Almost a perfect standard, but the prepended one byte header is a mistake IMHO. It makes it impossible to encode when the input size is unknown. Better to encode whether the last chunk is one byte or two at the end of the stream.
Please whoever is involved with this, revise the standard to not have a header and call this existing spec a beta. Otherwise, good work.
It would be easy to do. That 3-bit word just needs to start at 1 (since it encodes fewer than 8 options). Then the fixed 1-bit can instead encode whether another 7-bit segment follows.