I think it's just a case of leaky abstraction. I'd change the API to something like:

    robot.action = { action: 'jump' };
    
    robot.action = { action: 'move', amount: 4 };

    etc.

Or if you want to keep the action creators:

    robot.action = robot.jump();
    
    robot.action = robot.move(4);

Which makes it obvious that there can only be a single action per loop call.

Or the action could be the return value. IMHO, trying to guess the physical model by experimentation is painful. Why not propose a debug button to display the state of variables during the execution of script.

Exactly, the function should not be responsible to see if robot.action has any existing value. It should just return an action depending on the state.

Another idea is to return a map of actions.

    { jump: true, move: 4, shoot: true }

Making the user increment a nonce and use the remainder operator just to do multiple actions simultaneously is pretty heavy.

That's not right. You can put more than one action in the loop. This slowly completes the level:

    function loop(robot){
      // your code here
      robot.move(1)
      robot.jump()
    }

It shouldn't according to its own instructions.

I bet jump() has something like `if (robot.touchingFloor) { action = 'jump'; }` which makes your code work because it doesn't overwrite the robot action if it can't actually jump.

Switch the order of actions and you'll see the robot never jumps because its action gets overwritten (as stated in the instructions tab).

EDIT: Here it is:

https://github.com/reaalkhalil/JSRobot/blob/478848147a080d8b...

Ah I stand corrected, thanks. So I guess it's performing the last valid action.