Skip to main content
The think tool is a reasoning scratchpad that lets agents think step-by-step before acting. The agent can write out its reasoning without producing visible output to the user — ideal for planning complex tasks, breaking down problems, and reasoning through multi-step solutions. This tool has no side effects and produces no output visible to the user. It only improves the quality of the agent’s reasoning.

Configuration

toolsets:
  - type: think
The think tool has no configuration options.

Example agent

agent.yaml
agents:
  root:
    model: anthropic/claude-sonnet-4-0
    description: Coding assistant
    instruction: |
      You are a coding assistant. Before making any changes,
      use the think tool to plan your approach.
    toolsets:
      - type: think
      - type: filesystem
      - type: shell
Include the think tool in every agent. It adds minimal overhead while significantly improving the quality of reasoning on complex, multi-step tasks.

When to use it

The think tool is most valuable when the agent needs to:
  • Plan a multi-step task before executing it
  • Reason through competing options before choosing one
  • Decompose a complex problem into smaller steps
  • Verify its own reasoning before committing to an action
It is especially effective when combined with tools that have side effects (like shell or filesystem), where planning ahead reduces the chance of mistakes.

Todo

Track task progress across multi-step work.

Background agents

Dispatch parallel work to sub-agents.

Shell

Execute commands after planning.

Filesystem

Read and write files after planning.

Build docs developers (and LLMs) love