​
Overview

Plan mode is a read-only planning phase for complex tasks. In plan mode, the agent explores the codebase, considers approaches, and writes a concrete implementation plan before any files are edited. The /plan slash command toggles plan mode from inside the Gizzi Code TUI, or displays the current session plan when already in plan mode. Use plan mode when you want to review a strategy before the agent starts writing code, running tests, or making destructive changes.

​
Command reference

​
Basic usage

Enter plan mode from the Gizzi Code TUI:
Enter plan mode with a specific task:
View the plan that the agent has written so far:
Edit the plan in your default external editor:

​
How plan mode works

When you enter plan mode, the agent switches to a read-only stance:
  • It explores the codebase and existing patterns.
  • It asks clarifying questions if the approach is ambiguous.
  • It writes a design and implementation plan to the session plan file.
  • It does not create, edit, or delete files except the plan file itself.
When the plan is ready, the agent exits plan mode and presents the plan for your approval. After approval, the agent implements the plan using the normal permission rules.

​
Plan file

Each session in plan mode writes its plan to a Markdown file under the Gizzi Code configuration directory:
For subagent sessions, the file name includes the agent ID:
You can override the plans directory with the plansDirectory setting, but it must remain within the project root for safety. Plans are persisted across session resumes and forks.

​
Starting in plan mode from the CLI

You can also start a session directly in plan mode without using the TUI slash command:
In these cases the agent behaves the same way as if /plan had been invoked: it explores, designs, and writes a plan before making any changes.

​
Exiting plan mode

Plan mode ends when:
  • The agent calls the ExitPlanMode tool and presents the plan for approval.
  • You run /exit to leave plan mode manually.
  • The session ends.
After exiting, the session returns to its previous permission mode and the agent can proceed with implementation.