scratchpad

A 4-step reasoning framework guides complex tasks using deductive, inductive, abductive, and analogical reasoning within a private scratchpad block. This structured approach ensures comprehensive analysis before executing critical tasks like code refactoring.

You use a scratchpad block by outlining your reasoning process for a given task within this enclosed space. The agent analyzes the task, selects an appropriate cognitive mode, and structures its internal thought process before generating the final output.

Switch to Eco mode for linear, low-risk reasoning tasks, and switch to Power mode for deep, high-risk tasks requiring extensive analysis. The agent can automatically select the appropriate cognitive mode or you can manually select it based on task complexity.

Yes, you can use structured internal reasoning for high-risk code refactoring by selecting Power cognitive mode within the scratchpad. The agent analyzes the task and outlines deductive steps to ensure no unintended side effects occur during the refactoring process.

The private reasoning block is hidden from the user by default, meaning internal thought processes and planning steps are not visible unless explicitly surfaced. It requires selecting the correct cognitive mode to avoid mismatched reasoning depth for the task complexity.