Plan Mode

Plan mode is a read-only phase. The engine explores the codebase, designs an approach, and writes an implementation plan, all without touching a single file. You review the plan, then let it execute with full tools. This separates the thinking from the doing, so a change is reasoned through before it lands.

How it works

When plan mode is active, the engine restricts itself to read-only tools and follows a structured sequence:

1. The engine enters plan mode, which limits tool access to Read, Grep, and Glob.
2. It explores the code, identifies the files to change and the code to reuse, and composes a clear plan in its response.
3. It exits plan mode, which restores full tool access. The plan is optionally saved to disk.

So nothing is written while the plan is being formed. Write, Edit, and Bash become available again only after the plan is submitted.

The four phases

Inside plan mode the engine works through a fixed order:

1. Understand: explore the codebase with read-only tools.
2. Design: identify the files to modify and the code to reuse.
3. Write the plan: compose an actionable implementation plan.
4. Submit: restore full tool access so the plan can be carried out.

Configuration

Plan mode is registered by default. Control it and the save location under [plan]:

[plan]
enabled = true                          # register plan-mode tools (default: true)
plan_directory = ".wayland-core/plans"  # where plan files are written

Set enabled = false to remove the plan-mode tools entirely. The plan_directory is where a submitted plan is saved, so a plan stays on disk for review or to feed a later run.

From plan to execution

A typical loop is two steps: produce the plan in a read-only pass, review it, then run the work with the plan in hand. Because the plan is saved under plan_directory, you can read it back, edit it, or hand it to a fresh session for the implementation pass.

Next

• Built-in Tools
• Single-shot and REPL
• Configuration