​

Scheduled Tasks

​

Use Cases

• “Post no LinkedIn sexta 10h” — schedule a skill to run at a specific time
• “Roda o financial pulse amanha 8h” — run an existing script at a future time
• “Envia resumo pro Telegram as 14h” — send a prompt to an agent later

​

Task Types

​

Creating Tasks

​

Via CLI (skill)

agendar post no LinkedIn pra sexta 10h sobre o Summit

​

Via Dashboard

1. Click ”+ New Task”
2. Fill in: name, type, payload, agent, date/time
3. Click Schedule

​

Via API

POST /api/tasks
{
  "name": "Post LinkedIn -- Summit",
  "type": "skill",
  "payload": "social-post-writer LinkedIn post about Evolution Summit",
  "agent": "pixel-social-media",
  "scheduled_at": "2026-04-11T16:00:00Z"
}

​

Task Lifecycle

pending --> running --> completed
                   --> failed (can retry)
pending --> cancelled

• pending — waiting for scheduled time
• running — currently executing
• completed — finished successfully
• failed — error occurred (can be retried with “Run Now”)
• cancelled — manually cancelled before execution

​

How Execution Works

1. The scheduler loop (every 30 seconds) queries: WHERE status='pending' AND scheduled_at <= NOW()
2. Each matching task is executed in a background thread
3. For skill and prompt types, the task uses ADWs/runner.py (same infrastructure as routines)
4. For script type, the Python script runs as a subprocess
5. Results are saved in result_summary; errors in error
6. Status updates to completed or failed

​

API Reference

• status — filter by status (pending, running, completed, failed, cancelled)
• page — page number (default: 1)
• per_page — items per page (default: 50)

​

Dashboard Page

• Filter tabs — All, Pending, Running, Completed, Failed
• Task table — name, type badge, agent badge, scheduled time, status, actions
• Create modal — form with name, description, type, payload, agent, datetime
• View result modal — shows output/error for completed/failed tasks

​

Actions by Status

​

Agents

​

Tasks vs Routines