service install pins the absolute node path, so removing that node version disables the proxy at the next boot

portless service install writes the unit with the absolute path of the node that ran it (options.nodePath ?? process.execPath in the service install path), so the proxy is pinned to e.g. ~/.nvm/versions/node/v24.1.0/bin/node. Remove that version later (an nvm cleanup, switching version managers) and the service fails at the next boot, taking every routed hostname down with it, and nothing points at the cause: the URLs just stop resolving.

Re-running portless service install re-elevates, rewrites the unit, and restarts the proxy, so the fix is one command, but nothing in the failure suggests it.

I found this by reading 0.15.5's published source rather than by hitting the boot failure live, so if the service already re-resolves node somewhere I missed, close this. Otherwise: resolving node at service start (/usr/bin/env node, or re-resolving in a small launcher) would remove the class of failure, and short of that a portless doctor check that the pinned path still exists would at least make it diagnosable.