mats: claude code and opus 4.6 is shockingly good at doing well defined and annoyingly tedious tasks
mats: like batch, vbscript, or powershell code golf minification iterations to shrink a payload so it fits in a clickfix variant
mats: or generating obfuscation tooling that reliably frustrates existing published deobf tools
asciilifeform: mats: seems like a great way to shoot yerself in the foot: if it's actually good at generating obfuscations, you'll be hard pressed to verify that they preserve correctness of the input
asciilifeform: ( conversely, if it's easy to verify, then the obfuscations aint that great, are they )
asciilifeform: this problem aint limited to the use of bots, either. recall how the more clever of polymorphic virii would in fact often produce offspring that doesn't run (crashed, or 'fired blanks', but in any case not functionally equivalent to the parent)
asciilifeform: the reason you don't see these much any longer is that afaik ~nobody bothers to try to statically deobfuscate anything nao -- you just let the thing run and watch the i/o
mats: how is it hard to verify? it may hide but it must run
mats: anyway i'm not doing any serious initial access work, the intent is to evade edrs and train internal teams
asciilifeform: np-hard
asciilifeform: i.e. if it's a serious obfuscator, rather than a symbol renamer