Yes, and sandboxed execution doesn’t even work, once you’ve managed to extract the painful AppArmor molars: Linux sandbox: agent unable to run commands - #2 by Isaac_Dario_Blanco is still unresolved (linking to a reply to my original post; Isaac Dario Blanco thoroughly outdid me in the diagnostics department, thanks very much.)
(Unrelated to this, my experiences with quota is causing escalating tics and twitches in my eyebrow region: Claude base quota seems to burn up like flash paper; Gemini base quota lasts a looooong stretch (even with Gemini 3.1 pro high, which alas even after thorough initialization instructions requires fastidious babysitting to get e.g. scientific compute algos right, and still manages to hallucinate things akin to “I have made it perfect! 8 out of 14 tests pass!”); however once dipping into overages (purchased AI credits) both models seem to consume those very rapidly. Anecdote not data.)
I know I’m patient to a fault.