Skip to main content

​
Overview

The todowrite tool provides structured task list management within a conversation. It allows the LLM to plan, track, and report progress on multi-step work.

​
Parameters

The tool accepts an array of task items:
FieldTypeRequiredDescription
contentstringYesThe task description
statusenumNopending, in_progress, completed, or cancelled
priorityenumNohigh, medium, or low

​
Task Statuses

StatusMeaning
pendingNot yet started
in_progressCurrently being worked on
completedFinished successfully
cancelledSkipped or blocked (with reason)

​
How It Works

The task list persists across the conversation. Each call to todowrite replaces the full list, so the current state must always be passed. Tasks are rendered grouped by priority with status indicators.

​
Workflow

The recommended workflow for multi-step tasks:
1

Plan

Create a task list with concrete, actionable items. Each task should be specific (e.g. “Add X field to Y struct in Z.rs”), not vague (e.g. “Improve error handling”).
2

Execute

Work through tasks sequentially. Mark each task in_progress before starting and completed when done.
3

Track

If new work is discovered during execution, add it to the list. If a task is blocked, mark it cancelled with a reason.
4

Verify

After all tasks are done, review the work against the original request.

​
Example

A typical task list during execution: 
[
  {
    "content": "Add error variant for network timeouts in error.rs",
    "status": "completed",
    "priority": "high"
  },
  {
    "content": "Update provider to map timeout errors",
    "status": "in_progress",
    "priority": "high"
  },
  {
    "content": "Add retry logic with exponential backoff",
    "status": "pending",
    "priority": "medium"
  },
  {
    "content": "Write tests for timeout handling",
    "status": "pending",
    "priority": "medium"
  }
]

​
Best Practices

  • Keep lists short. 3-7 tasks is typical. Break larger work into phases.
  • Be specific. Each task should describe a concrete action, not a goal.
  • Update immediately. Mark tasks in_progress/completed as you work, not in batches.
  • Add discovered work. If you find new tasks during execution, add them.
  • Never stop early. Continue until all tasks are completed or cancelled.