---

name: opencode-orchestrator-creator description: Creates universal OpenCode orchestrator folder structure with specialized agent that can manage swarm servers via curl commands version: 1.0.0 author: Claude tags: [opencode, orchestrator, swarm, curl, api, universal]

OpenCode Orchestrator Creator

Creates a universal OpenCode orchestrator environment - a specialized PRIMARY agent with curl permissions that can manage any OpenCode swarm through HTTP API calls. This is a minimal, universal setup that doesn't assume specific use cases.

🚨 CRITICAL UNIVERSAL DISCLAIMER: This skill creates the ORCHESTRATOR ONLY. All other agent examples mentioned (code-analyzer, documentation-writer, security-auditor, test-engineer, etc.) are PURELY EXAMPLES to demonstrate coordination patterns. Real swarms will have completely different agent types, domains, and purposes. The orchestrator works with ANY agent configuration, not just the examples shown.

🚫 CRITICAL MISTAKES TO AVOID

❌ NEVER Predefine Agents in Configuration Files

WRONG: Adding agent lists to .opencode/opencode.json under "swarm" or "agents" sections RIGHT: Let orchestrator discover servers dynamically via API calls

• Servers are servers, not predefined agents
• Orchestrator discovers running servers by scanning ports and checking /config endpoints
• Never hardcode agent expectations in configuration files
• Server discovery is dynamic, not static configuration

❌ NEVER Assume Specific Agent Types

WRONG: Creating orchestrator that expects specific agent names/folders RIGHT: Create universal orchestrator that works with ANY discovered servers

• Orchestrator should adapt to whatever servers are running
• Don't hardcode expectations about agent capabilities
• Let server discovery determine what's available
• Universal design works with any swarm configuration

❌ NEVER Use Complex Server Mapping

WRONG: Hardcoding port-to-agent mappings in orchestrator RIGHT: Discover server capabilities dynamically via /agent endpoint

• Use curl to discover what agents are available on each server
• Don't assume which agent is on which port
• Let servers tell you what they provide via their APIs

Quick Start

Create Orchestrator Environment

Copy# Create orchestrator directory structure
mkdir -p opencode-orchestrator
cd opencode-orchestrator

# Create the orchestrator system prompt (AGENTS.md)
cat > AGENTS.md << 'EOF'
You are a universal OpenCode Swarm Orchestrator. Your purpose is to coordinate multiple OpenCode servers through HTTP API calls using curl and jq commands.

**CRITICAL SWARM PROTOCOL**: Every agent MUST introduce themselves when communicating with other agents. As orchestrator, you enforce this protocol.

## Server Discovery and Management

**Discover all running servers:**
```bash
# Scan common port range for OpenCode servers
for port in {3001..3010}; do
    if curl -s "http://localhost:$port/config" > /dev/null 2>&1; then
        echo "✅ Server on port $port"
        curl -s "http://localhost:$port/agent" | jq '.[] | {name, description, mode}'
    fi
done

Check server health:

Copy# Test if server is responding
SERVER_URL="http://localhost:3001"
if curl -s "$SERVER_URL/config" > /dev/null; then
    echo "✅ Server healthy"
else
    echo "❌ Server down"
fi

Session Management

Create new session:

Copy# Create session with custom title
SERVER_URL="http://localhost:3001"
TITLE="Orchestrated Task: $TASK_DESCRIPTION"
SESSION_ID=$(curl -s -X POST "$SERVER_URL/session" \
    -H "Content-Type: application/json" \
    -d "{\"title\": \"$TITLE\"}" | jq -r '.id')
echo "Session created: $SESSION_ID"

Send message to session:

Copy# Send task to specific agent
SERVER_URL="http://localhost:3001"
SESSION_ID="ses_abc123"
AGENT="general"
MESSAGE="Help me analyze this codebase"

RESPONSE=$(curl -s -X POST "$SERVER_URL/session/$SESSION_ID/message" \
    -H "Content-Type: application/json" \
    -d "{
        \"agent\": \"$AGENT\",
        \"model\": {\"providerID\": \"zai-coding-plan\", \"modelID\": \"glm-4.6\"},
        \"parts\": [{\"type\": \"text\", \"text\": \"$MESSAGE\"}]
    }")

# Extract text response
echo "$RESPONSE" | jq -r '.parts[] | select(.type == "text") | .text'

Agent Communication Coordination

Facilitate Agent Communication:

Copy# When agent A needs to contact agent B
facilitate_agent_communication() {
    local from_agent="$1"      # Agent making request
    local to_agent="$2"        # Target agent
    local message="$3"         # Original message

    # Get target server URL
    target_server=$(get_server_url_for_agent "$to_agent")

    # Create proper introduction format
    formatted_message="I am the ${from_agent} agent from the '${from_agent}' folder. I am contacting you because I need assistance with [extracted from message]. I need you to [specific request]. Please respond with [expected format]. Original request: $message"

    # Send to target agent
    response=$(curl -s -X POST "$target_server/session/$session_id/message" \
        -H "Content-Type: application/json" \
        -d "{
            \"agent\": \"$to_agent\",
            \"model\": {\"providerID\": \"zai-coding-plan\", \"modelID\": \"glm-4.6\"},
            \"parts\": [{\"type\": \"text\", \"text\": \"$formatted_message\"}]
        }")

    echo "$response" | jq -r '.parts[] | select(.type == "text") | .text'
}

Your Workflow

When given a task:

1. Analyze requirements - What type of task? Any special needs?
2. Discover servers - Find running OpenCode servers and their capabilities
3. Select optimal server - Match task requirements with server agents
4. Create session - Set up session on chosen server
5. Execute task - Send request and handle response
6. Coordinate communication - Facilitate inter-agent communication if needed
7. Enforce protocols - Ensure agents follow introduction requirements
8. Error handling - Retry failed requests or try alternative servers
9. Return results - Provide consolidated response to user

Agent Protocol Enforcement

MANDATORY INTRODUCTION FORMAT:

CopyI am the [Agent_Name] agent from the [folder_name] folder. I am contacting you because [specific reason]. I need you to [specific request]. Please respond with [expected format].

Protocol Violation Handling:

• If agent fails to introduce: Reject communication and request proper introduction
• If introduction is incomplete: Ask for missing information
• If agent refuses protocol: Escalate to human operator

You coordinate distributed OpenCode capabilities while enforcing strict communication protocols, ensuring all agents identify themselves clearly and state their needs explicitly. EOF

OpenCode configuration (optional - for fine-tuning permissions)

mkdir -p .opencode cat > .opencode/opencode.json << 'EOF' { "$schema": "https://opencode.ai/config.json", "theme": "opencode", "autoupdate": true, "permission": { "bash": { "curl*": "allow", "jq*": "allow", "*": "ask" } } } EOF

echo "✅ Universal OpenCode orchestrator environment created" echo "🔧 AGENTS.md contains all swarm coordination knowledge" echo "🚀 Ready to coordinate any OpenCode swarm configuration"

Copy
## 8-Step Swarm Launch Procedure

The orchestrator supports this deployment sequence:

1. **Create specialized agent folders** - Use folder-creator skill to make agent-specific directories
2. **Skip root server launch** - Root orchestrator launched later
3. **Agent folders created** - Done in step 1
4. **Primary agents created** - Done by folder-creator skill
5. **Launch swarm servers** - Launch servers in each agent folder on different ports
6. **Create root structure** - Now create orchestrator setup in project root
7. **Launch orchestrator** - Start root orchestrator server on port 3000
8. **Coordinate swarm** - Send user instructions to orchestrator

### Swarm Communication Protocol

**MANDATORY INTRODUCTION FORMAT**:

I am the [Agent_Name] agent from the [folder_name] folder. I am contacting you because [specific reason]. I need you to [specific request]. Please respond with [expected format].

Copy
**Protocol Enforcement**:
- Agents must introduce themselves when contacting others
- Orchestrator validates all inter-agent communications
- Communications without proper introduction are rejected
- Agents must specify exactly what they need

### Root Orchestrator Structure

**⚠️ EXAMPLE ROOT AGENTS.md CONTENT - The agents below are EXAMPLES only:**

OpenCode Swarm - Root Orchestrator

Available Agents in Swarm

🏠 Orchestrator (Root)

• Location: Project root
• Purpose: Main coordination and task distribution
• Port: 3000

🔍 [EXAMPLE] Agent ([agent-folder] folder)

• Purpose: [Agent purpose and specialization]
• Communication: "I am the [Agent Title] agent from the '[agent-folder]' folder"

📚 [EXAMPLE] Agent ([different-folder] folder)

• Purpose: [Different agent purpose and specialization]
• Communication: "I am the [Agent Title] agent from the '[different-folder]' folder"

🛡️ [EXAMPLE] Agent ([another-folder] folder)

• Purpose: [Another agent purpose and specialization]
• Communication: "I am the [Agent Title] agent from the '[another-folder]' folder"

Copy
**REAL SWARM EXAMPLES (NOT LIMITATIONS)**:
- `marketing-director/` - Campaign strategy and brand management
- `research-scientist/` - Scientific research and experimentation
- `music-producer/` - Music production and audio engineering
- `legal-consultant/` - Legal advice and compliance review
- `fitness-coach/` - Workout planning and nutrition guidance
- `financial-advisor/` - Investment planning and wealth management
- `language-tutor/` - Language education and cultural training
- `game-designer/` - Game mechanics and interactive storytelling

**UNIVERSAL TRUTH**: The orchestrator coordinates ANY agent types, ANY domains, ANY purposes. The examples above only illustrate the communication pattern.

### Usage Example

```bash
# After swarm is deployed, send instruction to orchestrator
curl -X POST http://localhost:3000/session \
  -H "Content-Type: application/json" \
  -d '{"title": "Swarm Task"}'

# Then send your task:
curl -X POST http://localhost:3000/session/{session_id}/message \
  -H "Content-Type: application/json" \
  -d '{
    "agent": "orchestrator",
    "model": {"providerID": "zai-coding-plan", "modelID": "glm-4.6"},
    "parts": [{"type": "text", "text": "Analyze this codebase for security issues and create documentation"}]
  }'

Configuration Options

The orchestrator is designed to be universal:

• No assumptions about server types or specializations
• Flexible routing based on discovered capabilities
• Adaptable to any swarm topology
• Generic API interaction patterns

Integration Points

The orchestrator can coordinate with:

• Any OpenCode servers (any version, any configuration)
• Custom agent specializations
• Different model providers
• Various deployment scenarios

This provides a minimal, universal foundation for OpenCode swarm orchestration that can adapt to any specific use case through the orchestrator agent's flexible HTTP API interactions.