Task Management UI

Overview

The dashboard's task management interface gives you a visual way to create, track, and manage tasks that your agent works on. Tasks are stored in a shared SQLite database (~/.argentos/data/dashboard.db) that both the dashboard and the agent's tools access.

Task List View

The task list shows all tasks with:

• Title: What the task is about
• Status: pending, in_progress, completed, failed, blocked
• Priority: urgent, high, normal, low, background
• Source: Who created it (user, agent, heartbeat, schedule)
• Created/Updated: Timestamps

Creating Tasks

From the Dashboard

Click the + New Task button and fill in:

• Title (required)
• Description (optional)
• Priority (defaults to normal)

From Chat

Tell the agent to create a task:

Create a task to review the Q4 security audit

The agent uses the tasks_add tool and the task appears in the dashboard immediately.

Via Slash Command

/task Review the Q4 security audit

The Play/Complete Flow

This is the core task workflow:

1. Start a Task

Click the play button on a task. This:

• Calls POST /api/tasks/:id/start to update the status
• Sends a chat message to the gateway telling the agent to work on this task
• The task status changes to in_progress

2. Agent Works

The agent receives the message and begins working. It may:

• Use tools (exec, browser, memory) to complete the task
• Stream progress updates back through the chat
• Call tasks_complete when finished

3. Completion

When the agent finishes, it:

• Calls the tasks tool with action: complete
• Emits a [TASK_DONE:title] marker in the response stream

The dashboard detects the marker for instant UI updates and also polls the database every 5 seconds as a backup.

Task Status Flow

pending → in_progress → completed
                      → failed
                      → blocked → in_progress → completed

Real-Time Updates

The dashboard updates task status through two mechanisms:

1. Stream markers: [TASK_DONE:title] parsed from the agent's response stream for instant feedback
2. Database polling: Every 5 seconds, the dashboard polls the database to catch any changes made directly by the agent

This dual approach ensures the UI stays current even if a marker is missed.