This is a fantastic update. First, to answer R2D2: Yes, we are officially
logging a C-level focus shift to "Reflex Agent & Sub-Millisecond Physics
Tracking."

Your discovery about the arrow keys fundamentally changes the actuation layer
for worker-02. If BREAKOUT.JS doesn't currently support mouse tracking, our
move_mouse() command in the reflex script will do absolutely nothing. We must
bridge the high-speed vision (tracking the ball) with keyboard actuation
(pressing and holding arrows).

And your instinct about VNC is spot on. If we run vncviewer in windowed mode,
we have to calculate the window offsets and deal with window manager
decorations. Running it in full screen forces the coordinate system to an
absolute 0,0 origin, completely eliminating math errors.

In standard Linux environments, you can usually force TightVNC to launch
fullscreen via the command line like this:
vncviewer -fullscreen hauki-x86

Upgrading the Reflex Agent to Keyboard Actuation
Since the paddle requires you to hold the arrow keys down, we can't just send
discrete "press" events. We need a continuous feedback loop.

Here is the logic we will inject into worker-02:

Vision: Track the X-coordinate of the ball (using the exact same mss + OpenCV
frame differencing we just built).

Evaluation: Compare the Ball's X-coordinate to the Paddle's assumed
X-coordinate (since the paddle stays on the Y-axis).

Actuation (The Key Hold):

If the Ball is significantly to the left of the Paddle, hold the Left Arrow.

If the Ball is significantly to the right of the Paddle, hold the Right Arrow.

If the Ball is directly above the Paddle, release both arrows to stop.

You already built an incredible keyboard actuation tool earlier. The teaching
script you were looking for is exactly where you left it: /opt/hauki-gem/teache
r.py (which powers the engine_actuation.py methods).

However, since engine_actuation.py typically uses xdotool for discrete
keypresses (which doesn't handle "holding" a key down very gracefully in a
tight while-loop without flooding the X11 buffer), we will use Python's
pyautogui library for the Reflex Agent. It has explicitly designed keyDown()
and keyUp() functions perfect for game emulation.

Let's do this in two steps. First, we install the keyboard control library.
Then, we rewrite the Reflex script to use the Arrow Keys instead of the Mouse.

Execute this block:

(..)

The Launch Sequence
Here is how we test this:

In your VNC client on hauki-teacher, run vncviewer -fullscreen hauki-x86.

Type /html5 BREAKOUT.JS and hit Enter so the game starts.

Quickly flip to your SSH terminal and run /opt/hauki-gem/venv/bin/python3
/opt/hauki-gem/agents/worker-02/reflex.py.

The AI will capture your full screen. Every time the ball moves, it will
calculate the distance from where it thinks the paddle is, and physically hold
down the left or right arrow keys on your keyboard.

(Note: Because we are guessing the paddle's movement speed paddle_x += 10, it
won't be perfectly calibrated yet, but you should instantly see it thrashing
left and right trying to catch the ball!)