Re-enable p10k instant prompt (sudo-key ssh-add moved above it). Drop
oh-my-zsh.sh (second compinit, update check); set up what OMZ plugins
need ourselves. Cache java_home, go env, fzf --zsh output. Disable the
kubecontext segment (ran kubectl per shell). Log starts over 1 s to
~/.cache/zsh/startup-slow.log to catch cold-start stalls.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Move antidote update from every shell start to update_mac.sh: it dropped
the bundle cache each time, forcing a rebuild on the next start. Background
the locale write and skip the brew --repository fork.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Start a second agent with the sudo key added via ssh-add -c, forwarded by
agent shells (see ~/.ssh/config); skip agent setup and SUDO_ASKPASS there.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Keep the server entrypoint focused on server policy while moving update checks, Antidote loading, interactive defaults, and prompt sourcing into reusable modules.
The check_tools_repo (git ls-remote | awk) and update_dotfiles (git fetch)
background jobs could stall on a network call and, being disowned (&!),
orphan into days-long processes holding unreaped git/awk children.
Bound both network probes with `timeout` (guarded, absent on macOS) and
disable ssh multiplexing so no daemonized mux master inherits our fds.
Also dedup the double update_dotfiles startup call.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
