EnterPlanModeTool: Enter Plan Mode

The purpose of EnterPlanModeTool is not to "make the model think for a while," but to switch the current session to a new runtime state: plan.

This means the system simultaneously changes:

• Permission mode
• Session objective
• User interaction expectations

So it is essentially a runtime state switching tool.

tools/EnterPlanModeTool/EnterPlanModeTool.ts:

import { handlePlanModeTransition } from '../../bootstrap/state.js'
import { applyPermissionUpdate } from '../../utils/permissions/PermissionUpdate.js'
import { prepareContextForPlanMode } from '../../utils/permissions/permissionSetup.js'

The actual core logic is in call():

context.setAppState(prev => ({
  ...prev,
  toolPermissionContext: applyPermissionUpdate(
    prepareContextForPlanMode(prev.toolPermissionContext),
    { type: 'setMode', mode: 'plan', destination: 'session' },
  ),
}))

Model determines task is too large/complex > EnterPlanModeTool > update permission mode to plan > prepare Plan Mode context > main thread enters planning state

There is a hard constraint in the source code:

if (context.agentId) {
  throw new Error('EnterPlanMode tool cannot be used in agent contexts')
}

This indicates that Anthropic does not want child agents to casually switch themselves into Plan Mode,
but rather leave this kind of workflow control to the main thread.

Plan Mode is not about writing a plan in isolation.
If requirements are unclear, you still need to use AskUserQuestionTool to gather information first before proceeding with that plan.

The value of EnterPlanModeTool lies in:

It upgrades "plan first, code later" from a mere prompting habit into an official state machine transition within the Claude Code runtime.