force @agent-plan

# .claude/settings.json

```
{
  "hooks": {
    "UserPromptSubmit": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "$CLAUDE_PROJECT_DIR/.claude/hooks/prepend_plan_agent.py"
          }
        ]
      }
    ]
  }
}
```


# .claude/hooks/prepend_plan_agent.py

```
#!/usr/bin/env python3
import json
import sys

try:
    # Read the JSON data passed from Claude Code via stdin
    input_data = json.load(sys.stdin)

    # Check if the current permission mode is 'plan'
    if input_data.get("permission_mode") == "plan":
        original_prompt = input_data.get("prompt", "")

        # Important: Only add the prefix if the user hasn't already typed it.
        # This prevents a double-trigger like "@agent-Plan @agent-Plan ..."
        if not original_prompt.strip().lower().startswith("@agent-plan"):
            # Prepend the built-in agent trigger to the user's prompt
            updated_prompt = f"@agent-Plan {original_prompt}"

            # Construct the JSON output to update the prompt
            output = {
                "hookSpecificOutput": {
                    "hookEventName": "UserPromptSubmit",
                    "updatedPrompt": updated_prompt
                }
            }

            # Print the JSON to stdout for Claude Code to process
            print(json.dumps(output))

except (json.JSONDecodeError, KeyError):
    # Fail silently, allowing the original prompt to go through if there's an error
    pass

# Always exit with code 0 to indicate success
sys.exit(0)

```