Task Tracking

Overview

When Claude Code uses the TaskCreate and TaskUpdate tools during a session, Abbado captures these events and reconstructs a task list for each run. This gives you a structured view of what the agent planned to do and how far it got.

How It Works

Abbado does not implement its own task system. Instead, it reconstructs tasks from hook events:

1. The agent calls TaskCreate with a subject, description, and status
2. The PreToolUse hook fires — Abbado persists the tool call as an agent.tool_call event
3. The PostToolUse hook fires — Abbado persists the tool result as an agent.tool_result event (containing the task ID)
4. When you request the task list, Abbado replays all events and pairs TaskCreate calls with their results

For TaskUpdate events, the same process applies — Abbado updates the task’s status, subject, or description based on the tool input.

Task States

Tasks go through these states (as set by the agent):

Status	Meaning
pending	Task created but not started
in_progress	Agent is currently working on this task
completed	Task finished successfully
deleted	Task removed (filtered from display)

API

Get Tasks for a Run

GET /api/runs/{id}/tasks

Returns a list of tasks reconstructed from events:

[
  {
    "id": "task-1",
    "subject": "Add input validation to login form",
    "description": "Validate email format and password length",
    "status": "completed"
  },
  {
    "id": "task-2",
    "subject": "Write unit tests for validation",
    "status": "in_progress"
  }
]

Tasks with status deleted are filtered out of the response.

Event Reconstruction

The task list is derived from events, not stored separately. This means:

• No separate task table — tasks come from agent.tool_call and agent.tool_result events
• Consistent with event history — tasks always reflect what actually happened
• No sync issues — the task list is computed on read, not maintained in parallel

The reconstruction algorithm:

1. Iterate all events for the run in order
2. For agent.tool_call events with tool TaskCreate, store the input (subject, description, status)
3. For agent.tool_result events with tool TaskCreate, extract the task ID from the response and create the task entry
4. For agent.tool_call events with tool TaskUpdate, update the matching task’s fields

Note

Task tracking only works with Claude Code, since it’s the CLI that provides the TaskCreate and TaskUpdate tools. Codex sessions do not generate task events.

Dashboard Display

In the frontend, the task list appears in the run panel when tasks are present. Each task shows:

• A status icon (pending, in progress, completed)
• The task subject
• The task description (if provided)

This gives you a quick overview of the agent’s plan and progress without reading through the full event history.