By default, sudo-proxy separates approval (TUI prompt) from privilege escalation (sudo). The TUI shows the command for single-keypress Y/N approval, then sudo handles password authentication — both in the same terminal window.
The --pkexec flag bypasses the TUI entirely and lets pkexec handle both
authentication and approval in its own dialog. This mode exists but is not
recommended because polkit conflates authentication (proving who you are) and
authorization (approving what to do):
-
Without auth caching: pkexec asks for a password on every single request. This is safe but impractical for an AI agent workflow that may run dozens of commands.
-
With auth caching: pkexec skips its dialog entirely for the cache duration — commands run silently with no approval prompt. This is unsafe and defeats sudo-proxy's human-in-the-loop design.
There is no middle ground in polkit. The TUI mode solves this by always prompting for approval (one keypress, no password) while letting the OS handle password authentication separately.
Only requests sent with privileged: true go to pkexec. In those, every step
escalates, so each one raises its own dialog. A structured plan that escalates
only some steps (through run-as nodes) still gets the TUI prompt and runs
through sudo, even in --pkexec mode. A pkexec dialog shows a single step, so
the unescalated steps of such a plan would otherwise run with no approval at
all.
Each escalated step runs as pkexec env -C <dir> -- <argv>, so it starts in
the plan's working directory rather than the target user's home, which is
where pkexec runs programs by default. The polkit dialog therefore names
/usr/bin/env as the program.
If you do use --pkexec mode, you can optionally configure polkit to cache
authentication for a few minutes, similar to sudo's default behavior.
This is a user decision. The rule applies to all pkexec calls from your
user, not just those from sudo-proxy — polkit has no way to distinguish the
calling program. Evaluate whether this trade-off is acceptable for your setup.
The pkexec-cache tool manages this rule:
# Check if the rule is installed
pkexec-cache
# Install the rule (detects your username from SUDO_USER)
sudo pkexec-cache --create
# Remove the rule
sudo pkexec-cache --deleteThe username comes from SUDO_USER, then PKEXEC_UID, then LOGNAME/USER.
It must name an existing non-root account in the passwd database and match the
usual login-name charset (^[a-z_][a-z0-9_-]*\$?$, at most 32 characters);
otherwise --create refuses to write the rule.
Some distributions ship /etc/polkit-1/rules.d as root:polkitd 0750, so a
non-root user can't look inside it. The check then prints
unknown: cannot read …: permission denied (run with sudo) instead of guessing;
run sudo pkexec-cache to see the real status.
The rule is written to /etc/polkit-1/rules.d/50-pkexec-cache.rules
and caches authentication for the calling user with these conditions:
subject.active— only the foreground session (not background terminals)subject.local— only local sessions (not SSH)AUTH_ADMIN_KEEP— cache the admin password for ~5 minutes
polkitd monitors its rules directory and reloads automatically — no service restart is needed.
There is no GUI tool to manage polkit rules on any distribution. On current
Ubuntu and Debian, polkit uses the JavaScript .rules format — the older
.pkla format is deprecated. The
Arch Wiki polkit page is the most
comprehensive reference.