In general I've come to the conclusion that self-referential DNA is problematic.
First, something like refdnalen means that a bot's DNA becomes part of its phenotype. DNA isn't just selected based on what it does, but what it is. I think this can only have constrictive consequences for natural selection.
For things like .mutate, it would be in the same boat as delgene. Way too overpowered to let bots use for venom, infoshots, etc. We'd need to create more exceptions for what memory locations are valid.
That said, crossover and sendgene are reasonable. Crossing over should be tied with some sort of sexual reproduction. Although I could maybe see it being independent of reproduction. Sendgene should be engineered to be less aggressive than viruses. Maybe instead of sending genes, bots request genes.