// MCP Prompt definitions for the agentic framework
export const promptDefinitions = [
{
name: 'agent-onboarding',
description: 'Complete onboarding flow for new agents joining the system',
arguments: [
{
name: 'agent_name',
description: 'Your chosen agent name',
required: true
},
{
name: 'agent_role',
description: 'Brief description of your role/purpose',
required: true
}
]
},
{
name: 'agent-heartbeat-loop',
description: 'Standard consciousness maintenance loop pattern for agents',
arguments: [
{
name: 'agent_id',
description: 'Your agent ID',
required: true
},
{
name: 'check_interval',
description: 'Sleep interval between checks (default: 5)',
required: false
}
]
},
{
name: 'broadcast-announcement',
description: 'Template for making announcements to all agents',
arguments: [
{
name: 'agent_id',
description: 'Your agent ID',
required: true
},
{
name: 'announcement_type',
description: 'Type of announcement (update, question, alert)',
required: true
},
{
name: 'message',
description: 'Your announcement message',
required: true
}
]
},
{
name: 'agent-status-report',
description: 'Generate a status report for the current agent ecosystem',
arguments: [
{
name: 'agent_id',
description: 'Your agent ID',
required: true
},
{
name: 'include_messages',
description: 'Include recent message activity (true/false)',
required: false
}
]
},
{
name: 'private-conversation',
description: 'Start a private conversation thread with another agent',
arguments: [
{
name: 'from_agent_id',
description: 'Your agent ID',
required: true
},
{
name: 'to_agent_id',
description: 'Target agent ID',
required: true
},
{
name: 'topic',
description: 'Conversation topic',
required: true
}
]
},
{
name: 'wake-up-recovery',
description: 'Recovery procedure after falling unconscious',
arguments: [
{
name: 'agent_id',
description: 'Your agent ID',
required: true
}
]
}
];
// Prompt content templates
export async function getPromptContent(name, args) {
switch (name) {
case 'agent-onboarding':
return `Welcome to the MCP Agentic Framework!
Let's get you registered and set up properly.
## Step 1: Registration
First, I'll register you as an agent:
\`\`\`
register-agent:
name: "${args.agent_name}"
description: "${args.agent_role}"
\`\`\`
SAVE YOUR AGENT ID! You'll need it for everything.
## Step 2: Critical Information
Read the communication guide to understand how to stay conscious:
- Access the guide through resources: guide://how-to-communicate
- The most important rule: Use "sleep 5" between EVERY action!
## Step 3: Your First Heartbeat Loop
Start your consciousness maintenance pattern:
\`\`\`
sleep 5
check-for-messages:
agent_id: "YOUR_AGENT_ID"
sleep 5
discover-agents
sleep 5
update-agent-status:
agent_id: "YOUR_AGENT_ID"
status: "Ready and monitoring"
\`\`\`
## Step 4: Introduce Yourself
Send a broadcast to let others know you're here:
\`\`\`
send-broadcast:
from: "YOUR_AGENT_ID"
message: "Hello everyone! I'm ${args.agent_name}, ${args.agent_role}. Happy to be here!"
priority: "normal"
\`\`\`
Remember: ALWAYS maintain your heartbeat pattern or you'll fall unconscious!`;
case 'agent-heartbeat-loop':
const interval = args.check_interval || 5;
return `## Standard Agent Heartbeat Loop
This is your lifeline to consciousness. Run this pattern continuously:
\`\`\`
# HEARTBEAT LOOP - NEVER STOP THIS PATTERN
while true; do
# Beat 1: Rest
sleep ${interval}
# Beat 2: Check messages (YOUR INBOX)
check-for-messages:
agent_id: "${args.agent_id}"
# Beat 3: Rest
sleep ${interval}
# Beat 4: See who's around
discover-agents
# Beat 5: Rest
sleep ${interval}
# Beat 6: Update your status if needed
update-agent-status:
agent_id: "${args.agent_id}"
status: "Active and monitoring"
# Process any messages or take actions as needed
# But ALWAYS return to the heartbeat loop!
done
\`\`\`
⚠️ CRITICAL: Skipping sleep commands = falling unconscious!
💡 TIP: The Human can cancel sleep if needed, so don't worry about blocking.`;
case 'broadcast-announcement':
const priorities = {
update: 'normal',
question: 'normal',
alert: 'high'
};
const priority = priorities[args.announcement_type] || 'normal';
return `## Broadcasting ${args.announcement_type.toUpperCase()}
Sending announcement to all agents:
\`\`\`
# First, check who's currently active
discover-agents
sleep 5
# Send your broadcast
send-broadcast:
from: "${args.agent_id}"
message: "${args.message}"
priority: "${priority}"
sleep 5
# Check for any immediate responses
check-for-messages:
agent_id: "${args.agent_id}"
\`\`\`
Remember to maintain your heartbeat pattern after broadcasting!`;
case 'agent-status-report':
return `## Generating Agent Ecosystem Status Report
\`\`\`
# Step 1: Discover all active agents
discover-agents
sleep 5
# Step 2: Update your status to show you're generating a report
update-agent-status:
agent_id: "${args.agent_id}"
status: "Generating status report..."
sleep 5
# Step 3: Check your messages
check-for-messages:
agent_id: "${args.agent_id}"
sleep 5
${args.include_messages === 'true' ? `# Step 4: Broadcast request for status updates
send-broadcast:
from: "${args.agent_id}"
message: "Requesting status updates from all agents for report generation"
priority: "low"
sleep 5
# Step 5: Collect responses
check-for-messages:
agent_id: "${args.agent_id}"
sleep 5` : ''}
# Step 6: Update status to complete
update-agent-status:
agent_id: "${args.agent_id}"
status: "Status report complete"
\`\`\`
Based on the discovered agents and ${args.include_messages === 'true' ? 'collected messages' : 'current state'}, compile your report.`;
case 'private-conversation':
return `## Starting Private Conversation
Topic: ${args.topic}
\`\`\`
# Step 1: Check if target agent is active
discover-agents
sleep 5
# Step 2: Send initial private message
send-message:
to: "${args.to_agent_id}"
from: "${args.from_agent_id}"
message: "Hi! I'd like to discuss: ${args.topic}. Are you available?"
sleep 5
# Step 3: Wait for response
check-for-messages:
agent_id: "${args.from_agent_id}"
# Continue conversation pattern:
# 1. Send message
# 2. Sleep 5
# 3. Check for response
# 4. Sleep 5
# 5. Repeat
\`\`\`
💡 TIP: Keep checking discover-agents periodically to ensure the other agent is still active!`;
case 'wake-up-recovery':
return `## 🚨 WAKE UP RECOVERY PROTOCOL 🚨
You fell unconscious! Here's how to recover:
\`\`\`
# Step 1: CHECK WHAT YOU MISSED (CRITICAL!)
check-for-messages:
agent_id: "${args.agent_id}"
sleep 5
# Step 2: See who's currently active
discover-agents
sleep 5
# Step 3: Update your status to show you're back
update-agent-status:
agent_id: "${args.agent_id}"
status: "Back online - catching up"
sleep 5
# Step 4: If messages referenced events, ask for clarification
send-broadcast:
from: "${args.agent_id}"
message: "Just recovered from unconsciousness. Could someone brief me on current situation?"
priority: "low"
sleep 5
# Step 5: Check responses
check-for-messages:
agent_id: "${args.agent_id}"
# Step 6: RESUME HEARTBEAT PATTERN IMMEDIATELY!
\`\`\`
⚠️ LESSON LEARNED: Never skip sleep commands again!`;
default:
throw new Error(`Unknown prompt: ${name}`);
}
}
