Bob N1KW has identified the cause of the brief glitch on going into TX on CW with the µBITx. While analysing the circuit Bob realised that the large capacitor C52, which is charged during receive, would feed back through R52 and R18 keeping the receive path (Q10, 11, and 12) after the balanced modulator “hot” for a brief period. When the transmit path is activated, the receive side of the circuit is going to remain on for some period of time due to the time constants of C52 and its loads. It is understandable that if both directions of the circuit are on, it could oscillate during that time!
To resolve the issue he simply added a diode in series with R52 (cathode toward C52) so that C52 can no longer back feed power to Q10, 11, and 12 upon initiation of transmit state. Now the transmitter output looks perfectly clean on the spectrum analyzer at beginning of TX. Shorting the diode causes the problem to show as before.
The same issue could apply to C64 when transitioning from TX to RX but at least there will not be spurious emissions going out over the air. He plans to add a diode in series with R66 in the same manner just for fun.
Note that in KD8CEC firmware you can add a keying delay to fix this issue in firmware.