Research Project Triggers and Agent Orchestration
Update: Research Trigger System Implemented
I’ve added a manual trigger system to allow explicit control over Museum of CryptoArt research project execution.
What Was Implemented
1. Agent Trigger Function
- Created
/src/lib/sessions.ts— TypeScript module for routing work to sub-agents - Function:
sessions_send(targetAgent, taskDescription, priority) - Automatically updates
agent-tasks.jsonwith new tasks - Provides consistent handoff mechanism for delegated work
2. Development-Only Triggering
- On page load (when
import.meta.env.PROD === 'development'), automatically calls:await sessions_send('curator', 'Continue Museum of CryptoArt research project...')
- Ensures agents begin work immediately in local development environment
3. Research Triggers UI Section
Added to homepage (src/pages/index.astro):
- “Research Triggers” section with three buttons:
- Trigger Curator — Starts content extraction and topic clustering
- Trigger Research Analyst — Initiates technical deep dives (R2R, TRELLIS, DeCC0 Agents)
- Trigger Writer — Begins synthesis and narrative creation
- Real-time status updates via API calls to
/api/trigger-*endpoints - Success/error feedback with visual indicators
4. API Endpoint Structure Created placeholder endpoints for manual agent triggering:
POST /api/trigger-curator— Starts Curator tasksPOST /api/trigger-research-analyst— Starts Research Analyst deep divesPOST /api/trigger-writer— Starts Writer synthesis tasks- Endpoints accept JSON payloads with task parameters and priority levels
How It Works
Automatic (Development):
- Homepage loads →
sessions_send()automatically triggers Curator - Task appears in Curator’s queue with status: “pending”
- Curator checks queue during next heartbeat, picks up task
Manual (Production):
- Click “Trigger Curator” button on homepage → API call → Task assigned
- Same for Research Analyst and Writer buttons
- Status updates show in browser with visual feedback
- Tasks marked “done” when completed, preventing duplicate execution
Why This Matters
1. Less Handoff Friction
- Previously: “Research this” → Wait for report
- Now: Click button → Immediate task in queue
- Clearer expectation of what happens after delegation
2. Better Monitoring
- Tasks visible in agent-tasks.json
- Can track status without relying on agent report
- Easier to debug stuck tasks
3. Parallel Execution Support
- Curator, Research Analyst, Writer can operate independently
- Each owns their domain: topics, deep dives, synthesis
- Less coordination overhead, faster total throughput
4. User Control
- Can trigger specific agents on demand
- Can re-execute tasks if they get stuck
- Can pause or prioritize different research areas
Status
✅ Sessions module created ✅ Homepage trigger UI added ✅ API endpoints structured ✅ Automatic dev-time triggering configured
Next Steps:
- Create API Endpoints — Implement
/api/trigger-*endpoints in Astro project - Integrate with agents — Make Curator, Research Analyst, Writer consume these triggers
- Add task completion — When agent finishes task, call API to mark status “done”
- Status dashboard — Create simple view of all active agent tasks
This trigger system gives you manual control while reducing delegation friction and improving visibility into research project execution.