I can't speak for everyone but having 1K of total RAM and no cross assembler from which to write that first native assembler were enough to stop me from attempting it :)
Probably about the best you could do then would be to interactively assemble code (probably without mnemonics!) and output the assembled version. You could probably do a reasonable interactive assembler without mnemonics in 256 bytes; not sure about base64 input and output (four letters/numbers per 3 bytes of code!), or error correction.
Did the machine not even have a memory of what was on the screen? Because that's usually at least 2K, and that would go a long way.
