[BUG] PostToolUse:EnterPlanMode hook doesn't fire for user-initiated plan mode

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

PostToolUse:EnterPlanMode hook only fires when Claude calls the EnterPlanMode tool. It does NOT fire when the user enters plan mode via:

• Shift+Tab (interactive mode toggle)
• --permission-mode plan (CLI flag)

This means hooks cannot inject plan-specific context for user-initiated plan mode sessions.

What Should Happen?

PostToolUse:EnterPlanMode should fire whenever plan mode is entered, regardless of whether:

• Claude calls the EnterPlanMode tool
• User presses Shift+Tab
• User starts with --permission-mode plan

All three methods result in the same state (plan mode active), so hooks should respond to all three.

Error Messages/Logs

# Hook log shows NO entries for user-initiated plan mode sessions:

# Session cd786c7c (Shift+Tab) - no plan_template_injector entry
# Session fac43fe0 (Shift+Tab) - no plan_template_injector entry
# Session 3045ee76 (--permission-mode plan) - no plan_template_injector entry

# Only Claude-initiated plan mode fires the hook:
2025-12-28 23:13:04.708 [INFO] [plan_template_injector] starting
2025-12-28 23:13:04.709 [INFO] [plan_template_injector] plan_template_injected via PostToolUse:EnterPlanMode session=e8be406e-...

Steps to Reproduce

1. Add EnterPlanMode to your allow list in ~/.claude/settings.json:

{
  "permissions": {
    "allow": ["EnterPlanMode", ...]
  }
}

1. Create a PostToolUse:EnterPlanMode hook:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "EnterPlanMode",
        "hooks": [{ "type": "command", "command": "echo 'HOOK FIRED' >> /tmp/plan-hook.log" }]
      }
    ]
  }
}

1. Test each plan mode entry method:

# Clear log
rm -f /tmp/plan-hook.log

# Test 1: Shift+Tab
claude  # press Shift+Tab twice to enter plan mode, type any prompt, exit
cat /tmp/plan-hook.log  # RESULT: empty - hook didn't fire

# Test 2: CLI flag
claude --permission-mode plan -p "Hello"
cat /tmp/plan-hook.log  # RESULT: empty - hook didn't fire

# Test 3: Claude calls EnterPlanMode
claude -p "Please use the EnterPlanMode tool now"
cat /tmp/plan-hook.log  # RESULT: "HOOK FIRED" - only this one works

Claude Model

None

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.0.76

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Verified Test Results

| Entry Method | PostToolUse:EnterPlanMode Fires? | Session ID Tested |
|--------------|-----------------------------------|-------------------|
| Shift+Tab | No | cd786c7c, fac43fe0 |
| --permission-mode plan | No | 3045ee76 |
| Claude calls EnterPlanMode tool | Yes | e8be406e |

Use Case

I maintain a plan template (~1,100 tokens) with formatting guidelines. I want to inject it only when plan mode is active, but:

• SessionStart: Wastes tokens on sessions that never use plan mode (~90%)
• PostToolUse:EnterPlanMode: Only fires for Claude-initiated plan mode (incomplete)

Suggested Fix

Option A: Fire PostToolUse:EnterPlanMode synthetically when user toggles plan mode

Option B: Add new PermissionModeChange hook event with matcher support:

{
  "hooks": {
    "PermissionModeChange": [
      { "matcher": "plan", "hooks": [...] }
    ]
  }
}