
       │
       ▼
┌─────────────┐
│   Result    │
│   Summary   │
└─────────────┘

Agent Types and Specializations

Agent Specialty Playbook
content Content Marketing Strategist playbook content
blog Blog Content Specialist playbook blog
linkedin LinkedIn Professional Agent playbook linkedin
medium Medium Narrative Agent -
hackernoon HackerNoon Technical Agent -
twitter Twitter/X Agent playbook twitter

Each agent follows the same execution model but with platform-specific adaptations.

Research Skill Rules

CRITICAL: Research skills must only be called ONCE per content piece.

After research is complete:

  1. You receive an artifact_id (research_id)
  2. Use that ID when calling content_create in a SEPARATE message
  3. DO NOT call research again unless explicitly instructed
  4. If you need more information, reference the existing research

Error Handling

If a step fails:

  1. Research fails: Report error, ask user if they want to retry or adjust parameters
  2. Create fails: Report error with research_id, ask if user wants to retry with same research
  3. Image fails: Report error, content is still created, image can be added later

Never continue to the next step if the current step failed.

Quick Reference

Step Tool Input Output
Research research_content topic, audience, type research_id
Create content_create research_id, title, type slug, content
Image generate_image slug, description image_url
Rule Description
One tool per message Never combine research and create
Wait for response Always wait for tool result before next step
Pass research_id Explicitly pass ID from research to create
Confirm before proceeding "CONFIRMED: [summary]. Proceeding with research."