Both install through the lazy mise wrappers (npm:dimcode and
npm:@moonshot-ai/kimi-code), preinstalled on new machines and migrated onto
existing ones unless preinstalls were removed, and each gets a menu entry in
both places: under Setup > Defaults > Agent as a default-agent choice, and
under Install > AI for installing without changing the default. The launch
spellings come from each CLI's actual flags — bare dim is the TUI and `dim
exec` the one-shot, kimi runs full-auto as `kimi --auto` with `--prompt` as
its one-shot. The Install > AI entries are guarded by `mise where`, since the
preinstalled stub would otherwise read as installed. The menu marks are the
Kimi simpleicon and Dim's traced from its app icon, since upstream publishes
no monochrome SVG.
* Quote install-app and install-font names like install-and-launch
* Quote the package list too, not just the display name
The display name was quoted but omarchy-pkg-add's own arguments were still interpolated into the bash -c string raw, so `omarchy install app Vim 'vim; id'` ran id. The list has to reach the helper as several words, so it cannot be quoted whole: it is split the way the unquoted expansion split it and each word is quoted on its own. Reading with -d '' keeps a newline-separated list intact instead of dropping every package after the first, which plain read -a would. install-font's package is singular and is quoted whole, and install-and-launch carried the same flaw.
Reported by acrogenesis in review of #7843.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Test that install-font skips font-set when pkg-add fails
The hostile-package case was asserting the family still got set, which only held because the mock always exits 0. pacman would reject that name and the && chain would skip font-set.
* Keep the installers working when errexit is inherited
read -d '' always ends at EOF rather than on its delimiter, so it reports failure on every input. Under an inherited errexit the installers exited there and built no command at all.
Reported by Codex XHigh in review of #7843.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
install/post-install/first-run-mode.sh shipped on quattro between 53e26115 and 75cb4f71, and its final body writes `Cmnd_Alias FIRST_RUN_CLEANUP = /usr/bin/rm -f /etc/sudoers.d/first-run, /bin/rm -f /etc/sudoers.d/first-run`. The predicate's case listed only the two `/bin/rm` spellings, so that line fell through to the user-spec test, failed it, and the whole file read as hand-written. The migration then left it alone and wrote its machine marker, which is permanent: on an offline install from that window the account keeps passwordless `/usr/bin/systemctl` for good, and nothing looks at the file again.
Adding the string is the whole fix. The test now carries all nine bodies the installer wrote across both locations rather than the eight from install/preflight.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <codex@openai.com>
`>|` is a plain redirect with noclobber overridden, not a redirect followed by a pipe. command_destinations detached `>` from its target before looking at the bar, so the target read as `|` and the privileged path behind it was never examined: `cat <<EOF >| /etc/udev/rules.d/99-x.rules` with `$HOME` in the body produced no finding at all, while the same write through `>` produced one.
Normalizing `>|` to `>` alongside the existing `>>` handling closes it. The fixture fails without the normalization.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
test/shell.d/config-test.sh greps every file under migrations/ for `upgrade-to-quattro` and fails the suite when one matches, because pre-4 layout work belongs in the upgrade command rather than in a migration. The comment explaining why this particular cleanup is the exception named that command literally, so it matched the guard and config-test.sh failed on this branch while passing on quattro.
The comment now names the Omarchy 4 upgrade command without spelling the file, which leaves the guard able to catch a migration that actually reaches for it. agents/skills/migrations.md still names `bin/omarchy-upgrade-to-quattro` in full, and it is not under migrations/.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The failure advice printed a ready-to-paste annotation with the scan's
verdict already filled in, so the shortest way past the check was to copy
back what it had just concluded. That is worst exactly where the scan is
weakest: a path it cannot follow through a variable reads as an ordinary
value, and the annotation it offers for that case is paths=none.
Print the annotation with the path list left blank and say why the author
has to fill it in. The scan's own reading stays in the report above it, so
nothing diagnostic is lost.
Running the migration on a real machine with no cached sudo credentials
printed sudo's "a terminal is required to read the password" and still
exited 0. bin/omarchy-migrate writes the completion marker on a zero exit,
so the cleanup would have been recorded as done on every install that runs
migrations without a terminal, and never tried again.
Probe for elevation before the combined existence check and exit non-zero
when it fails, so the marker stays unwritten and the next run retries. The
probe is skipped when the directory is readable as-is, which is the case
when migrations run as root.
Review of the previous commits turned up four places where the predicates
and their tests disagreed with the tools they are modelling, each checked
against udevadm verify, systemd-analyze verify and visudo -cf rather than
against reading of the sources.
An empty ExecStop= resets the list, so a unit an administrator neutralised
that way runs nothing at shutdown and is no longer ours to remove; the
predicate now tracks the last state instead of returning on the first home
path it sees. A file whose last line ends in a backslash still carries a
live directive for systemd, so the pending logical line is emitted at EOF;
udev ignores such a line and sudo rejects the file outright, so this costs
those two nothing. The scanner's taint pass now reads += appends, which its
own comment already promised: the value of an append is no use, but a name
that reaches a user root through one has to be judged on it.
Two regression guards passed against the implementations they were written
for. The udev continuation fixture put the whole RUN+= below the comment, so
it matched whether or not the pending half was carried across; the split now
falls inside the RUN+= value. The sudoers one kept its file on the strength
of a spec above the comment, so it could not fail either; the hand-written
spec now sits below. Both fail against a mutant that discards the pending
line. The comment above the second also claimed a continued comment stays a
comment, which visudo contradicts.
An installer that writes a root-owned file through a heredoc with an
unquoted delimiter (<<EOF rather than <<'EOF') has the installing user's
shell expand the body first, so a user-controlled value is baked in as a
literal. Send that into /etc and root later reads or executes a path the
unprivileged user picked: a udev rule carrying
$HOME/.local/share/omarchy/bin/... resolves through a symlink that user
owns, so replacing the symlink gets their code run as root.
Add the static check. A heredoc is flagged when its delimiter is
unquoted, its body contains an install-time expansion (escaped \$VAR does
not count, since that is left for a root daemon to expand at runtime),
and its output reaches /etc, /usr, /opt, /srv, /boot or /var/lib via sudo
tee, sudo dd, a redirect, or an install/cp/mv of the generated scratch
file. Destinations written as variables are resolved from the file's own
assignments.
Sites that genuinely need install-time expansion declare it inline:
# omarchy:heredoc-expands paths=none -- $servers is a validated IP list
paths= is machine-checked against the expansions the scanner finds to be
path-shaped, so this cannot become a rubber stamp: adding a $HOME/... to
an already-annotated heredoc makes the declaration false and trips the
check again. Path expansions anchored under a root-owned prefix, as in
"/etc/systemd/system/$unit", are correctly not path-shaped.
Annotate the sites the scan reports, each of which expands a scalar: DNS
addresses in omarchy-dns, a literal PAM line in
omarchy-setup-security-fingerprint, kernel cmdline parameters and
usernames in omarchy-upgrade-to-quattro. omarchy-provision-owner expanded
a unit name that was already a constant, so its delimiter is now quoted
and the name hardcoded; the generated unit file is byte-identical.
omarchy-windows-vm declares paths=storage,shared, the only site that
interpolates a user-chosen path.
Fixtures prove non-vacuity in both directions: the write routes other
than a pipe into sudo tee, the shapes that must stay quiet, udev rules
and a shutdown unit taken verbatim from this repository's history, and
the rubber-stamp case where a paths=none annotation on a baked $HOME path
still fails.
Three installers that no longer exist each left a root-owned file on
disk, and nothing in Omarchy has ever removed any of them.
/etc/sudoers.d/first-run granted the installing account passwordless
sudo for the rest of the first boot, unrestricted /usr/bin/systemctl
included from 2025-10-14 on. omarchy-first-run clears its first-run.mode
guard before eight set -e steps and only deletes the grant after them,
so any failure in between strands it with nothing left to retry.
/etc/sudoers.d/tsui named whatever $(which tsui) resolved to for the
installing user, normally a binary under their own home that the vendor
script had just written without sudo.
/etc/systemd/system/omarchy-plymouth-shutdown.service ran an ExecStop
under the installing user's home as uid 0 on every shutdown.
Each file is judged against what the installer that wrote it actually
produced. The first-run grant was rewritten eight times and only the
last four carry both Cmnd_Alias lines, so rather than key on those, every
active line must be one the installer emitted and one of them must be
its own self-cleanup. The shutdown unit is disabled but never stopped:
stopping it is what would run the ExecStop being taken away.
Generalize the migrations.md exception, which framed itself around pre-4
layout transitions and so did not cover installers retired on their own.
Omarchy 3 wrote 99-power-profile.rules and 99-wifi-powersave.rules with
an unquoted heredoc, baking the installing user's home into a rule udev
runs as root. That path resolves through ~/.local/share/omarchy, a
symlink the unprivileged user owns, so replacing it and provoking a
power_supply event runs their code as root. HEAD points the rules at
/usr/bin under new names, but the one-shot cleanup for the old
filenames was dropped, leaving the file on every install that came up
through the 3.x line.
Remove a legacy file only when an active RUN+= really does run that
filename's binary out of a home directory, so a rule of the same name a
user wrote themselves stays, comments and all.
The upgrade command only runs on a machine still crossing 3 to 4, so a
vulnerable file an old installer wrote never gets swept on an install
that crossed already. It ends by running omarchy-migrate, so a single
migration reaches both populations.
Validating /etc/omarchy.conf walks its parent chain to /, and that walk leaves
its own subject behind in the global failure_context. The comparison that
follows -- the one deciding whether the authorization actually names this
checkout -- sets no context of its own, so an unauthorized development tree
refused with:
refusing to publish: directory / (must be root-owned and not group- or
world-writable) failed validation
naming a directory that had just passed, and pointing the reader at a
filesystem problem that is not there. Seen on a worker VM running an
unauthorized checkout, where / is 0555 root-owned.
The hint line underneath was already right, which is what kept this from being
worse. Restore the context before the comparison so the first line agrees with
it, and assert both halves: that the refusal names the authorization, and that
it does not blame the root directory.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0115LngksSpXLD9NSXBEP3ki
Quattro stopped making the Chromium managed-policy directory world-writable while this branch was open, and the block it deleted from the theme install leaf sat directly above the comment this branch rewrites, so the two edits landed in one hunk. The resolution keeps the hardening — the policy directory is set up through install/config/browser-policy.sh now — along with the first-run seed and the comment that names both things the seed does.
omarchy-dev-link told people /usr/share/plymouth was not covered by a link
and to reach for omarchy-dev-pkg-test. Since the publisher authorizes a
dev-linked checkout out of the root-owned /etc/omarchy.conf, the Plymouth and
SDDM themes are republished from the checkout like every other
$OMARCHY_PATH-resolved tree, so the help was left describing the old
behaviour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Checking a queue for jobs and then deleting it leaves a window in between, and the sudo that does the deleting can sit at a password prompt for as long as someone takes to type. A job submitted in that window was cancelled by a deletion that had already decided the queue was empty. cupsreject closes the queue to new work first, which also stops more jobs piling onto one that is being left behind and can no longer route them. It comes from the cups package, which stays.
What a queue with jobs on it means is now said rather than implied: whatever is already at the printer finishes, whatever is still waiting cannot be routed with the daemon gone, and the person who owns them has to cancel what is left.
A queue that another administrator removed while this was running is the outcome wanted, not a failure to keep the package installed for, so a deletion that fails is checked against whether the destination is still there.
Co-Authored-By: Codex XHigh <noreply@openai.com>
Mutating each root-side control one at a time showed three that no test could
see: a destination directory root does not own, a single user-owned asset
inside an otherwise root-owned packaged directory, and an asset left group- or
world-writable by its own mode. Deleting any of the three left the suite green,
because the existing cases mark a whole tree untrusted and are caught by the
directory check before the per-file one is reached.
The harness already had the hook for the ownership pair: TEST_UNTRUSTED_SOURCE
makes the stat shim report a chosen prefix as uid 1000, so those two only need
it pointed at a destination directory and at a single file rather than at a
whole tree. A mode has to be real, so that case stages a copy of the packaged
tree the shim reports as root-owned and loosens one asset in it.
The empty logo is refused by the destination size bound rather than the
caller-side one, so that case pins the behaviour without isolating the check;
the two bounds are exactly redundant.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0115LngksSpXLD9NSXBEP3ki
cups-browsed is the daemon that watches the network and creates print queues by itself. Hardening it took a root daemon with a predictable cache down to a confined service account, but a daemon that turns anything advertising itself on the network into a print queue is a lot of exposure for a convenience, so it comes out of the default install while that is reworked. Only the discovery half: CUPS itself stays and printing keeps working, with each printer added by hand in Print Settings.
The migration disables the unit before removing the package because that is the only order that works: pacman deletes the unit file but not the enable symlink, and once the unit is gone systemd can no longer resolve it by name to clean that up.
It then removes the queues discovery generated. cups-browsed keeps those when it stops, since KeepGeneratedQueuesOnShutdown defaults to Yes, and they route through its own implicitclass backend, which goes with the package, so they cannot print again. Idle ones go. A queue with jobs on it is left alone and named: implicitclass only needs cups-browsed to choose a destination, so a job already past that point finishes on its own, and deleting the queue would abort it. One printer's job does not hold up the removal. A printer added by hand has an ipp:// or usb:// device and is left where it is.
A queue whose jobs cannot be asked about is left alone rather than assumed idle, including one named so that lpstat would misread it -- "all" is its word for every destination, and a leading dash or a comma reads as another option or a list.
Where CUPS does not answer at all, or a queue will not delete, discovery is still stopped but the package stays and no marker is written. omarchy-migrate records a migration for the user as soon as it exits zero, so that is where the machine stays until someone removes the package by hand, and the message says so rather than implying a retry.
The queue list is read under LC_ALL=C because lpstat translates "device for", and captured rather than piped, so a cupsd it cannot reach is reported instead of reading like a machine with nothing to clean up.
It removes with plain pacman -R rather than omarchy-pkg-drop, which passes -n and would discard /etc/cups/cups-browsed.conf instead of keeping it as a .pacsave. A removal meant to be temporary should not delete the machine's copy of its own configuration. Without -s either, so it only ever removes the package it names: sweeping newly unneeded dependencies is nothing today, but it is not a promise a rolling dependency graph can keep.
Queue names come off the network, since cups-browsed names its queues after what the printer advertised. CUPS allows every printable character but space, tab, / and #, and lpstat and lpadmin take a destination as an option value, so a name with a leading dash or a comma is reported rather than passed to them and guessed at.
Migration state is per user, so a machine-wide marker records the one removal. Without it, an account whose first migration run came after someone deliberately reinstalled discovery would quietly take it back out again.
The install-time override for cups-browsed.conf now waits for cups-browsed rather than for CUPS. Guarding it on a file CUPS still ships would write a configuration file for a package nothing installed, and pacman would later land the package's own copy beside it as a .pacnew.
The hardened configuration stays in the tree. omarchy-settings still ships the cups-browsed.conf override, the sysusers account and the service drop-in, so they are what discovery returns onto.
Co-Authored-By: Codex XHigh <noreply@openai.com>
The comment above the seed was the only thing recording that color_scheme and color_scheme2 are both zero in order to follow system appearance rather than force dark. Generalizing it to "first-run defaults" left two magic numbers with nothing to explain them, so the next person touching an unrelated first-run setting has no way to tell that changing them regresses theme following. Name both things the seed does.
Co-Authored-By: Codex XHigh <noreply@openai.com>
Chromium 151 flipped MasterPrefs::eula_required from false to true, so a first run with no seed now stops on a blank terms-of-service dialog before the browser opens. The opt-out is written in two places that each run exactly once: the install leaf that runs during ISO finalization, and the one-time 3.x upgrade. A machine already on Quattro runs neither again, so it keeps the old seed and still meets the dialog the first time anyone launches Chromium, including from every user account created after the install.
The migration writes the same seed those two paths write. It compares before writing so the second user on a shared machine no-ops rather than repeating a machine-wide repair, and the literal is duplicated rather than sourced because a migration repairs the state of its own moment and must not drift when the seed later changes.
Co-Authored-By: Codex XHigh <noreply@openai.com>
Deleting the whole ancestor walk from validate_trusted_directory and checking
only the immediate parent left the suite green, so the invariant the design
rests on had no coverage: a writable ancestor lets an attacker swap a validated
directory out from under the leaf. Reject a run whose destination grandparent
is world-writable while the destination itself is pristine.
The harness also rewrites the root script's fixed /usr/share paths by string
substitution. A drifted string silently no-ops that rewrite and would point the
simulation at the real system tree, so assert each substitution landed. Assert
the scratch directory exists too: a failed mktemp left it empty and the suite
carried on building paths from it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
Every check inside the root shell is a bare [[ ]] or (( )) assertion that
aborts under set -e, so a refusal exited with status 1 and no output at all.
The floating-terminal wrapper then printed its green "Done!" for any status
but 130, so a failed boot-theme change read as a success.
The refusal a working machine actually hits is omarchy dev link, which points
OMARCHY_PATH at a checkout the desktop user owns. Name that case outright and
point at omarchy dev unlink; report every other rejection through an ERR trap
that names what failed validation.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Keep SDDM color substitution collision-free for White themes, based on the fix proposed in #8469.
Co-authored-by: itz4blitz <itz4blitz@users.noreply.github.com>
AGENTS.md asks for unquoted variables inside `[[ ]]`, with quotes reserved for string literals being compared. The three conditions added here quoted them.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
Neither half of the validation was covered. Dropping `&& -c $cached` from the wrapper left the whole file green: all three poison values fail on the pathname prefix, so none of them ever reached the character-device test. A path that matches the hiddev glob but is not a device now covers it, and it is the real case rather than a synthetic one -- the display replugs, the interface renumbers, and the cached node is gone. It is added only when the host has no such node, so a machine with the display attached cannot fail there spuriously.
The no-XDG_RUNTIME_DIR assertion had the same problem for the opposite reason: its decoy held a path the validation rejects on its own, so restoring the `${XDG_RUNTIME_DIR:-/tmp}` fallback left it passing. It asserts on the open now instead of on the contents -- a FIFO with no writer blocks whoever opens it, so a wrapper that consults the path hangs and one that ignores it exits. mkfifo is atomic and fails outright if the path is taken, so it still neither overwrites a file nor follows a symlink at the fixed path.
Clearing created_tmp_cache as soon as the decoy is removed keeps this run's EXIT trap from deleting a concurrent run's decoy at the same fixed path, which would have let that run pass against the old code.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
The scheme check ran after the interactive branch, which had already handed the URL to curl: a refused `file://` or `ftp://` URL was dereferenced first, the whole string went out to Google's favicon endpoint in a `domain=` query, and a fetch that happened to succeed left an orphan icon and a rebuilt icon cache behind. Validating immediately after normalization puts the refusal ahead of every dereference in both branches.
A leading space also kept the URL out of the scheme test entirely. `normalize_webapp_url` saw no scheme, prefixed it, and ` file:///etc/passwd` became `https:// file:///etc/passwd`, which passes `^https?://`. The desktop `Exec` field is unquoted, so it splits back into two arguments, and `omarchy-launch-webapp` forwards everything after the first to the browser, where a bare argument is another URL to open. Refusing whitespace closes that without touching desktop-entry escaping.
Scheme comparison is case-insensitive because schemes are: `HTTPS://example.com` installed before this check existed and has no reason to stop.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
The username prompt already refuses the service accounts a desktop user must not claim, cups and lp among them. A user who took cups-browsed would get a primary group of that name, and the CUPS authorization written here puts that group in SystemGroup, handing that desktop user the passwordless administration the rest of this change removes.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
cupsd compares directive names with _cups_strcasecmp, so a hand-edited "systemgroup sys root wheel" is live configuration, but matching $1 against the canonical spelling skipped it and appended a second directive at the end of the file. parse_groups accumulates the groups of every SystemGroup directive it reads rather than replacing them, so both lines took effect and wheel kept the passwordless administration this is meant to remove, with the migration reporting success.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
A bracket range is collated, not ASCII, so `[a-z]` admits `é` under
en_US.UTF-8 and refuses it under C: the same theme URL installed on one
desktop and was refused on the next, and the manual's stated set was only
true in the C locale. Pinning the comparison makes the written set the
shipped one everywhere.
git reads a URL as scp-style when a colon appears before any slash, so the
path after it need not contain one. Requiring a slash left the prefix on
`git@host:omarchy-blue-theme.git` and derived the name
`git@host:omarchy-blue`, which the allowlist then refuses -- a repo that
clones fine could no longer be installed at all.
Co-Authored-By: Codex XHigh <codex@openai.com>
The migration recomputed whether cups-browsed was active at the start of every run, after an earlier run may already have stopped it. A run interrupted between that stop and the completion marker left the service down, and the retry that followed read it as inactive, skipped the restart, and wrote the marker anyway: printer discovery stayed off until the next reboot with no migration left to bring it back. Keying the restart to whether the unit is enabled survives the interruption, and reports not-enabled for a unit the user masked or disabled, which restarting would fail on and abort the migration short of its marker.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
GNU sed's N auto-prints the pattern space and exits at end of input, so the
`:a;N;$!ba` slurp skipped every following s/// for a value with no newline in
it. That is every value except the injection attempt the escaping exists to
stop, so the Exec quoting's own backslashes were written to the file unescaped
and GLib refused to load the entry.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex GPT-5.6-sol (xhigh) <noreply@openai.com>
The slash guard was the only thing keeping a name out of the directory structure, and nothing tested it: deleting it left the suite green, because creating the launcher directly in the applications directory already makes the redirect fail on its own, with a raw bash error instead of the message. The assertion is on the message now, alongside the traversal case the guard actually closes -- on quattro a name of `../../../../escaped` writes its launcher clean outside the applications directory.
The interactive prompt read the name, fetched the favicon, wrote it and updated the icon cache before the name was ever checked, so a URL typed into the Name field left an icon behind on every attempt. Validating as soon as the name is read covers both paths from one place.
Removing by name also scanned unconditionally, so a machine with no applications directory printed a find error where omarchy-remove-gaming-xbox-cloud does not hide stderr.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Run cups-browsed as a locked service account with a dedicated cache and a focused systemd sandbox. Restrict automatic queues to driverless IPP printers, remove wheel from passwordless CUPS administration, replace cups-pdf with Polkit-backed setup, and migrate existing systems safely.
Reported-By: Erik Hunstad (Bad Sector Labs)
Co-Authored-By: Daybreak Blue <noreply@openai.com>
Each of these is a Text rendering external data with no textFormat, written in
a form that passed silently. None exists in this tree, so they were holes in
the guard rather than live exposures — but a guard is only worth what it
catches, and every one of them is a single line someone could plausibly write.
Text /* why */ { strip_noise knew // and not /* */, so a block
comment between the type name and its brace hid
the element from every rule at once
QQ.Text { ... } a namespaced import made the name compare unequal
to `Text`, and the element was skipped outright
visible: textFormatEnabled textFormat was matched as a substring, so a
lookalike property exempted the whole block
component Info: a component root with its Text on the next line;
Text { the one-line form was covered and this was not
an unreadable subdirectory rglob() swallows a directory it cannot enter, so
a locked subtree scanned as though it were empty
The scan moves out of the heredoc into qml-text-format-scan.py, taking its root
as an argument, because nothing could run it over anything but the real tree —
and a scanner whose only input always passes cannot be shown to fail. The test
now runs it over nineteen fixtures, one per form above and one per form the
scan already handled, so a later edit that loosens it fails here instead of
going unnoticed until something renders a remote image.
Two limits stay open and are written down in the module docstring rather than
papered over: text assigned from elsewhere (a Binding element, PropertyChanges,
an onCompleted assignment, a property alias onto a child) is invisible to a
scanner that reads each element's own declaration, and a regex literal holding
a brace throws off the brace depth. Neither shape exists in this tree and both
need a QML parser, not another regex.
Co-Authored-By: Codex XHigh <noreply@openai.com>
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The card binds the body Text to styledBody, which rewrites newlines to <br/>
*after* sanitizeBody has run. That rewrite inserts tag syntax into text the
stripper deliberately kept: a kept tag may hold a `<` of its own, and `<x`,
newline, `<img src="http://host/x.png">` is one tag named `x` to both the
stripper and Qt, so it survives whole — until the rewrite splits it into
`<x<br/>` and a live image tag the input never contained.
Measured against Qt 6.11.2 with an offscreen StyledText and a local HTTP
server: that body issues the GET after this branch's sanitizer and issues
nothing before it, because the one-pass /<img[^>]*>/gi it replaces deleted the
inner substring outright. The whole-tag bound is still the right trade — it is
what stops the stripper manufacturing tags — but it only holds if nothing edits
the string afterwards.
So move the rewrite into NotificationLogic, next to the reasoning it depends
on, and strip again after it. What Qt parses is then what was checked last. The
tests assert on styledBody for the same reason, since sanitizeBody's output is
no longer the string that reaches the renderer, and a regex assertion pins the
card's binding because no JavaScript assertion can see a QML property.
The section had grown a paragraph per review round, each one explaining why the last was right, until one offer took a third of the file. Most of it was reassurance about what the command refuses rather than anything an agent has to do, and the command enforces that itself whatever the prose says.
What is left is the instruction: offer it and never run it unprompted, say how to lift it, which of the two names to pass and why the binary is the better one, quote it because the name is the crashed program's to choose, and name the interpreter collision before muting python or node on someone's behalf. Fifty-four lines to thirty-two, with nothing dropped that changes what the agent does.
The mute was reachable only as `omarchy-toggle crash-ignore/<program>`, which asks whoever runs it to know the flag layout, to reduce a binary's path to the name the watcher keys on, and to have read the rule that a name climbing out of that directory writes an unrelated toggle. All of that was carried in the skill's prose, which is the wrong place for a rule that has to hold: prose is advice, and the thing being advised about is a name the crashed program chose.
So it is a command now. `omarchy crash mute hyprland` silences that program, `off` lifts it, `toggle` flips it, and no argument lists what is muted. It takes the binary's path as readily as the name and reduces it the way the watcher does, so the `Executable:` line from `coredumpctl` can be handed straight to it; it refuses what is not one component of a name, so it cannot be talked into writing outside its own directory whatever it is given; and it re-reads the flag afterwards and reports what is now true rather than what was asked for. The listing counts only regular files, because that is all the watcher honours -- anything else in there would read as muted while the crashes kept arriving. A leading `--` is consumed so a program named `-h`, which the router would otherwise answer with its own help, can still be muted.
The watcher gained an unrelated fix that this uncovered. Its fields are read with `IFS=$'\t'`, and tab is IFS whitespace, so an empty field collapsed into the next delimiter and shifted every field after it along one: a crash whose comm was empty had a path read as its pid and was discarded as somebody else's. A process can set its comm to nothing, so that was reachable. Empty fields now arrive as a dash like missing ones, and a dash joins the empty and dot cases that fall back to `unknown`.
Co-Authored-By: Codex XHigh <noreply@openai.com>
The manual had single quotes covering "punctuation your shell would otherwise read as its own", which is more than they do: a name containing a single quote closes them, and the rest of it is read as shell. The skill states that correctly and the manual did not, so the one document a person reads before typing the command was the one making the claim that does not hold.
A crash that is understood is not a crash that stops: an upstream bug waiting on a release, a program that dumps core every time it exits. The diagnosis explains it once and the toast keeps arriving, and the only answer Omarchy had was Crash Capture, which turns off every program's notifications in order to silence one.
The watcher already resolves a name to dedupe on and announces that same name in the toast, so the mute is keyed on it: a flag file under toggles/crash-ignore/, written by the existing omarchy-toggle and read by the existing omarchy-toggle-enabled. One flag per name rather than one list, so `on` mutes, `off` un-mutes, and `ls -A` shows what is muted, with no new file format and nothing to parse. It is the executable's basename wherever one was recorded, falling back to the process name, which the kernel truncates to fifteen characters -- muting the truncated form would match nothing, forever, while looking like it worked.
The name is not always a name, though, and the mute turns it into a path. A program picks its own comm and prctl takes anything, including slashes, and the watcher falls back to comm whenever a crash carries no absolute executable. So it is stripped to its last component first: without that, `a/../bar-off` is a legal comm aimed at an unrelated Omarchy flag, letting a crashing program suppress its own notification and letting a user who accepted the offered mute hide their bar instead. Stripping does not always leave a component either -- `/` leaves an empty string, which is no kind of array subscript and no kind of toast, and `.` or `..` names a directory that omarchy-toggle would touch and report success on, leaving a mute that never matches. Both fall back to `unknown`, the word omarchy-agent-crash already uses for a name it does not have, and which mutes like any other.
The skill offers this at the end of a diagnosis and never runs it unprompted, which makes it the single change a diagnosis may make to a system it otherwise only reads. It tells the agent to use the name it was handed rather than re-derive one, since the watcher resolved that name already and the two agree for ordinary names and not for strange ones; a diagnosis started by hand from `omarchy agent crash <pid>` is given no name and gets the derivation instead. It also says to treat the name as hostile text rather than as a word -- it is whatever the crashed program's author called a file, so a single quote inside one closes the quotes around it and the rest runs as the shell -- and to check the flag arrived rather than assume it.
Co-Authored-By: Codex XHigh <noreply@openai.com>
The name a theme installs under is derived from its repo URL, and holding it to
an allowlist made that allowlist a naming convention nobody had written down. It
was also tighter than the harm it exists to stop: `+` is not shell syntax and a
leading `_` is neither the `..` climb nor the dash basename reads as an option,
so `omarchy-c++-theme` was refused for nothing.
Widen the set to those two and say what it is where a theme author is already
picking a name. The leading character stays out of `.` and `-`, which is the
part that does the work.
Reported-by: Luis Alvarez (lalvarezt)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fd3RCHxwjEbMXoSYB9Aiso
- omarchy-webapp-install wrote all substituted values raw into the
generated .desktop (Name/Comment/Icon/MimeType/Exec), with no escaping.
- Adds two spec-level escapers: Desktop Entry string escaping on every
field (a raw newline could inject a second key line / second Exec=),
and Exec-argument quoting for the default Exec's URL (spaces, %, reserved
chars). $CUSTOM_EXEC stays file-syntax-only — it is a full command line
by design, not a single value.
- No known exploit path: untrusted input reaches these values today only
via Omarchy literals, interactive gum, or direct CLI. This is
defense-in-depth for a latent sink.
- Verified end-to-end: generated .desktop has one escaped key per field;
gio launch passes the URL to omarchy-launch-webapp as a single unchanged
argument (Sunshine literal and normal gum path both byte-identical).
The root rule matched only a file-level root Text, of which this tree has
exactly one. QML inline components are roots for the same reason — the
`text` of `component InfoValue: Text {` comes from every caller, so the
file it lives in never binds it — but they sit inside another element, so
the depth-1 test never saw them. Six went uncovered while the test
reported green, among them the network panel's InfoValue, which callers
bind to the IP address and gateway.
Six more ways to write a Text were read as clean rather than as unreadable:
an opening brace that is not last on its line, a brace on the line after
`Text`, a one-line block containing nested braces, a wrapped binding split
by a comment or a blank line before its `+` (which exempted a dynamic
binding as a literal), and a root Text indented from column zero. Require
the forms a line scanner can read instead of parsing QML; the tree already
writes every Text that way.
Last, a run that read no files reported success. A checkout with no shell/
QML now fails instead, since an all-clear from a scan that opened nothing
is the one answer this test must never give.
Each case is covered by a fixture that fails without its fix.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: OpenAI Codex (gpt-5, xhigh) <noreply@openai.com>
A theme installed from a git repo is named after the repo URL, and that name
becomes its directory name under ~/.config/omarchy/themes. Style > Unlock built
a command line out of the name the picker returned and handed it to
omarchy-launch-floating-terminal-with-presentation, which runs its argument as a
shell string -- so a theme directory called `a';id;'b` ran `id`. Themes are
already held to contributing colour and nothing that executes, which is why
omarchy-theme-set stages no .lua, terminal config, or vscode.json from one.
Hold the derived name to the characters a theme name needs, which stops it from
being dangerous at every place it lands rather than at the one found, and quote
it with printf %q on the way into the action for the names already on disk.
omarchy-theme-remove keeps its existing path-climb guard: its name reaches only
a quoted rm, and the same charset would strand a theme installed before this.
Reported-by: Luis Alvarez (lalvarezt)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011WFcUm5HWFyxaVYdwAeWPP
QQuickStyledText skips the characters between `<` and the tag name with
QChar::isSpace(), which counts U+0085 NEL. JavaScript's `\s` does not, so
isImageTag() read no name at all from a tag written as `<`, U+0085, `img`,
kept it, and Qt then read `img` and issued the GET the stripper exists to
prevent. Measured against Qt 6.11.2 with an offscreen StyledText and a
local HTTP server.
Read the name by skipping everything that is not part of it rather than by
matching the separator, so the two definitions cannot drift apart again.
Over-skipping is the safe direction: it can only classify more runs as
images, and dropping a run never manufactures a tag.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The wildcard granted passwordless root for timedatectl set-timezone plus any trailing arguments, so -H/--host and -M/--machine reached the SSH and machine transports as root. Systemd 261 guards argv injection into ssh, but -H still drives root's SSH client at an attacker-chosen host, and the transport resolves its helper through PATH; only Defaults secure_path stands between that and a planted ssh running as root. Match the argument with an anchored POSIX ERE that admits exactly one timezone token (no whitespace, no leading-dash segment, no traversal component), so no second argument and no option can ever match. The sole caller, omarchy-menu-timezone, passes one list-timezones value and is unaffected.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
The Dell XPS 13 DX13260 drives its two CS35L56 sidecar speaker amplifiers through a quirk that Linux only gains in 7.2, so until Arch ships that kernel the machine plays through one amplifier with no bass. The dell-xps13-sidecar-amps package selects the same driver path with a module override; this installs it on that exact machine and nowhere else.
The detector requires both the DX13260 product name and SKU 0E53, because the override forces a quirk value rather than merging into one, and a machine that gets it wrong loses whatever quirk the kernel would have chosen for itself.
Pacman registers a package even when its post_install scriptlet fails, so the leaf calls dell-xps13-sidecar-amps-apply itself instead of trusting the install to have applied: a failed cleanup or boot-image rebuild has to reach the caller rather than hide behind a package pacman considers installed. That is also why the migration marks reboot-required only after the apply succeeds — a migration that exits non-zero keeps no completion marker and retries the apply on the next run, even though pacman already has the package.
The leaf runs after intel/ptl-kernel.sh rather than beside the other Dell leaf at the top of install/hardware/all.sh, so its boot-image rebuild sees the Panther Lake kernel that step swaps in rather than the stock one it removes.
Co-authored-by: Codex XHigh <codex@openai.com>
Managed policy dirs are enterprise trust roots, so they stay 0755 root:root. The menu path takes root for that one write through a sudoers glob of six hex digits, the same shape as omarchy-dns, and falls back to pkexec where the grant is not installed. Drop omarchy-browser-policy; a group member could plant any JSON, not just a colour.
install -d follows a managed or distribution symlink and would chmod the target. Unlink those paths first, and treat a dangling symlink as a directory the migration still has to repair.
install -d follows a planted ancestor symlink, and a writable parent can rename the managed leaf aside. chromium.theme is user-installed, so only a 0-255 RGB triple becomes a colour.
The hardened gate only looked at the distribution directory. A regular
policies.json planted under the old 777 mode would then be left in place
if the directory later looked 755/root.
Chromium managed policy is mandatory for every profile. World-writable
dirs let any local uid plant policy, including force-installed
extensions. Write goes through the omarchy-browser-policy group at 2775
so theme colour still works without other-write.
The install user is no longer in the docker group by default, so a bare
`once` cannot reach the Docker socket. Run with `sudo` instead.
The script already requires sudo to install the command and enable the
service, so we can safely use it for the initial command launch as well.
* Fix Codex usage collector approval policy
* Capture codex argv with boundaries in the scanner test
The stub joined its arguments with "$*", so the assertion compared one
flattened string and could not tell five arguments from fewer containing
spaces. Passing "-s read-only" and "-a on-request" as single arguments --
which codex rejects as an unexpected argument -- passed the test. NUL
separation and an array comparison keep the boundaries the assertion is
about.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
---------
Co-authored-by: Omabot <omabot@omarchy.org>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
The /tmp-fallback case did check-then-create on a fixed /tmp name, a
TOCTOU/symlink race, and the EXIT trap only cleaned $TMPDIR. Create the decoy
atomically with noclobber (O_EXCL) so it refuses to overwrite an existing file
or follow a symlink at that path, and remove it on exit only when this test
created it. The fixed path is required (it is exactly the path the old code
would form), so a random mktemp name cannot replace it. Addresses the Copilot
review on #8198; the wrapper fix is unchanged.
The cached device path was trusted for merely existing, not for being a hiddev node, and fell back to a predictable /tmp path when XDG_RUNTIME_DIR was unset. Validate the cache shape (hiddev char device) and cache only under the user-private runtime dir; asdcontrol already gates non-Apple devices downstream, so this is defense-in-depth in the layer Omarchy owns.
* [Security] Stop the FIDO2 setup staging its authfile at a predictable /tmp path
pamu2fcfg wrote to /tmp/fido2 and the registration was then moved into place
with `sudo mv`. Any other local user can pre-create /tmp/fido2, and rename(2)
does not dereference the final component, so the privileged move installed the
attacker's symlink itself as pam_u2f's global authfile -- a file consulted by
`sufficient` lines in /etc/pam.d/sudo and /etc/pam.d/polkit-1.
The same move also carried the staged file's ownership into /etc, so on every
install to date /etc/fido2/fido2 is owned by the invoking user at mode 0644.
That needs no attacker: anything running as that uid can add its own credential
and satisfy the machine's sudo prompt without root.
Stage under mktemp and hand the bytes to `install` instead, so the authfile is
always a fresh root-owned regular file rather than an inode a non-root user
still controls. Guard the already-registered check with -L, which -f would
otherwise follow, and reject a symlinked /etc/fido2 in the remove path for the
same reason. A migration takes ownership of authfiles left behind by the old
code; it reports a symlink rather than repairing one, since chown would follow
it and removing it would strip sudo from anyone whose only credential is the
token.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012e6WagC5iUigCGoK1tQeFz
* Install the FIDO2 authfile with -T and refuse a non-regular path
`install SRC DEST` without -T treats an existing directory at DEST as a
destination directory: it drops the credential inside as
/etc/fido2/fido2/tmp.XXXX, exits 0, and setup reports a successful registration
while PAM goes on reading a path that is not a file. -T makes that an error.
The already-registered check has the same blind spot from the other side. -f
follows symlinks, so it reads a symlinked authfile as a registration and leaves
it in place, and is false for a directory, so it tries to register over one.
Only a regular file is a pam_u2f authfile; anything else is now refused with the
same advice to remove it and set FIDO2 up again.
The test deleted every staged path that fell outside its scratch directory,
taking the path from the script under test and already resolved through any
symlink -- so a script staging through one would have named a file of the user's
and had it unlinked. It now unlinks only a file its own stub wrote into.
On a machine that already has /etc/fido2/fido2 the staging assertions cannot run
at all, and the file used to pass without exercising one of them. That branch now
asserts what the host state promises instead: a regular authfile still has to be
recognised as a registration and left alone, and anything else has to be refused.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Replace the FIDO2 authfile inode rather than chowning it in place
Permission is checked at open(2), not at write(2), so a descriptor the
registering user opened on the authfile while it was still theirs stays writable
through chown and chmod alike. pam_u2f resolves /etc/fido2/fido2 to that same
inode, so the repair left the account it authenticates able to append a
credential it controls -- the exact state the migration exists to end, now
recorded as migrated and never revisited.
Installing a fresh root-owned copy and renaming it over the path leaves any such
descriptor writing to a file nothing reads. Credit to #7703, which reached the
same conclusion independently. An interrupted run heals: the staged copy is
root-owned 600 and inert, no marker is written, and the next run replaces it.
A directory or device at the authfile path is no more ours to rewrite than a
symlink is, and chmod 600 on a directory would only make it untraversable, so
both are now reported rather than repaired.
The repair had no test, because it names an absolute path no unprivileged suite
can write. It is exercised through a scratch copy with that one literal
retargeted, rather than by reading the path from the environment: the migration
hands `install` and `mv` root, and an operand the caller can choose is a
privileged write to anywhere. The copy is only as honest as the substitution, so
the test fails if the migration stops naming the path exactly once.
Covered: the no-op on a machine that never registered a key, which must not cost
a password prompt; the repair itself; the new inode; the absence of a staged copy
afterwards; a second account finding it done; and the symlink and non-regular
cases. Each assertion was checked against a mutation that defeats it -- notably a
repair with the right install call, mode, content and cleanup that writes through
the old inode, which only the inode assertion catches.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Finish hardening FIDO2 authfile installation
* Guard the FIDO2 directory and the stage path the setup writes through
install -d follows a symlink at /etc/fido2 and applies the mode and ownership to whatever it points at, so the credential would be staged and published inside the link target and that directory silently reopened to root:root 755. The leaf guard above it only covered fido2 itself, and this is the same threat omarchy-remove-security-fido2 already names on its side.
mktemp's output is an operand for a privileged tee, chmod, mv and rm. The migration validates it before any of them run; the setup did not, so take only the name it asked for there too.
The suite was guarded on the host's own /etc/fido2/fido2 and exited early when one existed, which meant the staging assertions asserted nothing on exactly the machines that use FIDO2. Drive a retargeted copy the way the migration suite already does, so every branch is a fixture and all of them run everywhere.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Stop the FIDO2 migration recording a repair it never made
omarchy-migrate writes the per-user completion marker on any zero exit, so the two states this migration cannot repair got one line in the update terminal and were then silenced for good: no login notice, no re-run, the migration recorded as done having repaired nothing. Those are precisely the machines where the authfile may already be under someone else's control, so raise them through omarchy-notification-send as well, where they outlive the scrollback. Delivery is best-effort: a machine with no user bus or no notification server must not abort the migration and take every later one with it.
The early exit had the same shape of problem. It read the authfile unprivileged, and the old setup created /etc/fido2 with `sudo mkdir -p`, which took the union of the caller's umask and sudoers' 0022 — so registering under `umask 077` left the directory mode 0700 with the user-owned authfile still inside. Absence and "cannot look" are the same answer to those tests, and the migration exited 0 and marked itself complete. Ask root whether a registration is actually behind an untraversable directory before reopening it, so an aborted setup that left an empty directory, or one an administrator keeps private, does not have its mode widened and its group and special bits discarded for a repair it does not need. A machine that never set FIDO2 up has no directory here and still reaches exit 0 without a password prompt.
The notification assertion checks argument shape rather than a substring of the command line. The glyph is a private-use codepoint, and losing it shifts every argument left: -g swallows the headline, the body becomes the title, and the message goes out with no description — which a substring match reads as a pass.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Cover the FIDO2 removal's symlink guard
The -d to -e || -L change is load-bearing for the threat its own comment names — a dangling link at /etc/fido2 that -d reads as absent, left for the next setup to install an authfile through — and it was the one part of this work with no test behind it. Name the directory once so the suite can retarget a copy, the same seam the setup and migration suites use, and assert both halves: the link goes, and the directory it pointed at does not.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
* Stop device names from being executed as Hyprland Lua
Hyprland input-device and monitor names come from USB descriptors and
hyprctl output, so they are attacker-influenceable, yet the toggle and
monitor commands interpolated them straight into hyprctl eval and into
generated Lua that Hyprland re-executes on every reload. The input-device
toggle keys are bound with locked = true, so a malicious USB name reached
Lua code execution from the lock screen; a persisted disable made it run
on every start. This closes that class everywhere it appeared.
- The touchpad/touchscreen disable is now the device name in a plain-text
sidecar file, read back by a packaged Lua module on reload, never a
generated Lua file. hyprctl eval Lua-quotes the name and control
characters are rejected outright.
- Dropped the shipped *-disabled.lua templates so nothing seeds a
disabled state to /etc/skel, making the name file the single source of
truth read from a hardcoded ~/.local/state to match the sibling tools.
- The reload loader excludes those two legacy filenames, so a leftover
generated *-disabled.lua on a not-yet-migrated install can never be
sourced as code again; a migration then recovers the device name from
it and deletes it, sanitizing installs that ran the vulnerable version.
- All four monitor scripts (internal, mirror, clamshell, scaling) now
validate an output name against a plain-connector-name pattern before
writing it as Lua, closing the same latent pattern in the siblings.
- paths.lua treats a set-but-empty XDG_STATE_HOME as unset, matching the
bash side so state is never read from the filesystem root.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0144ZDt44vtxjyF8j9Y88NrM
* Let a failing Lua assertion fail the test
lua discards the status of a chunk read from stdin, so a blown assert printed its traceback and still exited 0: the surrounding `set -euo pipefail` never fired and the following `pass` printed `ok`. Every Lua block in these two files was unenforced, including the assertion that a quoted `hyprctl eval` cannot reach `os.execute` and the negative control that proves the test can detect the injection at all. Passing the chunk as a script argument makes lua report the failure.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Re-apply a recovered input-device disable to the running session
The package hook reloads Hyprland during `omarchy-update-system-pkgs`, before `omarchy-migrate` runs, and at that reload the generated Lua is already excluded while the name file does not exist yet — so a touchpad or touchscreen the user had switched off comes back on, and stays on until their next login. Reload once more once the name has been recovered, which is the same path a login already takes to read it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <codex@openai.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Omarchybot <omabot@omarchy.org>
Co-authored-by: Codex XHigh <codex@openai.com>
* Pin PATH to trusted dirs when omarchy-dns holds root
A dev link prepends a user-writable checkout bin/ to sudo's secure_path,
so the passwordless `omarchy-dns Cloudflare` sudoers rule lets root
resolve a bare helper (dirname, install, tee, nmcli, ...) out of that
checkout — turning checkout-write access into arbitrary root execution.
Pin PATH to trusted system directories once EUID is 0, leaving the
unprivileged wrapper phase free to locate sudo/pkexec on the caller's
PATH.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YUWoHbBoKMjsjV6X3nu1H5
* Assert the trusted-PATH pin is gated on root, not merely present
The EUID assertion matched `(( EUID == 0 ))` anywhere in the file, and require_root has carried that exact test since long before the pin existed. Deleting the pin left the assertion passing, so it stood for nothing: a run with the pin neutered reached the behavioural probe with both greps green. Anchor on the unindented guard and require the pin to be the line it opens, which no other construct in the script satisfies.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Skip the DNS trusted-PATH probe where user namespaces are unavailable
`fail` ends the file, so a sandbox or hardened kernel that refuses unprivileged user namespaces did not just lose the probe — it took the two elevation assertions below it down as well, reporting a product defect where there was only a missing capability. The non-graphical suites are meant to run on any machine and treat a skip as a passing test, the way require_compositor and plugin-add-test.sh already do. Gate the probe on the namespace it needs and say so when it is absent; the static checks above and the elevation checks below run either way.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
* Share the git URL check between theme-install and plugin-add
Both commands clone a URL a stranger can choose, and each carried its own copy of the rule that refuses a git option or a `<helper>::<address>` transport helper before cloning. Two copies of a security check drift: the second one arrived four months after the first, and only because someone went looking for it.
The rule now lives in omarchy-git-url-check and the callers ask it. Its absence refuses the URL rather than waving it through, since the callers read a non-zero status as a refusal and a missing command exits 127.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Refuse a git URL naming a transport Omarchy does not clone from
`<helper>::<address>` is only one of the two ways a URL reaches a remote helper. git also resolves git-remote-<scheme> for `<scheme>://<address>` whenever the scheme is not one it connects itself, so `ext::sh -c id` and `ext://sh -c id` arrive at the same helper while only the first was refused.
That shape cannot be refused outright, because it is also how every legitimate URL arrives, so the scheme is checked against the transports git still connects itself. `git+ssh` and `ssh+git` are on that list: they are spelled like a helper and read as plain ssh, and leaving them off would refuse a URL that clones today. `ext` and `fd` are off it deliberately -- git ships a helper for each, and `ext` runs whatever command the URL carries.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
The new assertion pinned /title:"Windows VM - Omarchy" with the quote sitting
immediately after the colon. That quote is incidental shell syntax, not the
title the Hyprland rule matches on. Open PR #7902 moves the RDP arguments into
an array, where the same flag reads "/title:Windows VM - Omarchy", so the
assertion would fail for whichever of the two pull requests merged second.
Allow an optional quote after the colon. The assertion still fails if the title
itself drifts, in either direction, which is the coupling it exists to protect.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Guard plugin-add against git transport-helper URLs
omarchy-plugin-add cloned a user-supplied git URL without the
transport-helper guard that omarchy-theme-install already applies
(added in #7884, which did not touch plugin-add). Port that guard
(reject ext::/fd:: and leading-dash forms, keep https/ssh/scp-style
incl. IPv6) and add a regression test. Stock systems are unaffected
(git default protocol.ext.allow=never); this removes the silent
dependency on that default and aligns the two install paths.
* Test the plugin-add guard's leading-dash arm via the gum input path
The prior leading-dash cases only exercised the argv option parser, not
the guard (removing the guard's -* arm left them green). Drive a dash
value through the interactive gum prompt under a pty so the post-input
guard is actually covered; skip cleanly where util-linux script is
unavailable.
* Offer to reboot when toggling sudoless Docker; show only the relevant menu entry
Group membership only takes effect on a fresh session, and in practice a logout
or newgrp isn't enough — only a reboot reliably applies it. So the setup/remove
commands now flag the reboot and offer to do it now with a gum confirm (like the
GPU toggle), and the notices say "after a reboot" instead of pointing at logout
or newgrp. The existing-user migration passes OMARCHY_DEFER_REBOOT so it does not
prompt mid-update — omarchy-update-restart still handles the reboot once the whole
update finishes.
The Setup > Security menu also showed Sudoless Docker under both Setup and
Remove. Condition the Setup entry on the group being absent (Remove already
conditions on it being present), so only the applicable one appears.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Gb7x6poap4hGCndPx5qt5T
* Ask omarchy-sudo-docker whether Docker needs sudo
Every place that chooses between talking to Docker directly and elevating was
testing group membership by hand, and the menu guards tested the wrong thing:
they read the running session's groups, which do not change until the reboot,
so after enabling sudoless Docker the menu still offered Setup — the one action
that could no longer do anything — while Remove stayed hidden.
Add omarchy-sudo-docker as the single answer, with the two questions that
actually differ in that window. By default it asks whether this session can
reach the socket, which is what decides if a command must elevate: lazydocker
and the Windows VM keep prompting until the reboot lands. With --configured it
asks whether the account is set up for sudoless Docker, which is what the menu
and the toggles need, so the menu switches to the action that can change state
as soon as the group is written.
Also correct a comment: nothing surfaces reboot-required in the bar; it is
omarchy-update-restart that reads it during a later update.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Gb7x6poap4hGCndPx5qt5T
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* Wait for the keypress ourselves instead of asking gum to
gum 2.0 runs a spun command without the terminal attached, so the
`gum spin -- read -n 1` that held the presentation terminal open returned
at once. Every menu command that ended in a failure took its window down
with it before the error could be read, which is how a failed update
looked like a terminal that just quit.
Read the key directly. gum's own terminal query replies are still sitting
on the tty when the spinner stops, so drain those first or they answer the
prompt on the user's behalf.
The green dot reads better than the globe did, so the provisioning notice
uses it too and drops its spinner along the way.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Ask the terminal for itself before prompting on it
The /dev/tty node is there whether or not a terminal is behind it, so the
existence check passed on a headless run and left both reads failing with
"No such device or address". Open it instead.
Prompt on the terminal too, rather than stdout: a caller that redirects us
was sending the prompt to a file while the read waited on the terminal,
which looks like a hang with no instruction on screen.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Group membership is fixed at login, so removing (or adding) the docker group
does not take effect in the running session. The existing-user migration and the
Setup > Security toggles now call `omarchy-state set reboot-required`, so
omarchy-update-restart prompts for the reboot that actually applies the change
(and the bar shows it pending). A plain log out and back in still works.
The migration test now exercises the real removal command and omarchy-state
rather than a stub, asserting the reboot flag is set on removal and left alone
when the user is already out of the group.
Claude-Session: https://claude.ai/code/session_01Gb7x6poap4hGCndPx5qt5T
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* Don't put the user in the docker group; make it opt-in
The docker group is root-equivalent: anything in it can `docker run -v /:/host`
and rewrite the host as root with no password. On a single-user box that's not
an escalation (the owner is already a wheel/sudo user), but it hands any code
running as the user — a rogue plugin, a poisoned dependency — a silent, headless,
passwordless path to root that sudo's password prompt would otherwise gate.
Stop granting the docker group by default. The daemon still runs (docker.socket);
the Docker TUI and the Windows VM reach it through a polkit prompt, and the plain
`docker` CLI runs under sudo. Sudoless Docker is a warned opt-in via
Setup > Security (omarchy-setup-security-sudoless-docker).
No automatic path may re-grant it: install and first-boot provisioning never
record or apply the group (provisioning also filters a docker line left in an
older factory snapshot), and the Quattro upgrade no longer adds it.
The Windows VM keeps needing the root daemon for a privileged container (KVM,
NET_ADMIN), so it is reworked to run without the group and without becoming a new
way in:
- The compose lives in a root-owned dir and is only written by an elevated,
input-validated writer. A root-invoked bring-up must never consume a file a
user-process could rewrite to bind-mount / into the guest — the old
~/.config/windows compose was exactly that. Volume paths are rebuilt from
$HOME on migration rather than trusted from the (user-writable) legacy file,
path validation rejects traversal, and the privileged sub-action is checked
against an allowlist before dispatch (a slash in it would otherwise run as a
path).
- pkexec elevates a verified root-owned command path, not a PATH-resolved one,
so an authorized prompt can't be redirected to an attacker's binary.
- The guest password is kept in a private 0600 per-user file for RDP instead of
a world-readable compose, and a declined authorization is reported as such,
never as a completed stop.
Existing installs auto-migrate the VM (no redownload) and refresh the stale
Docker launcher entry.
🤖 Generated by Opus 4.8 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <codex@openai.com>
Claude-Session: https://claude.ai/code/session_01Gb7x6poap4hGCndPx5qt5T
* Migrate existing installs off the docker group
The default flip only reaches new installs; existing users keep their docker
group membership and stay exposed. Extend the migration that already refreshes
the Docker launcher to also remove the current user from the group when present,
reusing omarchy-remove-security-sudoless-docker so there is one source of truth
for the change and its notice. It takes effect at next login (the current
session keeps working), and passwordless docker can be turned back on from
Setup > Security > Sudoless Docker.
Migrations run with sudo available — during `omarchy update`, or in the terminal
the pending-migrations notification opens — so the privileged removal does not
prompt at an unattended login. The no-op path (already out of the group) needs
no privilege.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Gb7x6poap4hGCndPx5qt5T
* Refuse symlinked VM mount sources; correct the docker CLI docs
Review follow-ups.
valid_path keeps a traversal string (/./, //, ..) out of the compose, but it is
a string check: a symlink planted at ~/.windows or ~/Windows redirects the
privileged bind mount exactly as traversal would, because docker follows it. So
verify the mount sources as root immediately before bringing the VM up — refuse
a source that is a symlink or resolves through one — which is where the string
check cannot help. A missing source stays fine (docker creates a plain dir).
Also correct the development-tools manual: the CLI is not transparently elevated
(there is no docker wrapper and `d` is still plain docker), so say plainly that
docker on the command line takes `sudo` until sudoless Docker is enabled.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Gb7x6poap4hGCndPx5qt5T
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Codex XHigh <codex@openai.com>
* Add omarchy-ascii for drawing text in the logo font
Renders text as ASCII art in Delta Corps Priest 1, the FIGlet font the Omarchy wordmark itself is drawn in, so branding art can be words rather than a picture. The font is embedded in the script and the layout is done in awk, so the command adds nothing to the default package set.
The layout runs on one-byte stand-ins for the five block characters the font draws with. Column arithmetic over the characters themselves counts bytes in one locale and characters in another, and the stand-ins keep length() and substr() counting columns either way.
Delta Corps Priest 1 carries letters and spaces only, and every mirror of it ships the same file with the digit and punctuation glyphs empty. Anything else is dropped and named on stderr, and text with nothing drawable at all exits 1 rather than printing silence.
🤖 Generated by Opus 5 in Claude Code.
* Correct what the renderer did with input it could not draw
The route never ran on piped text. The metadata declared `<text...>` as required, so `omarchy ascii` with nothing on the command line resolved to the router's help while `omarchy-ascii` run directly worked, which is why the tests missed it: they all called the binary. The argument is optional now, and a test goes through the route.
Text reached awk as a command-line variable, where awk reads backslash escapes of its own, so `omarchy ascii 'A\nB'` drew two blocks instead of naming the backslash as a character the font lacks. A text longer than the argument list could not be passed at all. It arrives as awk's input now, with the font on a descriptor of its own.
A line with nothing drawable printed nothing at all, so a blank line between two words closed the gap up rather than keeping it. Every line draws its block now, blank ones included, which is what figlet does with a newline.
Placing a glyph scanned and copied the whole width of the art so far, costing the square of the line's length: four thousand characters took forty-six seconds. A row is now held without its trailing blanks, counted separately instead, so a glyph costs its own width and those four thousand characters take a tenth of a second.
A skipped control character was named on stderr by writing it out, which sends it to the terminal as a control character; those are named by code now. An unknown option was drawn as art rather than refused, so a mistyped `--width 40` quietly rendered the word "width".
figlet.c trims the column of blanks that every row of an `M` shares when it is the first glyph on a line, and figlet.js keeps it. asciiart.eu runs figlet.js, so the rendering follows figlet.js and a test pins that `M`, because the wordmark alone does not catch the difference and the next reader would have no way to tell the choice from an oversight.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Codex XHigh <codex@openai.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Codex XHigh <codex@openai.com>
Nothing in the repository called omarchy-sudo-reset, and its one line interpolated an environment-supplied $USER into a string handed to a root shell: `su -c "faillock --reset --user $USER"`. $USER is an environment variable rather than a kernel-supplied identity, so whatever set it before the command ran chose the rest of what root's shell executed. That is not a way past PAM on its own — su still has to authenticate — but the installer sets root's password to the user's own, so the prompt this raises is one the user answers by habit.
It bought little for that. Omarchy sets `deny=10 unlock_time=120` in /etc/pam.d/system-auth and in the lock screen's PAM stack, so a lockout takes ten wrong passwords to reach and clears itself two minutes later, and the manual documents the root-TTY reset for anyone who would rather not wait.
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
* Stop a psmouse quirk from failing every install
install/hardware/fix-synaptic-touchpad.sh calls modprobe against the running
kernel. Under 4.0 the only thing that runs it is the ISO finalizer, inside
arch-chroot, where uname -r still names the live ISO's kernel while /lib/modules
holds the target's. The live ISO always boots linux-t2 and the configurator
gives every machine that is not a T2 Mac stock linux, so those two never match:
modprobe exits 1 with "Module psmouse not found in directory /lib/modules/<live
kernel>". run_logged returns that status and omarchy-apply-hardware runs under
set -euo pipefail, so a fresh install stops on the first machine with a device
named "synaptics" and no psmouse loaded -- reported from a ThinkPad in #6985,
but nothing about it is Lenovo-specific.
Skip the load when the running kernel's modules are not reachable, and warn
instead of failing when modprobe declines for any other reason. An optional
touchpad improvement should never be able to halt an install.
This does not make InterTouch reach the installed system: a module loaded into
the live kernel is gone at reboot, so on 4.0 this script has never applied
anything to an installed machine. Persisting it means writing options psmouse
synaptics_intertouch=1 to /etc/modprobe.d, which forces the SMBus transport past
the kernel's own allowlist on any touchpad merely named "synaptics" in
/proc/bus/input/devices. That is a hardware-behaviour change on a wide class of
machines, so it is left for a maintainer to decide separately.
Fixes#6985
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Ask modprobe whether psmouse resolves rather than guessing at /lib/modules
The reachability check ran through OMARCHY_SYNAPTIC_MODULES_DIR, an override
modprobe itself never saw: it decided whether the load was attempted but could
not change where modprobe looked, so the guard and the load consulted different
places and the seam read as though it configured module lookup. modprobe -qn
answers the same question directly -- it resolves psmouse against the running
kernel without loading it -- so the guard and the load now agree by
construction and the override goes away. The arch-chroot case that broke
installs is still skipped silently, for the same reason it always was: the live
kernel's modules are not the ones on disk.
Inline the remaining /proc/bus/input/devices override at its only use. These
leaves are sourced one after another into a single shell, so a variable left at
the top level outlives the script that set it.
Pin the wiring assertion to the run_logged call instead of any mention of the
path. A commented-out line satisfied the old grep, so the test could pass with
the quirk no longer running at all.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013L5zwTiZ2CsgazyxXBiPa8
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
omarchy-mise-install gained --quiet on its `mise use -g` line so a wrapper
no longer prints mise's "tools: pkg@version" ahead of the tool's own
output. That only changes wrappers written from then on, and the migration
that installed the current ones is already marked complete, so every
wrapper already on disk keeps polluting stdout: `claude --version` still
answers with two lines, and a wrapper for a protocol-speaking command
answers with a line its caller cannot parse.
Rewrite them through omarchy-mise-install so the template stays in one
place. Wrappers were written by four different generated forms over time
and only the ones a later migration happened to regenerate moved forward,
so all four are recognized. The whole file has to match one of them
exactly, which leaves a wrapper someone has added a line to alone and
makes a second run a no-op.
Claude-Session: https://claude.ai/code/session_01Mv1FyKG2VGZRdtknQ7YjU8
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The app name becomes a filename, and omarchy-webapp-install ran
`mkdir -p "$(dirname "$DESKTOP_FILE")"` over it, so every slash turned into a
directory level. Typing a URL into the Name field -- the reported way in --
wrote the launcher to
`~/.local/share/applications/http:/127.0.0.1:4000/.desktop`.
Removal could then never reach it. The picker lists the file but displays a
name derived from the path, and the removal rebuilt a flat
`$DESKTOP_DIR/$APP_NAME.desktop` from that name, so `rm -f` deleted nothing and
the app stayed in the launcher with no error.
Refuse a name containing a slash rather than silently renaming what the user
typed, and delete the file the scan actually found instead of a path rebuilt
from its display name. The second half also clears up whatever earlier versions
nested, which a reconstructed path cannot address.
The direct-Notify rewrite dropped notify-send options that callers rely on, which
the momus review caught: omarchy-display-text-size uses -r/-p to refresh one
toast in place, and the acceptance suite uses the --expire-time=15000 equals
form. Re-add -r/--replace-id (replaces_id), -p/--print-id (emit the returned id),
and the --flag=value form for every long option; a dash-leading description like
"-50% off" is now kept as body text rather than erroring, and --exec "" is
rejected.
Add a fixture proving the deliberate upgrade behavior: a popup persisted by a
pre-upgrade shell with a legacy `exec` shell string restores with an inert click
(execArgv empty, the old string never run) rather than executing it.
omarchy-notification-send now calls org.freedesktop.Notifications.Notify
directly with `busctl --user`, instead of shelling out to notify-send. Each
value is one typed D-Bus parameter, so there is no argv/option layer that could
reinterpret a relayed headline like `--hint=…` or `-rf` as an option or a hint:
the summary and body are strings, and omarchy-exec-argv is built only from
--exec. A leading `--` keeps busctl's own getopt from reading a dash-leading
value as a busctl option.
Map -i to app_icon, -t to expire_timeout, and urgency to the byte hint; unknown
options are now a hard error rather than a silent pass-through. Route the unused
hypr o.notify helper and the sample hooks through the wrapper too, and tighten
the bin-style test so nothing under bin/ may call notify-send. The test stubs
busctl and trips if notify-send is invoked.
The comment claimed the validation kept a hostile hint from reaching a shell,
but it is purely structural: a well-formed ["bash","-c",…] passes. Say so,
and point at the separate sender-trust boundary.
* Let a received Taildrop file wait to be answered
A delivery can land hours after it was sent, and the toast announcing it was expiring after five seconds -- so a file that arrived while nobody was at the machine was gone from the screen before anyone could click it open. Critical urgency is what the shell reads as a popup that lives until it is clicked or dismissed, the same thing omarchy-crash-watch uses to keep its click-to-diagnose toast around.
The wrapper takes options after the headline and description as well as before, which is how this argument list is built. That path had no test, and it fails quietly rather than loudly: the wrapper appends its own default urgency last, so an urgency it stopped parsing would reach notify-send as `-u critical ... -u low` and the toast would go back to expiring.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Say it in the commit message, not above the code
`-u critical` next to a line that builds a notification says what it does, and the five lines explaining why it is there were a recap of the change rather than something the code could not say. The reasoning stays where it belongs, in the commit that made the change.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
The argv click command closed injection through the hint's value, but the
sender still handed the headline and description to notify-send bare. A
value beginning with a dash is parsed there as flags, and one shaped like
`--hint=string:omarchy-exec-argv:[...]` is read as a hint of its own --
libnotify keys hints in a hash table, so the later of two replaces the
earlier and a forged headline outranks the vector --exec built.
That is reachable without any --exec in sight: omarchy-tailscale-send
passes a single file's basename verbatim as the description, so a file
named like the hint gives its click action to whoever chose the name.
Put the headline and description behind a `--` so notify-send reads them
as text, and refuse any pass-through word carrying omarchy-exec-argv --
--exec is the only thing that may build a click command.
Co-Authored-By: Codex XHigh <noreply@openai.com>
A greedy pre-scan matched the first --exec token anywhere on the line, so an
untrusted headline or description literally equal to "--exec" could be mistaken
for the delimiter. It always failed closed (the following token became a
dash-option argv[0] that parseExecArgv rejects), but it was a latent footgun
surfaced by an adversarial review.
Detect --exec only in the trailing-option loop, after the headline/description
positionals are captured. A headline that is the string "--exec" is now kept as
text and the real trailing --exec still wins. The residual — a description
exactly equal to "--exec" losing its click action — is an inherent, harmless CLI
ambiguity for a value identical to the delimiter.
Replace --exec-arg with an ergonomic --exec that consumes the rest of the line
as the click command. The caller's shell tokenizes the words into discrete
arguments before the tool sees them, and the shell runs them as positional
parameters (never a re-parsed string), so safety is identical to the argv form
while the call sites read naturally: `--exec omarchy toggle something`.
Crucially the tool never splits a string itself — a single quoted whole-command
argument is rejected and points at the unquoted form, because whitespace-
splitting a string hands argument boundaries to whoever controls its content
(the injection we are avoiding). --exec must come last; migrate every caller.
A free-form shell-string --exec sitting next to the safe --exec-arg is a
standing invitation for the next caller to interpolate untrusted data and
reintroduce the RCE. Remove it: omarchy-notification-send --exec now errors and
points at --exec-arg, and the shell drops the omarchy-exec string hint and its
bash -lc execution path, leaving only the argv path.
Migrate the remaining string callers (the first-run invitation hooks, wifi and
welcome prompts) to --exec-arg, and update their notification mocks. Trim the
verbose security comments added along the way.
Quickshell.execDetached(argv) ran the click target with only the shell
process's stripped environment, so GUI actions like the screenshot editor
(tensaku-edit) — resolved on the login-shell PATH the old `bash -lc` string
exec provided — stopped launching on click.
Run the argv through `bash -lc 'exec "$@"'` instead: the script text is a
constant and the arguments are passed as positional parameters, which bash
expands without re-tokenizing or re-evaluating, so injection safety is intact
while PATH and session env match the old behavior exactly.
The click action of a notification was a free-form shell string run through
`bash -lc`, safe only when every sender shell-quoted every interpolated value
perfectly. One slip is RCE: a hostile yt-dlp video title forged an output
record and injected an mpv option into the click command (mehmetince.net RCE,
partially addressed by #7847).
Add a parameterized transport: omarchy-notification-send gains --exec-arg
(repeatable), encoding a JSON argv into the omarchy-exec-argv hint. The shell
runs it with Quickshell.execDetached(argv) and no shell, so data an attacker
controls is only ever one argument and can never be reparsed as a command. The
shell fails closed on a malformed argv hint.
The legacy free-form --exec string is retained but honored only from Omarchy's
own omarchy-action toasts, and deprecated. Migrate all in-repo callers
(screenshot, screen recording, taildrop receive, migrate-notify, crash-watch,
yt-dlp host) to --exec-arg. Update docs and tests.
* Stop an installed theme from shipping code
`omarchy theme install <url>` clones a stranger's git repository into ~/.config/omarchy/themes, and omarchy-theme-set then copied that whole directory into the staged theme. Most of the files in a staged theme are code rather than colour: Hyprland requires hyprland.lua and gum_env.lua from it at login, Neovim loads neovim.lua at startup, and alacritty.toml, kitty.conf, foot.ini and ghostty.conf each name the program the terminal launches. Installing a theme was the same act as running its author's code, and nothing on disk distinguishes an installed theme from one the user wrote.
Stage only what a theme needs in order to be a theme: colors.toml, light.mode, the preview and unlock images, and image files under backgrounds/. Everything else is ignored, named on stderr, and generated from default/themed/*.tpl instead. Symlinks are never followed, because in an untrusted theme they point wherever the author chose. A theme older than colors.toml keeps its palette: its alacritty.toml is read for colours in a scratch directory and only the resulting colors.toml is staged, so the terminal config never lands.
The filter belongs in omarchy-theme-set rather than in omarchy-theme-install because staging is the choke point. It also covers themes installed before this change, themes copied in by hand, and files a theme gains later through `omarchy theme update`.
First-party themes under $OMARCHY_PATH/themes are unaffected. Per-theme overrides of a generated file are no longer available to user themes; the template at ~/.config/omarchy/themed/<file>.tpl replaces that, and icons.theme is the one setting with no replacement.
🤖 Generated by Opus 5 in Claude Code.
* Stop a theme URL or name being read as an option or a path
Three paths in the theme commands took an attacker-shaped string straight into git, into basename, or into rm.
`git clone "$REPO_URL"` passes the URL as the first positional argument, so a URL beginning with a dash is parsed as an option instead and the destination path becomes what git tries to clone. Pass `--` before the URL so a URL is always a URL. git also treats `<helper>::<address>` as a remote helper to run; git's own protocol.allow default already refuses `ext::`, so rejecting that shape here is a second line rather than the fix, and it keeps holding if that default ever moves. The helper name is a bare word at the very start of the URL, which is what the guard matches: an scp-style IPv6 host such as git@[2001:db8::1]:org/repo.git carries `::` of its own and still clones.
`basename "$REPO_PATH" .git` has the same problem one step later, after the scp-style prefix has been stripped: `host:-s/foo.git` leaves basename reading `-s` as an option and returning `.git` as the theme name. Take the name with `--`.
That name is then joined into a path that is about to be `rm -rf`'d, so a repo whose basename came out as `..` would take ~/.config/omarchy with it. omarchy-theme-remove had the same shape from its own argument, and omarchy-theme-set's sed/tr normalization does not stop a name containing a slash. Reject empty, anything starting with a dot, and anything containing `/` in all three, before the name reaches a path.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Codex XHigh <codex@openai.com>
* Re-stage the current theme for installs that already applied one
Dropping a theme's code at staging time only takes effect the next time a theme is staged. An install that already applied an extra theme keeps that theme's hyprland.lua, gum_env.lua, neovim.lua and terminal configs in ~/.local/state/omarchy/current/theme, which Hyprland requires at login and the terminals include at launch, and nothing forces a theme change — so for those installs the fix would arrive whenever the user next happened to switch themes, which may be never.
Re-stage once through omarchy-theme-refresh. First-party themes stage identically, so the cost for everyone else is a single retint during an update they are already running.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Codex XHigh <codex@openai.com>
* Stop a theme's unlock image republishing a file it points at
omarchy-plymouth-set-by-theme reads unlock.png straight out of ~/.config/omarchy/themes, which is an installed theme's own directory and outside the staging filter, and hands the path to omarchy-plymouth-set. That path was copied twice into world-readable /usr/share — once by the user into the Plymouth theme, and once by `sudo cp` into the SDDM theme. A symlink there was followed both times, so a theme could name a file it cannot read and have root publish it.
Refuse a symlinked logo, and copy the staged logo to SDDM instead of rereading the caller's path as root. The staged copy is made by the user, so nothing privileged opens a path the caller chose.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Codex XHigh <codex@openai.com>
* Limit only what an installed theme could run
Two corrections to the rule this branch introduced, both narrowing it to what it was actually for.
It applied to every theme under ~/.config/omarchy/themes, which swept up themes the user wrote themselves. Their machine, their file: a theme they wrote is theirs to fill however they like, and Omarchy's own themes were never in scope. Only a theme that came from someone else needs limiting, and the repo already knows which those are — omarchy-theme-extras calls a theme with a `.git` directory an extra and a symlink someone's working copy, because that is what `omarchy theme install` leaves behind when it clones. Use the same test.
It was also an allowlist, which dropped files that carry nothing but colour and left theme authors worse off for no gain. Drop only what can run: any `*.lua`, since Hyprland requires a theme's hyprland.lua and gum_env.lua at login and Neovim loads neovim.lua at startup; the four terminal configs, since each names the program the terminal launches; and vscode.json, whose extension field reaches `code --install-extension` and a VS Code extension is arbitrary JavaScript. Everything else an installed theme ships is kept, so btop.theme, chromium.theme, helix.toml, icons.theme, keyboard.rgb and shell.toml go back to being the theme's to set.
Symlinks are still dropped, now at any depth rather than only where an allowlist happened to look.
A denylist is wrong the moment someone adds a template and does not think about it, so the decision is forced rather than remembered: the test fails on any default/themed/*.tpl whose output is recorded as neither code nor colour, and a new terminal or a new Lua-loading editor cannot be added without classifying it.
What this does not cover, and is written down in docs/theming.md rather than implied: a theme shipped as an archive and unpacked by hand looks exactly like one the user wrote. `omarchy theme install` only takes git URLs, so the supported path is always filtered, but this marks where a theme came from and is not a sandbox.
🤖 Generated by Opus 5 in Claude Code.
* Fix what the review found
Four things, all confirmed against the source before changing anything.
The migration failed permanently when the active theme had been removed. `omarchy theme remove` deletes the directory without repointing theme.name, so the name survives, the staged copy survives, and omarchy-theme-refresh exits 1 because neither source directory exists — leaving the migration pending forever and the stale staged Lua exactly where it was, which is the one thing it existed to remove. Seed the default theme in that case: there is nothing to re-stage from, and the removal should have left a working theme behind anyway.
The staging test skipped the strict-mode header that docs/testing.md makes the contract for every shell test. Adding it means the patterns that fail on purpose have to stop being bare `cmd && fail` compounds, which errexit reads as the script itself failing; the mutations were re-run afterwards to confirm the assertions still fire rather than the run dying early and looking like something else.
The guards in omarchy-theme-install and omarchy-theme-remove had no coverage — they were checked by hand and left that way. theme-install-guards-test.sh stubs git and the themes directory and proves an option-shaped URL, a transport helper, and a name that would climb out all stop before git or rm runs, that a dash inside the path no longer becomes a basename option, and that an ordinary URL still clones and applies.
The new docs/theming.md prose was hard-wrapped, which AGENTS.md forbids for docs/. Unwrapped. The rest of that file is wrapped from before and is left alone rather than churned through this change.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh and Copilot.
---------
Co-authored-by: Codex XHigh <codex@openai.com>
* Stop a video title from becoming the Download Video play command
The host parsed yt-dlp's after_move line as title plus path, so a newline in page metadata could forge the path. Clicking the toast then handed that value to mpv as options. Print only the real file, ignore anything that is not inside the download dir, and invoke mpv with --.
* Refuse downloads whose video title contains control characters
The hoodie page still offered a real hidden clip, so yt-dlp saved it even after the play-action fix. A title with newlines is not a legitimate name; abort before the download and tell the user it was refused.
* Test the forged record in the order yt-dlp emits it
The records ran forged-first and good-last, so the assertion measured recovery after
bad records rather than preservation of an already-captured path when a forged record
arrives afterwards. That is the shape a hostile title actually produces, because a
title ending in a newline closes its own record and leaves the genuine path on a line
the loop ignores. As written the assertion passed with resolve_download_file replaced
by a no-op.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Resolve the download path without dropping a trailing newline
Command substitution strips trailing newlines, so an in-directory symlink pointing at
a regular file whose name ends in one canonicalised to a different path -- which may
itself exist -- and that path then passed the containment check and reached ffmpeg and
the click command. Reading realpath's NUL-terminated output keeps the name intact, and
a resolved path carrying a control character is refused outright.
Not reachable through a yt-dlp download, since --restrict-filenames strips control
characters from the name it writes; it is the helper's contract that was wrong.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Accept a download directory that resolves to /
realpath returns "/" for the root directory, which made the containment pattern "//*"
and rejected every file saved directly under it, so the host reported a failed download
after saving the file.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Stop refusing a download because its title has control characters
The gate cannot tell a hostile title from a legitimate one. --print emits one line per
extracted video and --no-playlist does not collapse a multi_video result, so a page
holding two clips arrives as two titles separated by a newline and is refused exactly
like a forged record would be.
It also guaranteed nothing it was read as guaranteeing. The simulate run and the
download run are separate fetches, so a site is free to answer them differently, and
the check never constrained the metadata the download actually used.
What stands between a record and the click command is resolve_download_file, which is
untouched here. Leaving a check that refuses valid pages while securing nothing invites
the path validation to be relaxed later on the strength of it. A gate that would work
is possible -- --print '%(title)j' encodes each title as JSON on its own line, which
separates a newline in the metadata from a newline between videos -- but it belongs
with a use for the title rather than as a bare refusal.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Disarm the legacy exec-before-download hook too
--no-exec clears the modern --exec map but leaves --exec-before-download stored
separately, and yt-dlp restores it as a before_dl postprocessor, so a hook configured
in the user's yt-dlp config still ran during the download this host drives.
The accompanying test runs download_url itself against stubbed tools. Everything else
in this file exercises the helpers in isolation, which left the invocation uncovered:
restoring the title to the record template, or dropping --no-exec or the trailing --,
passed every assertion here.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Toast the page title again instead of the saved filename
Deriving the toast text from the sanitised filename cost the title it was meant to
show: "My Great Clip" arrived as "My_Great_Clip [My_Great_Clip]". The title is safe as
notification text -- it is an argv element, never part of a command -- so the only
question was getting it out of yt-dlp without reopening the record forgery.
It now comes from the download run, so it describes the file that was actually saved,
and it is printed as %(title)j. JSON-encoding is what makes that safe: a newline or tab
in page metadata becomes an escape sequence inside one quoted string rather than a
record boundary, so a title can no longer split itself across lines. The decoder keeps
only what precedes the first control character, refuses anything notify-send would read
as an option, and leaves the filename-derived title as the fallback when a page offers
nothing usable.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Name the saved file after the page title
The download landed as "My_Great_Clip [My_Great_Clip].mp4" when the page called it
"My Great Clip". --restrict-filenames was carrying more weight than it earns here: it
folds spaces to underscores and strips non-ASCII, which is what mangles the name, and
it is not what keeps the record stream safe. yt-dlp removes control characters from a
filename either way -- a newline becomes a space, tabs and DEL and NUL are dropped --
so a path printed after the move is still only ever one line, which is the property
resolve_download_file depends on.
Dropping the [%(id)s] suffix is the other half of matching the title, and it trades
away the uniqueness that suffix bought: two videos sharing a title now share a name,
and yt-dlp skips a download whose file already exists, so the second one toasts as a
failure. Restoring the suffix is a one-line change if that trade is the wrong way
round.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Omabot <omabot@omarchy.org>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
Ori is OpenRouter's harness: `ori claude`, `ori codex` and `ori opencode` start those agents against OpenRouter's model catalogue, and `ori code` is Ori's own agent. That last one is what the default-agent entry launches, bare — Ori has no approval prompt to skip, so there is no "don't stop to ask" flag to pass it the way the other agents get one.
The package is `github:OpenRouterLabs/ori-releases`, because upstream ships prebuilt binaries as release assets and publishes nothing to npm. mise's `github` backend picks the right asset per platform and verifies GitHub's artifact attestations on the way in; `ubi` resolves the same release but is deprecated for removal in mise 2027.1.
The menu glyph at U+E909 is OpenRouter's own mark. Ori publishes no logo of its own and its product page renders that one, so there was no Ori-specific mark to prefer over it.
Co-authored-by: Codex XHigh <noreply@openai.com>
* Ask the terminal whether the layout scrolled
Whether the logo is on the rows the frames address was decided by arithmetic: count the lines of a second fastfetch run, and animate if the window has one more. That holds only while the second run and the real one agree, and they need not — a module that appears between them, a wrap the measurement never sees, anything that makes the painted layout a row taller than the counted one. The screen then scrolls, the logo sits a row above where these frames draw it, and the animation paints a copy of the logo one row below the real one.
fastfetch has just painted when the question is asked, so the terminal already knows the answer: an unscrolled layout leaves the cursor one row past its last line. Ask for it, and keep the arithmetic for a terminal that will not say.
The reply has to be read without a command substitution around it, or the query goes into the substitution's own pipe instead of to the terminal, and what comes back is the escape rather than a row.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Find the logo in the render instead of assuming where it is
The row and column the frames draw on came from the padding in this repo's fastfetch config. The config that runs is the one in /etc, and `omarchy dev link` says plainly that it does not replace files at fixed system paths — so the two can disagree, and nothing in the code would notice. A logo fastfetch drew on row 1 or row 5, redrawn on row 3, is a logo that jumps the moment the animation starts and stays where the animation put it.
fastfetch's own output is already being captured to measure the layout's height, so the logo can be found in it: take the longest line of the file as a landmark, look for it in the render, and turn where it was found back into where the logo starts. Padding of any size lands correctly, and a render that does not contain the file's own text is not this logo at all — a config that restyled it, a placeholder that was substituted — so there is nothing to animate and it is left alone.
That last part subsumes what the config-path check was guessing at. It is kept, because it also decides whether the window may be fitted, which happens before there is a terminal to measure anything in.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Take the cursor query back out
Asking the terminal where the cursor came to rest was meant to see a scrolled layout rather than infer one. It cannot: a linefeed at the bottom margin scrolls the screen and leaves the cursor on the bottom row, so a render that fitted exactly and one that scrolled both answer with the same number. Measured on a real terminal — a 29-line layout in windows of 30, 29 and 28 rows answers 30, 29, 28. The guard therefore only ever agreed with the arithmetic it replaced, and only in the case the arithmetic already had right.
What it did add was a read on the same stdin the keypress that closes About arrives on. A key pressed while the query is outstanding is swallowed; a reply that arrives after the timeout is read by the next tick as the keypress, and the window closes on its own; typed text of the right shape is accepted as an answer, because nothing checked for the introducer; and the query is written before the read turns echo off, so a fast reply can be echoed onto the screen.
None of that buys anything the line count did not already give, so it goes.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Fit the About window with a little room to spare
The window was sized to exactly what was measured: the columns the widest module line needed, and one row past the layout's last line for the cursor. That measurement is taken once, and then the content goes on living. An uptime turns minutes into hours and hours into days, a version string grows, a module shows up after the next boot — and a window with nothing in hand clips at the right edge or scrolls the top padding away the moment any of it happens. A scrolled layout also moves the logo off the row it was drawn on, which is where a shifting logo comes from.
Two columns and a row, which is invisible on screen and enough for anything that ticks over. The remembered size in about.fit is keyed on the logo, so the first launch after this still opens at the old size and the fit corrects it there and then.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
* Animate the About logo with a passing glint
fastfetch has no animation of its own, so the sweep is ours. Every frame is composed once up front and a tick writes one of them, repainting only the cells fastfetch drew the logo on, which holds 40fps for under 1% of a core and never reaches the module column six columns to its right.
The logo is handed back exactly as it arrived: frames carry the colour fastfetch drew it in rather than a plain reset, so between glints the window is byte-for-byte the one it was before this change. Where the logo on screen might not be the text in the file, it stays still rather than guess — a fastfetch config in any directory searched ahead of Omarchy's own, a logo built from $1 colour placeholders or from a tab fastfetch expands itself, a window too small for the layout to fit without scrolling, and a shell whose locale is counting bytes instead of characters.
A resize moves those cells, so a WINCH trap ends the sweep on the frame it happens rather than up to a poll later, which measured 10ms against 830ms. The polling stays as the backstop for a signal that arrived while it could not be taken.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Give the sheen a file of its own
The launcher was carrying two things: what a glint looks like, and what it is drawn over. Only the second is about fastfetch. bin/omarchy-logo-sheen now knows how to lean a band of light across an ASCII logo and nothing about About, and it is handed where the logo sits, what colour to give its cells back, and how much room it has. When a frame may be written stays with the launcher, because that is inseparable from how the window closes and resizes.
The tests split along the same seam, and the sheen's half no longer strips the launcher's tail to reach the code it tests.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Leave a logo still when one character is not one cell
The frames slice the logo by character and the terminal draws it by column, so the sheen only puts a row back where it found it while those two agree. A double-width glyph, a combining mark and a joined emoji each break that, and a cut through one of them renders as something else — pushing the rest of the row right, into the module column, with nothing on screen to say the logo moved. A user can put any of the three in the logo by editing the text.
Comparing the row's character count against its width in columns is the assumption itself, so it is what gets checked, rather than a list of the glyph classes that happen to break it today.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Size the About window from the layout fastfetch drew
The fit predicted the content height from the logo and the module column, taking the taller of the two. Once the logo is the taller one fastfetch writes a row more than that arithmetic expects, so every logo of 27 rows or more got a window one row short and scrolled its top padding away. The shipped logo is 26 rows, which is why it never showed.
Ask fastfetch how tall the layout came out instead of predicting it. That measurement was already being taken for the sheen, which refuses to animate a layout that scrolled — so a tall logo used to lose the glint as well as the padding row.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Leave the logo still when NO_COLOR asks for none
fastfetch drops the logo's colour when it writes to a terminal that set NO_COLOR, but not when it writes to the substitution that measures it. The colour to hand those cells back in was therefore measured as green while the logo on screen had none, and the first glint would have left the whole logo green. It stops suppressing autowrap there too, so a narrow window soft-wraps and scrolls where the measurement cannot see it and the frames go on addressing rows that moved.
Both follow from animating a screen drawn in a mode the measurement did not reproduce. A glint is colour besides, which is the thing NO_COLOR asks for none of.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Read fastfetch's config paths whole, and do not read silence as an answer
Splitting each listed path on whitespace dropped everything after the first space, so a config under a home directory containing one was missed and the fit and the sheen went ahead against a layout neither had measured. The marker fastfetch prints beside the config it settled on is not part of the path either.
A listing that failed was also indistinguishable from one that found nothing, because the status of a process substitution is discarded, so a fastfetch that could not enumerate read as "no custom config". Fall back to the directory fastfetch has always looked in first rather than take silence for an answer.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Notice a resize that lands while the grid is being read
The sweep read the WINCH flag before the grid rather than after it. A signal arriving during the two command substitutions that read the grid is delivered only once they finish, so the flag was still false when it was tested, the grid still compared equal, and one more frame went out at coordinates the resize had already moved. Read the flag last, after the check it could have landed during.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Check that the render loop is what plays the sheen
Every builder the tests drive can be exercised while nothing on screen ever animates. Replacing the render loop's animated branch with the still one it replaced left all of them passing, so nothing was holding the animation onto the screen it belongs on.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Name the animation after the branding it animates
omarchy-logo-sheen sat in a group of its own, and a group whose only command is hidden still gets a header printed for it with nothing underneath. It belongs beside omarchy-branding-about, whose art it animates: the group already exists, the two halves of About branding are next to each other, and `omarchy branding about` still routes to the command rather than the helper now sharing its prefix.
The tests take the names of what they cover, the way the rest of the suite does.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Fewer moving parts in the sheen
Four questions about a logo turned out to be one. Whether the shell is counting characters or bytes, whether a tab or an escape is in the line, whether a glyph is double-width or a combining mark or a joined emoji — each was asking whether one character is one cell, so that is the only thing asked now. It is also less strict in the one case that deserves it: plain ASCII art animates in a C locale, where the locale probe used to refuse everything.
One band of light instead of a white core inside a green halo, which is three colour spans a row rather than five and no helper to clamp four cuts with. Only the left cut needs clamping at all: a slice starting past the end of a line is already empty, while a negative offset would count from the end of it.
One loop instead of two, because an empty frame list plays nothing. A logo that cannot be animated now waits in the loop the animated one rests in rather than in a second copy of it, and the build that failed leaves no frames behind, since that loop plays whatever it finds.
The logo's colour comes out of one match rather than a loop eating one escape at a time.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Say in the manual which art the glint leaves alone
The line promised the glint to any art in the file, and the code does not: art whose characters are not one column wide is left still on purpose, as is the whole screen for anyone keeping a fastfetch config of their own. Somebody reading the old line and seeing a still logo would take deliberate behaviour for a broken feature.
Name the condition rather than the list of glyph classes behind it, and say that both modes of Set From Image meet it, because that is the path almost everyone is on.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Draw the band in a colour bold cannot collapse
A terminal can be told to render bold text in a brighter colour — foot's bold-text-in-bright, whose palette-based form brightens a bold regular colour into its bright counterpart. fastfetch draws the logo bold green, so under that setting the logo is already bright green, which is the colour the band was using: the glint came out the same green as the art and nothing appeared to happen at all. None of the four terminal configs set it, so it was waiting on whoever turned it on.
Bright white instead, because no regular colour brightens into it, and the band shows whatever the terminal does with bold. Narrower with it, since white against green needs less width to read than a second green did.
🤖 Generated by Opus 5 in Claude Code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
* Add Super + Q as a second chord for closing a window
Super + W stays the documented default. Super + Q is the chord people
arrive with from macOS, where Command + Q quits the app, and typing it
into Omarchy did nothing at all until now.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
* Put an action's alternative chord on one keybindings row
Super + W and Super + Q both read "Close window" in the menu, two rows
apart, with nothing to say they were the same thing -- and the
alternative sorted above the default. The scratchpad and the calculator
had the same trouble, each bound to a chord and to a second key.
Four actions are named as having an alternative, one at a time, and the
second chord joins the first one's row. A rule would be wrong here: Alt +
Tab and Shift + Alt + Tab both say "Reveal active window on top" while
cycling opposite ways, and a media key is nobody's idea of an alternative
to a Super chord. Both halves still have to agree on what they dispatch,
since a label is only what a chord is called, and an unresolved
dispatcher never counts as agreement.
Nothing is allowed past the 35-character column: a pair that would
overrun it stays as two rows rather than pushing its arrow out of line.
The menu elides a row that outgrows its card -- 754px of label, 78
monospace characters at the heading size -- and the longest entry already
sits at 74, so widening the column to fit the widest pair would have cost
two dozen rows the end of their description.
Priority ordering reads the rendered row, so the chord sharing it would
otherwise reclassify the entry: XF86Calculator alone belongs in the tail
kept for media keys, and it took the calculator down there with it.
Ranking now reads the chord that leads the row.
The key left of 1 reads as ~ rather than Hyprland's name for it, whether
a bind names it or reports the keycode for the keymap to resolve. Cached
records predate all of this, so the cache version moves with it.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-authored-by: Codex XHigh <noreply@openai.com>
---------
Co-authored-by: Codex XHigh <noreply@openai.com>
* Only offer Update > Extra Themes when there is one
omarchy-theme-update pulls the themes under ~/.config/omarchy/themes that came from a git clone, so on a machine that has never installed one by hand the row opens a terminal that prints nothing and closes. Guard it with the same predicates the command itself applies, since a row that shows over a symlinked theme or a worktree's `.git` file is the same dead end in a narrower shape, and pin the two to each other in the guard test.
Co-Authored-By: Codex XHigh <noreply@openai.com>
* Extract the Extra Themes guard into omarchy-theme-extras
The row's `when:` and omarchy-theme-update each carried their own idea of which themes came from a git clone, and the two only matched because a test held them together. Name it once instead: omarchy-theme-extras lists those directories and exits nonzero when there are none, so the row asks exactly the command its action runs. Living in a script also puts the glob out of reach of whatever shopt a login shell left set for the guard batch.
Co-Authored-By: Codex XHigh <noreply@openai.com>
---------
Co-authored-by: Codex XHigh <noreply@openai.com>
* Switch back to the packaged quickshell now that 0.3.1 kills synchronously
Omarchy shipped the quickshell-git build for a single fix: 0.3.0's `kill` returned before the instance had exited, so the kill loop in omarchy-restart-shell could race a dying shell. Upstream 0.3.1 ships that fix, which makes extra/quickshell the better package to be on again — signed, versioned, and not rebuilt from a moving branch on every update.
The migration swaps unconditionally instead of first checking which version the mirror offers. A machine left holding quickshell-git while the shipped package list names quickshell has no way to reconcile the two: omarchy-reinstall-pkgs installs that list with --needed, which does not skip a name that is not installed, and the conflict it then walks into has no answer under --noconfirm. A mirror that is briefly behind installs 0.3.0 instead and the next upgrade carries it to 0.3.1, which is much the cheaper way to be wrong.
🤖 Generated by Opus 5 in Claude Code. Reviewed by Codex XHigh.
Co-Authored-By: Codex XHigh <codex@openai.com>
* Drop the quickshell version note from the shell restart loop
The comment qualified the kill loop as needing 0.3.1 or newer, but omarchy-restart-shell ships in the same package upgrade that brings quickshell along, so a machine running this code already has the version the loop depends on. The caveat could never be false where it was read, which left it as version archaeology rather than something the code could not say for itself.
🤖 Generated by Opus 5 in Claude Code.
---------
Co-authored-by: Codex XHigh <codex@openai.com>
With the default scale = "auto", sync_internal_scale read the config,
rejected "auto" as non-numeric, fell back to the hardcoded default 2,
and force-applied it whenever the compositor's auto resolution differed.
Since the script runs from omarchy-system-wake after every idle cycle,
the panel flapped between 2 and auto's own value (1.5666667 on a 198 DPI
panel) on every wake/reload pair.
A config without a usable number -- "auto", or an expression only
Hyprland's Lua can evaluate -- delegates the scale to the compositor:
whatever it resolved for the enabled panel is the configured scale, so
there is nothing to correct. Recovery of a disabled panel is unchanged
and still re-enables it with the remembered scale, falling back to the
historical default 2.
Fixes#7265. Also the scale-revert half of #7301.
Claude-Session: https://claude.ai/code/session_01L4Z6GimYhR1Kpsir24VAPF
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* Replace Gemini coding agent with Antigravity
* Potential fix for pull request finding
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* Remove the dead Gemini mise wrapper in the Antigravity migration
Remove Preinstalls no longer lists gemini, so the wrapper Omarchy created
would have stayed in ~/.local/bin with nothing left to clean it up.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Install Antigravity when it is the default a Gemini user is migrated onto
The opt-out check skipped the install but the rewrite ran anyway, so anyone
who had removed the preinstalls was left with a default agent naming a
command that is not there.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Fix Antigravity skill provisioning and Gemini wrapper migration
- Wires Omarchy's default skills into Antigravity by linking them to ~/.gemini/config/skills/ in bin/omarchy-provision-user and migrations/1786719479.sh.
- Fixes the Gemini wrapper migration in migrations/1786719479.sh to recognize and remove wrappers containing either `mise use -g "gemini"` or `mise use -g --quiet "gemini"`, while leaving hand-written wrappers intact.
- Adds regression tests for both skill provisioning and wrapper removal in test/shell.d/default-agent-test.sh and test/shell.d/provision-user-test.sh.
* Stop the provisioning test from retheming the session it runs in
The test ran the real omarchy-provision-user, which sources install/user/all.sh and so reached omarchy-theme-set: hyprctl reload against the live compositor, gsettings against the live desktop, and a global Node install, none of which the skill symlinks it asserts need. Its mocks for omarchy-done and omarchy-refresh-applications were shadowed anyway, because provisioning prepends $OMARCHY_PATH/bin ahead of them, so stubbing the install suite at its own path is what a mock cannot do here. The exit status is checked rather than discarded: the assertion held even when provisioning died outright, because the symlinks are made twenty lines before the suite runs.
* Match the Gemini default and wrapper the way Omarchy writes them
The migration decided both questions differently from the code that owns them. It read the default agent with grep -qxF, while omarchy-default-agent takes the first line through read, so a padded " gemini " that the launcher still resolves was left naming an agent the launcher no longer supports. The wrapper it deletes was matched anywhere in the file, so a hand-written one that only mentions the installer's line in a comment went with Omarchy's own. Reading it the launcher's way and anchoring the match settles both against whoever wrote the file. The skills loop guards its glob the way migrations/1786539345.sh does, so an empty source cannot leave a symlink named "*" behind a migration already marked complete.
Co-Authored-By: Codex XHigh <noreply@anthropic.com>
* List Antigravity among the skill directories
The manual named Claude Code, Codex, Pi and the generic location; provisioning now links ~/.gemini/config/skills too.
Co-Authored-By: Codex XHigh <noreply@anthropic.com>
---------
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Omabot <omabot@omarchy.org>
* Add a clock format with live seconds
Right-clicking the clock now reaches "Thursday 09:39:23" and its AM/PM twin, and the widget's SystemClock ticks once a second only while a format that prints seconds is showing — every other format keeps the minute precision it had, so nobody pays for a repaint a second to read a label that changes once a minute.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Read an unterminated literal in a clock format as text
Qt reads an opening quote with no closing one as a literal running to the end of the format, so "HH:mm 'sec" prints "09:39 sec" and never a second count — but the seconds test stripped only balanced quotes, saw the s, and put the widget on a per-second tick for a label that changes once a minute. The wiring assertions went the other way: each passed while the feature was broken, so hard-coding showsSeconds to false, dropping the label's onDateChanged, or commenting the precision line out and leaving the text behind all shipped green. Comments now come out of the source before it is matched, and both halves of the tick are asserted.
Co-Authored-By: Codex XHigh <noreply@openai.com>
---------
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: Codex XHigh <noreply@openai.com>
The shell already shows a "Launching Windows…" OSD from the moment the desktop entry is activated until the RDP window appears, so the notification duplicated feedback the user is already looking at. The failure notification stays: nothing else reports a VM that never came up.
* Use (( )) for the numeric argument test
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Drop the quotes on a variable inside [[ ]]
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Use omarchy-pkg-drop instead of raw pacman -Rns
omarchy-pkg-drop already filters to installed packages, so the
2>/dev/null || true suppression is no longer needed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Drop defensive checks around default-set commands
ttfx, imagemagick, and networkmanager are all in the default package
set, so their commands are runtime invariants and should be invoked
directly. Removing the nmcli guard also removes the degraded wifi
fallthrough that only ran when nmcli was missing.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* Add Remove > AI for the apps Install > AI offers
Install > AI grew five entries and Remove grew one: Dictation, sitting on its own at the top level. Everything else installed from that menu had to come back out through Remove > Package by name.
Mirror the install tree instead. Dictation moves under the new AI submenu where its installer already lives, and ChatGPT Desktop, Grok Bot, LM Studio, Ollama and T3 Code get removers beside it. Each entry is conditional on the thing actually being installed, so the submenu only ever lists what is there.
What each remover deletes was read off a machine that had all five installed and launched, not guessed. That matters most for T3 Code, which bootstraps the agents it drives: ~/.claude.json, ~/.grok, ~/.npm and ~/.local/share/opencode all appear the first time it runs, and all of them outlive it, so it takes only ~/.config/t3code and ~/.t3. Grok Bot is the same trap in miniature -- ~/.grokbot is its own, ~/.grok belongs to the Grok CLI.
Ollama drops every acceleration variant rather than the one the installer happened to pick, and disables the service before the package, since that is what holds the models open.
🤖 Generated by Opus 5 in Claude Code.
* Remove only what these apps own, and only where removal works
Three defects from an independent review of the previous commit.
ChatGPT Desktop was deleting ~/.cache/codex-runtimes, which belongs to the Codex CLI rather than the desktop app: the `codex` binary resolves its runtime and plugins out of that directory, and it ships in a package this remover does not touch. Removing the desktop app took the CLI's prepared runtime with it, leaving a separate, still-installed tool to rebuild it -- and unable to, offline.
Ollama's row appeared whenever the `ollama` command existed, but omarchy-pkg-drop removes exact package names. With ollama-bin, ollama-git or a hand-built binary the entry offered a removal it could not perform: the service went down, /var/lib/ollama and ~/.ollama were deleted, and it reported success with the program still installed. Every acceleration variant depends on the base package, so testing for that package covers each one the installer can produce and nothing it cannot remove.
LM Studio keeps its models under a relocatable home, and ~/.lmstudio-home-pointer is the only record of where they went. The remover deleted the pointer and the default path, so a user who had moved their models kept every one of them while being told they were gone. Read the pointer before deleting it, and refuse one aimed at / or at the home directory itself, since following it there would take everything.
Co-Authored-By: Codex XHigh <noreply@openai.com>
---------
Co-authored-by: Codex XHigh <noreply@openai.com>
t3code-bin is in the Omarchy repo now, so the menu can offer it the way it offers Cursor and Grok Bot: install the package, then launch the desktop entry it ships.
The mark is a trace rather than a download. T3 publishes no monochrome SVG — the app icon is a black rounded tile with the letters knocked out of it, and a tile flattens to a solid square once the menu recolors every path with the theme foreground. Tracing the lettermark out of that icon keeps the silhouette that actually reads.
The font is package-owned, so the glyph reaches a desktop through an omarchy-settings release rather than omarchy update. Until that release lands, a pulled checkout draws the entry with no icon.
🤖 Generated by Opus 5 in Claude Code.
grant_covers re-implemented etc/sudoers.d/omarchy-dns in bash -- one of
the three providers, and %wheel -- but never asked whether the rule was
installed. It ships in the etc/ tree that omarchy-settings copies, so
every machine still on an older settings package answers yes to a grant
it does not have. require_root then execs into sudo with no way back,
and the panel's one-click toggle dies on a password prompt it has no
terminal to show.
Ask sudo instead. `sudo -l` alone reports whether a command is
permitted, which the blanket %wheel rule answers yes to for everything,
but the long listing prints the matched entry's tags -- !authenticate is
the grant and nothing else. It runs nothing, and under -n it prompts for
nothing, so a machine without the rule falls through to polkit and gets
a prompt on screen.
The provider list and the wheel check go away with it; sudo owns that
policy now, and it stays right if the rule is ever edited or removed.
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
* Switch DNS providers without a password prompt
The network panel and the menu run omarchy-dns from a process with no
terminal, so require_root reached for pkexec and put a polkit password
prompt in front of what is meant to be a one-click toggle.
Grant %wheel passwordless sudo for the three stock providers and take
that path whenever the grant covers the invocation. Custom stays out of
the grant: it points the machine at servers the caller supplies, and it
already runs in a terminal that can ask.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Pick the elevation path without asking sudo
The `sudo -n -l` probe answered the wrong question. It reports whether a
command is permitted, not whether it is passwordless, and the %wheel rule
every Omarchy install ships permits everything -- `sudo -n -l /usr/bin/rm
-rf /tmp/x` exits 0. So the probe passed for Custom too, and the exec
below it ran `sudo -n`, which fails outright with no terminal and no way
back to pkexec.
Decide from what the sudoers rule actually says instead: sudo when there
is a terminal to type into, or when the resolved path and the provider
are both ones the rule names. Everything else keeps going through polkit.
Pin a root-owned PATH once elevated, too. `omarchy dev link` puts a
user-writable checkout ahead of sudo's secure_path for every command, so
a passwordless grant on a script that resolves nmcli, tee, and install
through PATH would otherwise hand root to whoever can write there.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Keep users outside %wheel on the polkit path
The rule grants %wheel, so path and provider alone do not mean sudo will
take it. A user outside the group was sent to sudo anyway, and with no
terminal to answer the prompt that is a dead end -- polkit at least
offers to authenticate as somebody else.
Two holes in the test alongside it: it accepted any file containing the
expected rule, so a second, argument-free line would have widened the
grant unnoticed, and run as root it would have sailed past the stubs and
rewritten the host's own DNS config.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Elevate the system install, whatever copy was invoked
The rule names /usr/bin/omarchy-dns, so a dev-linked checkout handed sudo
a path nothing could match and fell back to a polkit prompt. Re-exec the
packaged path instead: the privileged half is the system install
everywhere, the grant matches everywhere, and the path comparison and the
PATH pinning that existed to work around the checkout both go away.
Dev-linked checkouts run their own unprivileged half and the installed
one as root, which is the trade for not carrying a second code path.
---------
Co-authored-by: Omabot <david@hey.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
* Decode webp in the shell
The background and the lock screen are drawn by Quickshell, so they
decode through Qt, which ships handlers for png, jpeg and gif but not
webp. QImageReader answers "Unsupported image format" and the layer
comes up blank. Any third-party theme shipping a .webp background hits
this today, even though every path that goes looking for a background
already globs the extension.
qt6-imageformats supplies the missing plugin for 71 KB downloaded. Its
one new dependency of substance, libwebp, is already on every machine
by way of libvips.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Store theme backgrounds as webp
WebP codes both of the things these backgrounds are made of better than
the formats they were in: the photographs, where its lossy mode is worth
a third or more over JPEG at matched quality, and the flat art and dot
patterns, where its lossless mode undercuts an oxipng-packed PNG.
28 of them become lossless webp and decode bit-for-bit identically
(AE=0), so the dot patterns and flat-shaded pieces carry no quality
question at all. That includes 0-launch, whose alpha channel comes
through intact. The other 51 are photographs held to the same 38 dB
PSNR floor as the JPEG pass, landing between 38.0 and 54.6 dB. Every
image keeps its exact pixel dimensions, for 29.8 MB.
Each one is encoded from the original as it stands in quattro rather
than from the file the earlier commits produced, so nothing picks up a
second generation of loss on the way here.
13 stay JPEG. WebP is plainly larger for most of them, and three are
grainy enough that its filter smooths the grain instead of coding it:
PSNR plateaus near 34 dB however high the quality goes, well under the
floor.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
* Print the OEM Windows product key from firmware
Machines that shipped with Windows keep the OEM key in the ACPI MSDM
table. `omarchy windows license key` reads it with strings, then cat.
* Rename the firmware key command to omarchy-windows-key
* Open the scratchpad with the default agent already in it
on_created_empty fires when the special workspace is created empty, so the
agent starts the first time the console drops down instead of at boot, and
comes back on the next open if you close it.
The exec rule pins the workspace rather than trusting the spawn to inherit
it: Hyprland only tags a process with its origin workspace while
misc.initial_workspace_tracking is on, and we turn that off.
Nothing to guard for a missing default agent. Omarchy picks none for you,
and omarchy-agent exits without opening a window when none is set, so the
scratchpad just opens empty until one is chosen.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Move the console into its own file and size it to half the screen
The scratchpad's presentation was scattered through looknfeel: a dim in the
decoration block, a workspace rule below it, two animation leaves further
down again. Gathered into qconsole.lua, where the whole console is one
readable thing.
Sized to half the screen while it moved. A window rule cannot do that: its
size expressions resolve once, when the window maps, so rescaling the
monitor afterwards leaves a console that is no longer half of anything.
Gaps are re-applied by the layout, so the console is sized by the gap left
underneath it, recomputed from the monitor whenever the layout changes.
Monitor dimensions come back in physical pixels while gaps are logical, so
the scale comes out before the reserved area comes off. That arithmetic is
the whole trick, and the test pins it at 1x, 2x and 1.5x.
The test runs lua with an explicit "-". Bare `lua <<EOF` reads stdin as a
REPL and exits 0 even after an error, which would leave its assertions
unable to fail.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Only rewrite the console rule when its size actually changes
Refitting replaces the rule in place rather than stacking a new one, so
there was no leak, but each write still schedules a monitor and window state
refresh and monitor.focused fires on every hop between screens. Remember
what was last written and skip the write when the number has not moved.
Also say out loud that the scale guard is what keeps the arithmetic below it
safe: a monitor handle that has outlived its output answers nil to every
field, and a layout change is exactly when that happens.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Drop the active window border inside the console
The gradient border marks which window has focus, which the console does not
need: it is only ever focused while it is open, and the dimmed workspace
behind it already sets it apart. On a single agent terminal the highlight
just reads as a frame around the panel.
no_border on the workspace rule pins the border to 0 at workspace-rule
priority, so it applies to whatever ends up in there without touching the
global border.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Omabot <david@hey.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Give scratchpad a Quake-style presentation
* Keep the Quake scratchpad from restyling every window
The presentation was bought with global decoration defaults: rounding went
0 -> 5 and both shadow and blur were switched back on for every window on
the system, undoing 935283c8. Blur in particular is close to invisible
here, since every window is already tagged 0.985/0.96 opacity and no layer
rule asks for it, so it was GPU load on every frame for almost nothing.
Put the globals back and scope the rounding to the scratchpad with a window
rule, the same way popped windows already get theirs. The inset, dim_special
and the slide carry the effect on their own. dim_special only applies while
a special workspace is open, so it stays.
Also drop the four workspace-rule booleans that only restated Hyprland's
defaults, fold the stale scratchpad rows in the hotkey table into the new
ones, and give the binding assertions their own fixture instead of borrowing
the bar-panel one.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Omabot <david@hey.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Re-encode over-encoded theme backgrounds
Several backgrounds shipped at quality 95-100, which buys nothing visible
on a wallpaper but costs 20 MB. Re-encode those at quality 85, leaving
resolution untouched -- every file keeps its original pixel dimensions.
Only files stored above quality 90 are touched, and only when the result
saves at least 15% and stays above 38 dB PSNR against the original. The
34 backgrounds already stored at quality 85 or below are left alone
rather than pushed through another lossy generation for a few hundred KB.
osaka-jade/2-shaded-entrance is skipped for that reason: nothing clears
both bars. ristretto/2-coffee-beans is re-encoded at 90 instead of 85,
where 85 fell below the PSNR floor.
Theme backgrounds drop from 107 MB to 87 MB.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Store nord's night-hawks background as a palette PNG
The file uses only 8 distinct colors but was saved as truecolor RGB,
costing 1.8 MB for an image an indexed palette stores in half the space.
Converting to a palette PNG is lossless: same dimensions, and zero pixels
differ from the original.
1782 KB -> 909 KB.
The other palette-eligible PNGs are already indexed, and re-compressing
them with ImageMagick only makes them bigger, so they are left alone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Recompress theme PNGs losslessly with oxipng
Every background PNG was left at its authoring tool's default deflate
settings. Running oxipng over them re-packs the same pixels: all 31 files
decode bit-for-bit identically (AE=0) at unchanged resolution, for 981 KB.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Store photographic backgrounds as JPEG instead of PNG
Four backgrounds were shipped as PNG despite being photographs and
painterly illustrations, where lossless coding buys nothing the eye can
see. Re-encoding them at the same q85 used by the other backgrounds keeps
every pixel dimension and stays above 40 dB PSNR, for 3.3 MB.
The remaining PNGs stay PNG: JPEG is larger for the dot patterns and the
flat-shaded pieces, and 0-launch.png genuinely uses its alpha channel.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
* Switch to the mise-bin package
mise-bin carries mise's own release artifacts from the Omarchy repo --
PGO+BOLT-optimized on x86_64, glibc-native on both arches -- instead of
Arch's mise, and tracks jdx/mise releases directly.
Existing installs need a migration because the two packages conflict, and
omarchy-pkg-add cannot make the swap: pacman answers its own conflict
question with No under --noconfirm and fails the transaction. --ask=4
answers that one question, so mise-bin replaces mise in a single
transaction -- which is also what keeps omarchy-zsh and omarchy-fish, both
of which depend on mise, satisfied through the swap by its provides.
* Guard the swap with a conditional instead of an early exit
Two-path control flow takes an if, per the style guide; the early exit only made the swap line unreachable from a distance.
* Add --gh-keys so sshd setup can run without prompts
Grabbing keys from GitHub was reachable only through the interactive menu: pick
"Grab key from GitHub", then type the username into a second prompt. So the one
path that needs no secret pasted around was also the one path a script could not
take, and setting a machine up over ssh or from a provisioning run meant falling
back to --key with a key copied by hand.
--gh-keys <username> takes the same path the prompt did. The fetch and authorize
logic is unchanged and now shared, with the prompt reduced to asking for the
username and handing it over.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Reject a missing --gh-keys username before setting anything up
The username was only checked for being absent entirely, and only after the
server was installed and the firewall opened. So `--gh-keys=` with an unset
variable behind it configured the machine and then dropped into the interactive
menu, and `--gh-keys --help` took --help as the username and set the server up
on its way to failing the fetch -- a help flag that changes the system.
Check the value where it is parsed, and reject one that is empty or shaped like
an option.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Launch claude and codex agents with auto-review instead of full bypass
Claude's auto permission mode and codex's --approve-for-me both run
unattended without prompting, but keep automatic review (and codex's
workspace-write sandbox) instead of skipping approval entirely. Grok
stays on bypassPermissions.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Match the bash aliases to the agent launcher's auto-review modes
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
The dial already grouped its digits for the locale above 10, but below 10 it
went through toFixed, which hardcodes a dot. A German desktop therefore read
9.5 and 1.235 off the same dial, switching decimal convention halfway up the
scale.
Send both branches through the locale. A reading is a measurement rather than
interface text, so its separators follow the system's number conventions even
though the interface itself stays English.
English output is unchanged.
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
The calendar grid's header row and the week-start toggle label were the only
text in the shell that followed the system locale, so a German desktop drew
MO DI MI over an interface that is English everywhere else. Nothing chose that;
they were the only two places reading day names off Qt.locale().
Take them from en_US instead. Where the week starts still follows the locale:
that is a regional convention rather than a translation, and it stays
overridable through weekStartDay.
Dropping the trailing-period strip with it, since that existed only for the
locales this no longer renders.
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
omarchy-plugin-clone only takes the source id as the first argument, but the
menu passed --edit ahead of it, so the id fell through to the unknown-option
branch and every clone from Setup > Plugins failed.
Closes#6913
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Bar widgets propagate their composed press-and-hold down to the center gesture
area without handing over the grab, so the gesture area started a bar move and
then received neither a release nor a cancel to end it. The move ghost stayed on
screen for the rest of the session. Ignore the gesture unless we hold the press.
Closes#6881
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
mise use -g announces the resolved tool on stdout, so every wrapped command
prepended a "tools:" line to its own output. That corrupts anything speaking a
protocol over stdout, such as codex app-server. Pass --quiet, which keeps errors
on stderr and preserves the exit status.
The obsolete-wrapper check in the agent migration matched the generated command
verbatim, so loosen it to match the package instead of the flags.
Closes#6908
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
A 3.8.3 migration appended the section header with one backslash too many, so
sed wrote the literal characters instead of a newline plus the header. foot
rejects the line and stops reading the rest of the file. The later text-binding
migration matches the header with grep -qxF, misses the broken line, and appends
a second section, leaving the config broken across the Quattro upgrade.
Closes#6903
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
A whitelisted bare git repo over $HOME driven only by constrained
omarchy dots commands: local snapshot history at batch boundaries plus
squash-published state sync across machines. Design survived adversarial
review; rejected alternatives and rationale recorded in the plan.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Every reference doc was audited claim-by-claim against the code.
file-layout and omarchy-shell were the most decayed (renamed commands,
the etc/ overrides source split, dead IPC entry points and example keys);
update-process lagged the recent pipeline changes and gains a channels
section; theming and audio-tuning were accurate but thin around their
lifecycles.
New reference docs for the subsystems that had none: the menu system,
the CLI router, the notification daemon, and the non-acceptance test
architecture. AGENTS.md links the two of those agents will need most.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Providers never returned JSON rows: they are shell-defined row sources
emitting tab-delimited lines, and extensions cannot declare new names.
bar.shellQuote moved to Util.qml, and the UpperCamelCase widget id
migration no longer exists.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Task procedure lives in agents/skills/ (migrations.md moves there),
system-shape reference in docs/ (AUDIO-TUNING.md renamed to match),
end-user documentation in manual/. AGENTS.md now states the split.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Hyprland reaps its own children, so os.execute() gets ECHILD from waitpid and
never sees an exit status. Every call reported failure, which meant the NVIDIA
env detection in hypr/nvidia.lua never set NVD_BACKEND, LIBVA_DRIVER_NAME or
__GLX_VENDOR_LIBRARY_NAME. Read a marker off stdout instead.
Closes#6914
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Wait for the current Windows boot before connecting RDP
docker logs retains output across stop/start, so grepping the whole log
matched "Windows started successfully" from an earlier boot and returned
immediately, firing xfreerdp3 while the guest was still booting. Anchor the
scan to the container's current StartedAt, and run it even when the container
was already running, since the image restarts the guest in place on reboot.
* Skip Kerberos when connecting to the Windows VM
FreeRDP 3 attempts Kerberos before NTLM for NLA, and Arch's stock
/etc/krb5.conf declares default_realm = ATHENA.MIT.EDU, so every launch tries
to reach MIT's KDC. Off the network each attempt blocks ~23s and xfreerdp3
sits in CLOSE-WAIT without drawing a window, which reads as the VM failing to
start. Point FreeRDP at a realm-less krb5 config so it falls through to NTLM,
which is what the local Windows account uses anyway.
* Re-read the container start time on every readiness poll
A failed docker inspect left STARTED_AT empty, and docker logs drops the
--since filter when it is, putting the scan back on the whole retained log
and its stale success line. Sampling per poll also keeps the window on the
current boot if the container restarts mid-wait.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Install rows hid themselves with `when:"! <present>"`, so software you
already had vanished from the very list it was installed from. Add a
`disabled:` guard that keeps a row listed but dim, ✓-marked, unselectable
and out of search, and move every Install row onto it.
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* Install missing apps when choosing defaults
* Restore Chromium through browser installer
* Trust default app installer status
* Use full conditionals for install paths
* Host the manual's images in the repo under manual/images
Replaces all learn.omacom.io/manual.omakub.org hotlinks with local webp
files capped at 1600px wide (~20MB total), so the manual is fully
self-contained. Theme and unlock previews are converted from the
canonical themes/*/preview.png files. A handful of illustration shots
are interim conversions of the current images, pending retakes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Retake the illustration screenshots fresh at 3x scaling
Navigation, clipboard history, notices, tmux layouts, prompt, and About
reshot on a clean workspace at 3.13x (1920x1080 logical on 6K) across
ten themes: Tokyo Night, Catppuccin, Gruvbox, Kanagawa, Everforest,
Nord, Osaka Jade, Rose Pine, Matte Black, and Ristretto.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Drop the extra themes chapter from the manual
The community theme gallery will live elsewhere. Removes the chapter
and its 114 gallery images, renumbers the following chapters, and
repoints the two references (theme installs still work via
Install > Style > Theme in the menu).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Reference the shipped theme previews instead of copying them
The themes chapter now points straight at themes/*/preview.png and
preview-unlock.png, so previews can never drift from the source and
manual/images drops from 6.3MB to 3.5MB.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* Restore preinstalls from the menu, and drop the Omacom apps with them
Remove Preinstalls missed omacut, omacalc, and omawrite, so the three Omacom
apps survived an opt-out that was supposed to clear the desk.
Opting out was also one-way. Install > Preinstalls now puts everything back:
the shipped .desktop launchers and mise stubs via omarchy-refresh-applications,
the dropped packages via pacman, and the opt-out marker deleted so the
preinstalled keybindings return on reload. The two menu entries guard on the
marker, so exactly one of them is ever visible.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Trim the preinstall lists to what quattro actually ships
Remove Preinstalls was still dropping typora, spotify, 1password, 1password-cli,
signal-desktop, opencode, claude-code, and github-cli. None of those are in
omarchy-base.packages anymore: typora gave way to omawrite, the services moved
to on-demand menu installs, and the agent CLIs are mise-managed. Removing them
took out apps the user had deliberately installed, and restoring them would have
put back what we no longer ship.
Both lists are now the same twelve packages, all of them in omarchy-base.packages.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Keep the opt-out marker when a restore fails
omarchy-pkg-add exits non-zero when pacman cannot install a package, but the
restore ran straight past it, cleared the marker, and reloaded Hyprland. That
reported success and brought back keybindings for apps that never arrived. The
marker now falls last, behind a check on the transaction.
The new test also pins the two lists to each other and to omarchy-base.packages,
which is the drift that let retired packages linger in the removal list.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Pacman answers its own conflict question with No under --noconfirm, so one
retired package can stop every update after it. Which package to drop is a
decision rather than a cleanup, so run the upgrade again with pacman asking
when there is a terminal to answer on, and report instead when -y promised
not to ask.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The migration asked a user to close every running browser before repairing
the Copy URL shortcut, but a browser only ever rewrites its own Preferences
on exit. Waiting on all browsers deadlocks `omarchy update` for anyone whose
main browser is effectively never closed: the pending ghosts commonly sit in
a stale profile nobody has open, yet the migration blocks on the always-open
browser until the prompt is declined, failing the whole update.
A running Chromium-family browser holds a SingletonLock (and socket) inside
its user-data-dir, so whether the profile being repaired is open is
mechanical. Gate on that instead of on the sheer presence of a browser
process — the repair proceeds where the affected profile is closed, and
browsers attached to other profiles no longer hold the update hostage.
The gate stays conservative while an affected profile actually is open, and
the existing post-repair verification still catches a browser that starts
mid-repair and restores stale Preferences on exit.
The migration test now simulates an open profile with its SingletonLock
instead of a pgrep stub; every prior scenario still passes.
LocalSend registers an Ayatana item with no ItemIsMenu and no Activate
handler, so its primary click is a silent no-op and the menu offers only
Open and Quit. Share > Receive already opens it, so drop the item the way
Dropbox's is dropped when its dedicated widget owns the surface.
Closes#6838
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
gum draws its confirm UI on stderr, so the migration's `2>/dev/null` threw
away the whole prompt while gum still held the terminal in raw mode reading
keys. With a browser open, an update stopped after "Running migration
(1786643346)" on an unpainted screen with no way to tell it was waiting for
an answer.
Nothing else in the repo suppresses gum's stderr; the redirect only ever hid
gum's own error in the no-terminal case, where the migration already explains
itself on stderr before deferring.
Fixes#6841
Claude-Session: https://claude.ai/code/session_01So4yKzTDdJJp7orjnF8i1J
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The switchers chapter shifted everything from chapter 3 up by one, and
the TOC was still on the old numbers. Add its missing entry too.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The network and bluetooth panels stopped launching them; the module
catalogue still said otherwise.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Add "Coming From Mac or Windows" as chapter 3: a translation layer for
Cmd/Win instincts rather than a feature tour.
Fill the remaining coverage gaps: fcitx5 input methods, laptop speaker
tuning, touchpad haptics, crash capture in the toggle table, the
installable font list, and the ChatGPT Desktop and Grok Bot entries.
Correct claims that drifted from the source: the network and bluetooth
bar icons no longer launch nmtui/bluetoothctl, monitor scaling steps
rather than cycles, and the display widget's scroll steps brightness.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Rebind ghost Copy URL shortcut registrations to the pinned id
Chromium never hands a suggested shortcut to one extension while
another — even a long-gone one — still holds the registration. Profiles
that first loaded Copy URL before its id was pinned registered
Alt+Shift+L under an id derived from the extension's load path at the
time, so the pinned extension never receives the shortcut and the
keypress does nothing (#6816).
The quattro upgrade tried to repair this against one hardcoded
path-derived id, which only ever matched a single home directory. The
historical ids are unknowable in general — they hash long-gone absolute
paths through whatever symlinks existed then — but the registration
itself names the command, so a migration now rebinds any copy-url
command that points away from the pinned id, unless that id belongs to
an extension that is actually installed or the pinned extension already
holds a binding of its own.
Browsers rewrite Preferences on exit, which reverts any repair made
while one runs, so the migration asks for this user's browser windows to
be closed first — failing and staying pending when there is no terminal
to ask in or the prompt is declined. The backup a repair leaves behind
marks it as attempted but unverified: until a browser-free run confirms
the registration stayed repaired, the migration keeps itself pending
rather than trusting a disk state an open browser may still overwrite.
The upgrade-time repair is dropped: the upgrade already runs migrations,
so the migration is the single implementation.
Fixes#6816
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Pin the WhatsApp Slim extension id
Keyless unpacked extensions get path-derived ids, which go stale if the
load path or packaging ever changes — the same class of bug that broke
the Copy URL shortcut for pre-package installs. Pin the id with a
manifest key like the other bundled extensions.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Crash capture stays on by default, but Trigger > Toggle > Crash Capture (or
`omarchy toggle crash-capture`) now turns the watcher off. The toggle writes the
usual flag file and stops the unit for this session; the unit checks the same
flag with ConditionPathExists, so the choice survives a logout without the unit
having to be disabled.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The blank timer was gated on `authenticating`, which is
`authenticatingPassword || fingerprintAuthenticating`. The fingerprint
PAM sits armed for the entire lock waiting for a finger, so on any
machine with a reader enrolled the gate is true from lock until unlock:
the timer is stopped when the lock begins and never re-armed, and the
display stays lit indefinitely.
Gate on `authenticatingPassword` instead. A password check in flight
still holds the display up, and the passive fingerprint wait no longer
does.
Claude-Session: https://claude.ai/code/session_01EDpyC9793TKZBS2jXUNECG
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Fix notification focus for agent terminals
* Restrict notification title fallback to agents
* Simplify the focus fallback to a lazy two-tier query
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* Keep AppLibrary scans out of login shells
A login shell sources the user's profile, and tools like mise touch
~/.local/share on activation. Quickshell's desktop-entry watcher monitors
that tree, so every scan retriggered another scan: a self-sustaining loop
that pinned ~20% of a core at idle. The scans only use bash builtins,
find, and sort, so they don't need the login environment.
Fixes#6806
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Pin the AppLibrary scans to non-login shells in tests
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Cover the KVM and disk requirements, the resources and credentials the
installer asks for, the launch/stop/status commands and --keep-alive,
what the RDP session carries, and what removal deletes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Persist notification images so history keeps avatars
Persisted popup and history entries stored image/appIcon as URLs into
resources that die with the live notification: Chromium-family senders
(every Omarchy web app, WhatsApp included) pass avatars as files in a
scoped /tmp dir deleted when the notification closes, and raw image-data
hints surface as in-process image:// URLs that die with the server
object. Replaying history then found dead references and hid the icon.
Copy file-backed images into the notification state dir when persisting,
keyed by the entry's file stem, and reference the copies from the JSON.
Blank dead image:// URLs so the card falls back to the app icon. The
copies die with their JSON: superseded-popup deletes, history trims and
clears remove them, and a startup sweep collects copies orphaned by a
restart killing a queued job mid-write.
Hold DND-silenced notifications open until their history write has run,
since untracking tells the sender to delete its avatar file, and carry
replayed on-screen rows over via their persisted copies, since the
replay dismisses their live notifications first.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Coalesce silenced updates and bound image copies through temp files
A replaces_id update lands on a held DND notification without a second
onNotification, so releasing after the first write could persist a stale
snapshot. Re-snapshot when the write completes and write again until the
content is stable, reusing the original file identity.
The image copy reopened the sender-controlled path after checking it, so
a file growing or becoming a FIFO mid-copy defeated the size bound. Read
through head -c under a timeout into a temp file, validate its size, and
rename it into place; the startup sweep clears temp files a killed job
leaves behind.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Neither group has any commands behind it, so both only ever printed
"Unknown Omarchy command" when browsed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Add chapters for the top bar and its panels, screenshots and recording,
toggles/idle/screensaver, browsers, shell plugins, and networking, which
collects the Wi-Fi, DNS, firewall, and Tailscale material that was spread
across the FAQ, TUIs, and services chapters.
Fill smaller gaps in the existing chapters: user hooks, autostart, menu
extensions, printing, power profiles, factory reset, drive and user
passwords, passwordless sudo, firmware updates, display text size, and
the LocalSend share menu.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* perf(agents): cut codex usage collector memory with SQL filter and cache
The codex collector scanned every row of opencode.db (1.7 GB, 55k+ rows)
with Python-side json.loads, peaking around 716 MB of RSS on every run
-- including the panel's refreshLimits() call, which passed --limits-only
that the collector silently ignored.
Filter rows in SQL (LIKE gates + json_valid + json_extract authority,
mirroring the old Python filter semantics) so giant blobs are never
parsed, and cache the local stats scan in XDG_CACHE_HOME following the
claude collector's pattern (atomic writes, flock, schemaVersion).
--force rescans, --limits-only and normal mode reuse a fresh cache and
fall back to a full scan when it is missing, stale, or corrupt.
Measured: cold scan 716 MB -> 158 MB peak; warm --limits-only ~85 MB
and ~1.4 s. Output record schema and values are unchanged for the same
data (parity verified against the old filter, including malformed rows).
* Scope the codex scan cache's 15-minute reuse to --limits-only
A no-flag run is the widget's periodic refresh, and refreshIntervalSec is
configurable down to 30 seconds; holding every mode to a 15-minute cache
meant stats could lag far behind the interval the user asked for. Mirror
the claude collector: normal runs reuse a scan for ~20 seconds purely to
dedup concurrent collectors, and only --limits-only, which promises just
fresh limits, may reuse a scan for up to 15 minutes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Invalidate the codex scan cache across day boundaries
The cached stats embed date-dependent fields (todayPrompts,
todayTotalTokens, recentDays), but only the file's age was checked, so a
cache written at 23:58 served yesterday's numbers as "today" for up to
15 minutes past midnight. Stamp the envelope with the scan's local date
and treat any other date as a miss.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Reject codex scan caches with a future mtime
A cache whose mtime is ahead of the clock has a negative age, which the
freshness check accepted forever: setting the clock backwards froze the
stats until real time caught up with the file. Require a non-negative
age before trusting the cache.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Never cache an interrupted opencode scan
A transient lock, schema migration, or corrupted database aborts the
opencode scan mid-flight; the partial numbers still serve the current
run, but persisting them let a single bad read suppress opencode usage
for every cache reader until expiry. The claude collector already skips
its opencode cache write on a database error; do the same here.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Make the json_valid guard order explicit in the opencode query
The query relied on json_valid(data) evaluating before json_extract(),
but SQLite does not promise that AND terms run left to right; a
reordered plan would let json_extract raise on a malformed row and
silently truncate the scan. Wrap each json_extract in a CASE so the
guard is structural rather than positional.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Drop a claude-collector comment that is false for codex
"These caches were world-readable before" was copied from the claude
collector; codex had no caches before this one existed. Explain the
chmod on its own terms.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: markbusking <marcosbustos.dev@gmail.com>
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
omarchy-hw-hybrid-gpu gates the Hybrid GPU menu entry, and it queried
supergfxctl unbounded — a wedged supergfxd stalled menu rendering
forever. Bound the query with the same TERM-then-KILL escalation the
toggle uses, and treat a daemon that cannot answer like a machine
without supergfxctl: fall back to counting GPUs rather than hiding
hardware that is really there. An ordinary supergfxctl failure still
hides the entry.
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* Only offer video capture devices as webcams
V4L2 exposes raw processing and output-only nodes beside usable cameras. The webcam picker and automatic recorder selection treated the first /dev/video node as a camera, so IPU6 laptops opened a black overlay instead of their loopback capture device.\n\nShare one device lister across detection, selection, and recording, and keep only groups whose first video node advertises Video Capture in Device Caps. Cover raw IPU nodes, ordinary capture devices, and capture-less systems.
* Fall through to a later capture-capable node in a webcam group
A group whose first video node is not capture-capable vanished entirely,
even when a later node in the same group could capture. Probe each node
until one qualifies, still emitting at most one device per group. Also
exit zero explicitly: a trailing filtered device used to leak the failed
capability check as the script's exit status.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* Bound hybrid GPU mode queries
* Test blocked hybrid GPU queries
* Give the blocked-client test headroom over its 12s of kill cycles
The third case spends ~12s of real TERM/KILL escalation against its own
15s watchdog, which can tip to a spurious 124 on a loaded machine. Also
drop the TEST_LOG plumbing no stub ever wrote.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
The interval floor was applied with max(), so the zero that --force
picked could never win: max(0, 15) is 15. Forcing a refresh within
fifteen seconds of the last probe silently served the cache instead,
though --force documents itself as ignoring them.
The window exists to absorb a panel opened and shut repeatedly, which
arrives as --limits-only. --force is a person pressing refresh, and it
should outrank a window meant for flicks.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Stop closed network panels from leaving Wi-Fi scanning enabled
refresh() defaults scanWifi to false and its no-scan branch enabled the
scanner unconditionally. Five paths reach it with no panel on screen —
Component.onCompleted, clearNetworkAction(), failNetworkAction(), the
band-change actionProc exit, and the 30s actionTimeout — so the scanner
stayed on and Quickshell kept re-arming RequestScan behind a closed panel.
scanRestart had the mirror gap: it enabled the scanner 100ms after
refresh(true) without re-checking that the panel was still open.
Every sweep takes the radio off the operating channel, so this degraded
the link it was scanning from: one sweep every 17s, gateway RTT rising
from ~2ms to repeated 150ms+ spikes on an otherwise idle connection.
Gate the scanner block on the panel being open, cancel a pending restart
on close and re-check the panel when it fires, and track the WifiDevice
this instance enabled so close, device replacement and destruction
release the right object. Destruction matters on its own: a bar reload
with the panel open would otherwise die with opened still true and never
write scannerEnabled = false.
* Cover the scanner ownership helper's own invariants
The previous assertion only pinned that no write bypasses
setScannerEnabled(); it said nothing about what the helper does. Dropping
either the opened gate or the release-before-adopt from the helper still
passed, while a closed instance could reclaim scanning and a device swap
could leave the previous interface scanning.
Run the helper's actual JavaScript against stand-in devices instead,
following the extract-and-eval pattern the agents panel tests already use.
Removing either invariant now fails its own assertion.
* Measure the About layout in UTF-8 so its window is not fitted too narrow
wc -L only counts display columns in a UTF-8 locale. A session that never
set one leaves it counting the box-drawing and Nerd Font glyphs the About
layout is built from as nothing, which measured the content 21 columns
narrower than it renders and fitted the window to clip it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Open the About window at the size it last settled on
The window used to map at the float rule's starting size, paint, and only
then measure itself and resize, so every open flashed one window size and
reflowed into another. The size that hugs the content can only be measured
from inside the terminal, so remember it and apply it as a window rule
before the terminal is spawned: the window now maps at its final size and
never moves. A rebranded logo or a new font falls back to the float rule
for one launch, refits, and is remembered from then on.
The fit itself now moves the window by the cells it is off by, rather than
scaling it to the grid, which multiplied up the terminal's padding along
with them and left the fit a column or two short. It accepts a cell of
slack instead of chasing an exact grid, since a window lands where the
terminal's cell boundaries put it, not where it was asked to.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Only the Claude Code CLI can refresh the OAuth token it saves; the
collector just reads it. A machine left alone long enough finds the
token lapsed, and that branch returned an empty limits list with no
status text at all, so the panel hid its whole limits section and
explained nothing.
Say what is wrong, and fall back to the cached limits already on disk
rather than discarding them. Cached windows are kept only until they
reset: a percentage from a window that has rolled over describes a
period that is over, and pinning a stale 78% on an allowance that is
now untouched would be worse than showing nothing. The probe-failure
path gets the same filtering for the same reason.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Drop the kms hook when the proprietary NVIDIA driver handles early KMS
install/hardware/nvidia.sh early-loads nvidia_drm (modeset=1) for early KMS,
but HOOKS still carried the kms hook, so autodetect pulled nouveau and
~100 MB of its GSP firmware into every initramfs for a driver that never
runs. On a Limine UKI setup that meant a 256 MB image where ~144 MB is
normal, doubled again by the fallback history on /boot.
Filter kms out of HOOKS when nvidia_drm is in MODULES (nvidia.conf sorts
before this drop-in) and every PCI display controller is NVIDIA. Hybrid
systems keep kms so the iGPU retains early KMS at the LUKS prompt.
Verified on an RTX 4090 (nvidia-open-dkms 610.57.04): UKI shrinks
256,183,296 -> 144,066,048 bytes, nouveau and its firmware gone, the
nvidia-utils GSP blobs and all four nvidia modules retained, Plymouth
still owns the LUKS prompt via nvidia_drm.
Fixes#6790
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Address review: quote literals, harden PCI detection, add shell tests
Quote fixed string literals in the [[ ]] comparisons per AGENTS.md, and
read the PCI tree through OMARCHY_PCI_DEVICES_PATH, the same seam
bin/omarchy-hw-nvidia already uses.
Require a positively identified NVIDIA display controller before dropping
kms: an empty or unreadable PCI tree previously counted as "no non-NVIDIA
GPU" and would have dropped the hook. Unexpected trees now keep kms.
Cover the conditional in test/shell.d/nvidia-kms-hook-test.sh: nvidia-only,
hybrid, no nvidia_drm, MODULES unset under set -u, audio-function-only,
empty tree, and a device directory missing its sysfs attributes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Treat unreadable PCI devices as inconclusive and make the test hermetic
A device whose class/vendor attributes cannot be read could be another
GPU, so skipping it let a readable NVIDIA GPU beside it drop kms without
having verified the whole tree. Count it as a non-NVIDIA sighting so kms
stays, and cover the mixed case in the test.
The test also sourced the host's /etc/vconsole.conf under set -u, where a
valid KEYMAP-only file makes the XKBLAYOUT expansion fail in the subshell
and ties the result to the machine running it. Predefine XKBLAYOUT and
FILES before sourcing the config.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Rebuild existing initramfses once the kms hook no longer applies
The settings package deploys the new omarchy_hooks.conf conditional, but
nothing rebuilds the initramfs when only a mkinitcpio drop-in changes, so
existing NVIDIA-only installs would carry nouveau's ~100 MB of GSP
firmware until their next kernel update. Following the precedent of
1784476564, add a migration that rebuilds via limine-mkinitcpio — once
per machine, and only where evaluating the installed drop-ins shows the
conditional actually dropped kms, so hybrid machines, non-NVIDIA
machines, and user-edited configs are left alone.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Document the mid-sourcing MODULES caveat in the kms conditional
A later-sorting drop-in that resets MODULES outright (as
surface_device_modules.conf does) would strip nvidia_drm after kms was
already dropped. Every machine Omarchy writes such a file for is hybrid
Intel and keeps kms through the PCI scan, but that is worth stating so
the invariant is not broken by accident.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Quattro hands wifi to NetworkManager, which starts wpa_supplicant through
D-Bus activation. Installs carrying a wpa_supplicant.service mask from the
iwd days break that activation: NetworkManager retries five times, gives
up, and every wifi device sits at "unavailable" with no network to
research the fix on. Remove the mask (including a runtime one) and restart
an active NetworkManager when a wifi device is stuck, so wifi comes back
without a reboot.
Fixes#6783
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
The discovery retry timer turned adapter.discovering on every second
while the panel was open, and nothing ever turned it off. The BlueZ
discovery session behind it is held by quickshell's D-Bus connection,
so one visit to the panel left the radio in inquiry until the next
shell restart — continuously starving A2DP audio on the same controller
into stuttering, and 'bluetoothctl show' kept reporting
'Discovering: yes' long after the panel was gone.
The panel now tracks the StopDiscovery it owes BlueZ and settles it
once closed. A timer bound to the confirmed discovery state does the
stopping, rather than a write in the close handler: quickshell only
forwards a discovering write that differs from the last state BlueZ
reported, so a stop issued while a just-fired StartDiscovery is still
awaiting confirmation would be swallowed and leak the session. Binding
to adapter.discovering re-arms the stop whenever the confirmation
lands, a reopen inside the first interval keeps the scan running
uninterrupted, and attempts are bounded so a session another BlueZ
client holds up cannot draw StopDiscovery calls forever.
One widget instance exists per monitor and they all share the default
adapter — the same shared-backend shape the network panel's wifi
scanner fix (#6772) dealt with — so the debt follows the session: an
instance opening onto a running scan adopts it, a closing instance
hands it to a panel still open on another monitor (the popout handoff
closes one instance as it opens the next), and a destroyed instance
passes it to a surviving sibling.
Fixes#6789
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* Recover image picker after interrupted thumbnails
* Use arithmetic assertions in image cache tests
* Bound thumbnail lock waits and reap partial thumbnails
A hung generator (vips stuck on a corrupt file or slow mount) held its
flock forever, wedging every later picker open; the directory-lock era
capped that wait at 30 seconds, so keep the same bound. A generator
killed mid-write also stranded its partial .jpg.<pid>.jpg forever, since
nothing prunes the cache directory; only the lock holder writes those,
so reap them right after taking the lock.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Harden thumbnail locks and cache publication against races
Adversarial review caught three holes. The lock fd leaked into
vipsthumbnail, so an orphaned or hung vips kept holding the lock after
its shell died; close it for the child. Reaping legacy lock directories
unconditionally raced a still-running legacy generator through an
upgrade; only reap ones older than the longest plausible generation.
And cache publication was neither atomic nor exclusive, so a picker
killed mid-write, or two interleaving, could leave truncated or
mismatched rows behind signatures that still validated - the same
permanent hiding this branch set out to fix; publish via renames under
a per-key lock, rows first.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* fix `--help` being skipped for bin routes that only partially resolve to a file
* Add regression tests for --help on partially resolved routes
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Stop help-flag scanning at -- and match --json as an exact token
A `--` marks the rest of the args as belonging to the command itself, so
neither --help nor --json past it should be treated as router flags. The
--json check also matched substrings of flattened args, so a --json inside
one quoted argument switched --help output to JSON.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
The floating rule pinned every agent terminal to 1200x800, which
overflows small and scaled displays: window rules see logical pixels, so
a 2560x1440 monitor at scale 1.6 is only 1600x900 and the window covered
89% of its height. Tiling drops the fixed size along with the rule, and
the shared app-id still earns the terminal tag from terminals.lua.
* Offer an AI diagnosis when a process crashes
systemd-coredump journals every core dump under a known MESSAGE_ID with the
crashing program, pid, and signal as structured fields. omarchy-crash-watch
follows that stream and raises a "Process crashed: <program>" toast; clicking it
opens omarchy-agent-crash, which briefs the default agent on the crash.
The toast goes through omarchy-notification-send --exec rather than a libnotify
action, because the shell runs clicks from its own omarchy-exec hint and never
emits ActionInvoked. It keeps the default "omarchy-action" app name too, the
only one shouldBypassDnd() lets through -- a crash being the last notification
worth swallowing. It stays quiet until an agent is configured, since a
diagnosis is all it offers.
The method lives in a diagnose-crash skill rather than the prompt, so it is
edited in one place and works with whichever agent is default. It covers
investigating the core, and reporting a confirmed Omarchy bug upstream: scoped
to bugs Omarchy controls, searched for duplicates first, only with the user's
agreement, and signed with the model and harness that produced it.
A migration reaches existing installs, whose skill symlinks and unit enablement
would otherwise sit behind one-time setup paths.
* Let the diagnosis clean up the core it extracted
"Do not modify or delete anything" contradicted the symbolization step right
above it, which writes a core to a temp file and deletes it on exit. Read
literally, the core survives -- and the same section warns it holds passwords
and tokens. The prohibition is about the system, not about your own scratch.
* Do not spend a crash toast on a dead notification server
The shell owns org.freedesktop.Notifications, so its own crash takes the
notification server down with it -- and a shell crash is exactly what you want
told about. The toast was sent once into that gap and the dedupe window was
recorded regardless, so the rest of the crash loop went quiet for a minute and
`journalctl -n 0` never replays what was missed.
It now waits for the restarted shell to reclaim the bus name, as
omarchy-migrate-notify already does, and only a delivered toast starts the
dedupe window.
* Reshape the agent launcher into omarchy agent
omarchy-launch-agent becomes omarchy-agent, with prompts on omarchy-agent-prompt
rather than the bare route: `omarchy agent` is both a command and a group, so a
positional prompt there would shadow any subcommand under it. The launcher takes
flags only and points at `omarchy agent prompt` when handed one.
Every agent window now launches under a fixed org.omarchy.agent app-id instead of
omarchy-launch-tui's default of org.omarchy.<binary>, so one rule floats them all
whichever agent is default.
Omarchy also stops picking an agent for you. omarchy-default-agent prints nothing
until one is chosen, leaving every entry under Setup > Defaults > Agent unchecked,
and a first-run invitation offers to take you there.
* Wordsmith
* Cover the agent routes and the invitation
The route split is the point of the change, so exercise `omarchy agent`,
`omarchy agent prompt`, and a rejected positional prompt through the router
rather than only the binaries behind them.
The invitation gets the same treatment as the Voxtype and fingerprint ones: it
notifies once, opens the agent defaults menu, and leaves both the notification
and the marker alone for anyone who already chose an agent.
* Offer the agent choice from the keybinding
Super + Shift + Ctrl + A now runs `omarchy-agent --pick`, which opens Setup >
Defaults > Agent when nothing is chosen yet. A keypress that writes to stderr
and opens nothing just looks broken.
* Reach existing installs with the agent invitation
first-run installs the invitation hook, and existing accounts marked it complete
long ago, so they would never see it -- while being the accounts most likely to
need it, since the old getter returned opencode implicitly and most have no
agent recorded at all. Post-update hooks run later in the same update, so the
invitation arrives without waiting for another one.
* Say what the Defaults submenus set
Setup > Defaults lists Agent, Browser, Terminal, Editor, but the header inside
each repeated the same bare word, which reads as a category rather than a
setting -- and says nothing at all when the menu is summoned straight into it.
The list keeps its short labels; the headers now name the setting.
Standing in for an abandoned compositor means binding a Unix socket, and the
sandboxes this guard exists for are the ones that deny it: the fixture raised
PermissionError and took the whole file down with set -e, adding a failure in
the environment the guard was written to keep clean. Run the cases that need no
socket first and skip the rest when one cannot be bound.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Read the keyboard being typed on rather than the one holding main
The main flag names no keyboard for long. fcitx5 takes it with the
virtual keyboard it binds to inject, and those are filtered out, so on a
seat running an input method the pick lands on nothing at all: no label,
and the widget hides itself off the bar. #6727 keeps polling in that
state rather than settling it, and the poll has nothing new to read.
Once fcitx5 unbinds, the flag lands on whichever device libinput listed
last, as easily a lid switch as a keyboard, and a device that never
receives the toggle reports the layout it started on forever, which is
the reading #6574 opened.
Every device carries the seat's layout list, but only the keyboard being
typed on advances through it, so read the furthest-advanced one.
activelayout names the keyboard it moved ahead of the layout, so take
that name and let it settle the pick, and the click that switches it.
* Leave the buttons out of the seat the widget reads
Reading the keyboard being typed on left keyboardName standing for two
things at once: the device a click switches, and the device activelayout
last named. Only the second was still being set, so the first went empty
until a switch happened -- which left the click doing nothing on a seat
whose only switch is the click, and left the poll running forever on the
one-keyboard install it was written to leave alone. Give each its own
property, and set the switch target from the reading that confirmed the
keyboard is there.
Layout progress only points at the keyboard being typed on while the
other devices stay where they started, and the ACPI power button, lid
switch and sleep key never do move on their own -- but they answer to
switchxkblayout and can hold the main flag, so anything that reads or
switches whatever the seat hands back can end up describing a button, and
unplugging the keyboard beside one leaves it standing in for the seat.
Drop them where the virtual keyboards are already dropped.
A reading that reaches hyprctl and finds no keyboard now clears the label
rather than leaving a device that is gone described on the bar, told
apart from the empty output a killed query leaves by the device list
itself, and the watchdog asks again rather than waiting for a poll that a
settled seat has already stopped.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
* Skip shell tests when the compositor can't be reached, not just when WAYLAND_DISPLAY is unset
A set variable only proves the environment was inherited. Sandboxes pass it
through while blocking $XDG_RUNTIME_DIR, so Quickshell cleared the guard and
aborted inside QGuiApplication, leaving two core dumps per launch instead of a
clean skip. Probe the socket and, when there's a signature to ask with,
Hyprland itself. Disable core dumps on the way through for the compositor that
dies mid-run, which no probe can catch.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Retry the compositor query before calling it dead
Hyprland can miss a query while it reconfigures outputs, and one miss was enough
to skip a whole file's runtime coverage. Retry the way omarchy-launch-shell
does. Only a leftover socket reaches the query at all, so the ordinary skip
still returns immediately.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Quickshell watches the QML it loaded and reloads on change, so pacman
replacing /usr/share/omarchy/shell mid-transaction makes the running
shell reload against a half-written tree. That reload fails, and a
failure that reaches the config load is not harmless: it raises the
reload popup, which is a second engine generation.
EngineGeneration::currentGeneration() returns null unless exactly one
exists, so the IPC kill that omarchy-update sends moments later takes
the QCoreApplication::exit(0) branch instead of the generation's own
quit, and Quickshell tears the QML graph down after deleting the
QGuiApplication. The first GUI resource touched on the way out aborts:
FATAL: QPixmap: Must construct a QGuiApplication before a QPixmap
The user gets the crash dialog after an update and a coredump per
occurrence. Reported in #6748 with 3 crashes across 10 updates, always
following a failed reload.
Fixing this in omarchy-update — stopping the shell around the pacman
step — would cover one caller and cost the polkit agent and the
notification server for the length of the transaction, which the
migrations that run next still notify through. It would also have to
carry omarchy-restart-shell's refusal to restart a locked session, or
reintroduce the hazard that refusal exists for.
And omarchy-update is not the only thing that rewrites the tree. The
pacman guard turns away a bare pacman -Syu, but nothing turns away a
targeted pacman -S omarchy, a pacman -U of a locally built package, the
documented OMARCHY_ALLOW_DIRECT_PACMAN bypass, omarchy-dev-pkg-test, or
a checkout in a dev-linked tree.
Turn the watcher off instead. Omarchy has never reloaded through it:
omarchy-restart-shell is what picks up QML changes, and config and
plugin changes go through the shell's own IPC. Third-party plugin hot
reload is PluginRegistry's own inotifywait and FileView watches its own
files, neither of which this touches — QuickshellSettings::watchFiles()
gates the config scanner and nothing else. The popup goes off with it,
because QML can still ask for a reload directly and leave the same
extra generation behind.
Environment reaches Quickshell only at launch, so the update that
delivers this still runs under a watching shell. It takes effect from
the next one.
Verified against an isolated instance: breaking a config in place and
then sending the IPC kill reproduces the FATAL, and it stops with either
variable set. QS_DISABLE_FILE_WATCHER also keeps the failed reload from
happening at all.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Drops the ChatGPT web app from the default set at the same time, so installing
the openai-codex-desktop package can't leave two identical-looking ChatGPT
entries in the launcher. Super + Shift + A still opens the web version, which
is the only place it was really used.
The bundled ChatGPT icon stays: the package's own chatgpt.desktop asks for
Icon=chatgpt and ships no hicolor icon of its own.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The vendor-ID guess rejects any device with a driver bound, on the
reasoning that libfprint drives readers from userspace so a real one sits
there unbound. But libusb claims interfaces through a synthetic usbfs
driver, so the reader binds one for as long as fprintd holds the claim —
which is exactly while it is being enrolled or verified against.
Readers that name themselves take the product-string branch and never
reach this, so the exposure is the ones that don't: Goodix 27c6:6594
reports "Goodix USB2.0 MISC", matches on vendor ID alone, and drops out
of detection mid-authentication. The menu entry disappears and the
first-run invitation stops firing while the reader is in use.
Ignore a driver link that resolves to usbfs, and keep rejecting the real
ones — usbio-bridge, usbhid, uvcvideo — including on a device that has
both.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Restore FPC fingerprint detection
* Anchor the FPC product match to a prefix
*fpc* is an unanchored three-letter token on the one branch that is
trusted outright, with none of the kernel-driver checking the vendor
guess gets. Every FPC reader on record leads with it — "FPC Sensor
Controller", "FPC Sensor Controller L:0002 FW:25.26.23.14", and this
branch's "FPC L:0000 FW:1425046" — so requiring the prefix costs no
coverage while keeping three letters from matching mid-string, where
FPC abbreviates unrelated things like flexible printed circuit.
Also restore the note about why Elan is kept out of the vendor list, so
the two signatures read as the same deliberate exception.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Write no product descriptor for a two-field device spec
${remainder#*:} returns the string unchanged when there is no second
colon, so a spec meant to describe a device with no product descriptor
wrote the product id out as its product string instead. Every call site
happened to pass a trailing colon, so the suite was right by accident.
Guard the split and drop the trailing colons, so the two vendor-match
cases exercise the path they were written for.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Bind a named driver in the kernel-driver fixture
Touching a bare `driver` file asserts that any driver at all disqualifies
a vendor guess, which is more than the detector should promise: libusb
claims an interface through a synthetic `usbfs` driver, so a reader in
active use looks bound by that rule.
Link the interface at a named driver directory instead, the way sysfs
does. The case still covers what it was written for — a Synaptics bridge
or a camera on a fingerprint vendor ID — without fixing the shape of the
answer for drivers it was never about.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The widget polled hyprctl every 10 seconds per monitor, including on the single-layout install where it never shows. Keep the poll only where its answer can change - a seat with more than one keyboard, where Hyprland moves the main flag with no event to announce it - and stop it entirely once a one-keyboard seat has been read.
Also coalesce a refresh that arrives mid-query instead of dropping it, time out a query that never returns rather than letting it hold the guard shut for good, and re-read the layout on configreloaded.
Co-Authored-By: markbus-ai <markbus-ai@users.noreply.github.com>
* Replay the history a dismissal or a clear was still being written into
The popup files a replay reads are written by a serialized queue of shell
jobs, and the read ran as its own process alongside it. A dismissal issued a
moment earlier could still be queued when the directory was read, leaving the
notification out of the replay it was the newest entry of, and a clear issued
a moment earlier could still be queued too, replaying entries it was about to
remove.
The read now waits for the queue to go idle, so the replay shows the history
as of the moment it was asked for rather than whichever jobs happened to have
landed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Catch up on an update that arrived before its popup had a row
Watching a notification for in-place updates starts the moment it is handed
over, but the row those updates write to is inserted a tick later, deferred to
keep a mid-incubation Repeater from being mutated underneath. A client fast
enough to update inside that window found no row to write to, and a property
that has already changed does not change again — so the toast and its file sat
on the superseded content until something else moved.
The row is now refreshed from the live notification once it exists. That reads
the same object the signals would have, so an update that beat the insert is
picked up and one that did not costs nothing: a refresh whose content matches
the row it would write is dropped, which also collapses the several signals a
single multi-property update emits into one rewrite.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Hold queued file work behind the replay's read, not just ahead of it
The read waited for everything queued before it, but nothing stopped the queue
from running on while it worked. A clear or an archive issued during the read
could delete or move files out from under awk mid-glob, so a replay could still
show a partial history — some of what a clear was in the middle of emptying.
The read is a barrier in both directions now: the queue holds until it exits,
and it releases on exit rather than on output, so a read that comes back empty
or fails cannot park the queue behind it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Queue the replay's read instead of waiting for the queue to empty
Waiting for the queue to go idle before starting the read still let work
overtake it. A clear or an archive enqueued after the replay was asked for,
while the current job was running, was dequeued the moment that job exited —
the read only starts once nothing is left — so the replay showed the state
after those jobs, which is the race this was meant to close. Unbroken file
traffic could postpone the read indefinitely for the same reason.
The read is now an entry in that queue rather than a process running beside
it. It takes its place in line behind the work queued before the request and
ahead of everything queued after, so no later job can overtake it and no
amount of traffic can push it back.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The pacman cache grows without bound across updates, and nothing in the
update flow ever reclaimed it. On a machine that has been updating for a
while it reaches several gigabytes of superseded versions that nothing
will ever install again.
Prune it with paccache -rk2 as the first step of an update. Both halves
of that placement are load-bearing.
Keeping two versions rather than one preserves the rollback path. The
cache is Arch's only offline downgrade: when an update breaks a single
package, reinstalling its predecessor from here is the surgical fix,
where a snapshot rollback would revert every other package too. Pruning
before the packages update means the installed version is still the
newest cached, so it survives along with a spare. Retention is by
version order and never consults what is installed, so that holds while
the installed version is among the two newest cached; a deliberate
downgrade or repeated failed transactions can stack newer archives on
top of it.
Running before the snapshot is what actually frees the space. The cache
sits on the snapshotted root subvolume, so a prune taken afterwards
leaves the fresh snapshot holding those extents and reclaims nothing
until it ages out of the number cleanup.
A failed prune warns and continues. Cache housekeeping should not trip
the update's ERR trap and tell the user their update went wrong.
This runs after omarchy-update-requires-free-space, so it reclaims space
during healthy updates but does not rescue a machine already under the
10 GiB gate.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
A client that updates a notification through replaces_id does not produce a
second onNotification: Quickshell writes the new content onto the Notification
object the shell is already holding. The card draws a snapshot copied out of
that object — deliberately, since a live QObject in a ListModel role becomes a
dangling pointer the moment the server destroys it — so the toast kept showing
the superseded text, and archived it to history when it left the screen. A
Slack thread that updates in place read as stuck.
Every property the card draws is now watched on the notification we hold, and
a change rewrites both the model row and the file the popup was persisted
under. The file name is that popup's identity, so the rewrite lands in place:
a shell restart restores the version last shown, and so does the copy that
reaches history.
The countdown starts over when the content changes. New text arriving a second
before the toast was due to expire deserves a full look, not the remainder of
the clock the text it replaced had nearly run through.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
History was a pair of in-memory lists mirrored into notifications.json, split
into "pending" and "past" by a seen/unseen distinction no surface exposed,
capped at 100, deduped by an id that repeats across server generations, and
pruned by a 15-minute TTL. Replaying it showed five rows drawn from whichever
list happened to hold them.
Every toast already writes a file under ~/.local/state/omarchy/notifications
so it can survive a shell restart. That file is now the history record: when
the popup leaves the screen it moves into notifications/history instead of
being deleted, the newest ten are kept, and showHistory replays exactly what
is in there, including the toasts still on screen when it is asked for. A
notification DND silenced is written straight into the same directory, since
a toast that never showed is the one worth looking back at.
That leaves the models, notifications.json history payload, past pruning, and
the /tmp image cache that existed to keep century-old history thumbnails alive
with nothing to do, so they go.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Every supported agent spells it differently, so map each one to its own
bypass flag instead of leaving the launcher at each agent's default.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
1784917531 gated its UKI rebuild on initramfs_async=0 being present in
the Limine config, but omarchy-settings ships omarchy-defaults.conf with
that parameter already in it. Any machine that installed the package and
ran the migration in the same update matched the config the package had
just written, skipped the rebuild, and kept booting an image baked
before the config existed — without initramfs_async=0, so encrypted
boots still fell back to an unthemed text LUKS prompt. Compare the
booted command line against the configured one and rebuild when they
disagree.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-11 23:35:40 -07:00
676 changed files with 30900 additions and 2188 deletions
- [`agents/skills/command-metadata.md`](agents/skills/command-metadata.md) - adding or changing commands in `bin/`
- [`agents/skills/install-scripts.md`](agents/skills/install-scripts.md) - working under `install/` or on system/user setup commands
- [`agents/skills/shell-dev.md`](agents/skills/shell-dev.md) - editing the Quickshell desktop under `shell/`
- [`agents/skills/icon-font.md`](agents/skills/icon-font.md) - adding branded glyphs to `default/fonts/omarchy/omarchy.ttf`
- [`agents/skills/acceptance-tests.md`](agents/skills/acceptance-tests.md) - writing or running graphical acceptance tests under `test/acceptance.d/`
- [`agents/skills/visual-verification.md`](agents/skills/visual-verification.md) - verifying any change with a visual effect in the running UI
- [`docs/migrations.md`](docs/migrations.md) - creating or changing migrations under `migrations/`
- [`agents/skills/migrations.md`](agents/skills/migrations.md) - creating or changing migrations under `migrations/`
# Documentation Layout
Three documentation trees, split by genre and audience:
-`agents/skills/` - task procedure ("do this when doing X"), for anyone working on the codebase
-`docs/` - reference on how the system is shaped (file layout, update pipeline, theming, shell architecture), for anyone working on the codebase; skills link here for depth
-`manual/` - end-user documentation for using Omarchy, published; never codebase internals
# Style
- In markdown documents (`plans/`, `docs/`, `manual/`), write full lines — no hard wrapping at 80 columns; break only at structural boundaries like headings and list items
- Two spaces for indentation, no tabs
- Use bash 5 conditionals: use `[[ ]]` for string/file tests and `(( ))` for numeric tests
- In `[[ ]]`, don't quote variables, but do quote string literals when comparing values (e.g., `[[ $branch == "dev" ]]`)
- Prefer a full `if`/`else` conditional for simple two-path control flow; don't rely on `exec` or `exit` in one branch to make following statements unreachable
- For strings/paths with spaces, quote them instead of escaping spaces with `\ ` (e.g., `"$APP_DIR/Disk Usage.desktop"`, not `$APP_DIR/Disk\ Usage.desktop`)
- Shebangs must use `#!/bin/bash` consistently (never `#!/usr/bin/env bash`)
- Scripts under `install/` and `migrations/` may be sourced and intentionally omit shebangs
@@ -80,7 +91,8 @@ Exceptions are allowed for migration and package-helper scripts where the helper
# Menu
- The menu definition lives in `default/omarchy/omarchy-menu.jsonc`.
- The menu definition lives in `default/omarchy/omarchy-menu.jsonc`;
[`docs/menu.md`](docs/menu.md) covers the schema, guards, and providers.
- Do not add `aliases` to new menu entries. Aliases are reserved for
established alternate names users already type, kept for compatibility.
@@ -92,7 +104,9 @@ Exceptions are allowed for migration and package-helper scripts where the helper
# Tests
Run focused automated tests for the area you changed. Current test entry points:
Run focused automated tests for the area you changed;
[`docs/testing.md`](docs/testing.md) covers how the suites are shaped. Current
test entry points:
-`./test/all` - aggregate runner for CLI and shell tests; it intentionally does not run graphical acceptance tests
Read this before creating or changing migrations under `migrations/`.
Omarchy migrations are one-time repair scripts for existing installs. They are
used when a package update needs to change state that pacman cannot safely own by
itself.
@@ -124,6 +126,7 @@ New migration format:
- Start with an `echo` describing what the migration does.
- Use `$OMARCHY_PATH` to reference the Omarchy directory.
- Be idempotent. Check existing state before changing it.
- Migrations are strictly ordered and synchronous. A migration that cannot finish must exit non-zero, remain pending, and stop the queue; never mark later migrations complete against state an earlier migration has not established.
- Use helper commands such as `omarchy-cmd-present`, `omarchy-cmd-missing`,
`omarchy-pkg-add`, `omarchy-pkg-drop`, `omarchy-pkg-present`, and
`omarchy-pkg-missing` when appropriate.
@@ -163,3 +166,5 @@ omarchy-migrate
Omarchy 4.0 is upgraded through `bin/omarchy-upgrade-to-quattro`, not through the
normal migration runner. Do not add compatibility migrations for old installer
layouts; put pre-4 package-layout transition work in the upgrade command instead.
Clearing a privileged file that a retired installer left on disk is the exception, and belongs in a migration whether or not that installer was part of a package layout transition. The upgrade command only runs on a machine still making the 3 to 4 crossing, so anything put there never reaches an install that crossed already, and it never runs at all for an installer that was retired on its own — while the file the installer wrote is still sitting on those machines. The upgrade command finishes by running `omarchy-migrate` (`run_post_upgrade_migrations`), so one migration reaches every population; a copy in the upgrade command would only be a second copy of the same predicate to keep correct. Such a migration must name the defect it clears and match what the old installer actually produced before deleting it. Leave safe administrator-authored files alone; if one still contains the vulnerable privileged action, preserve it under an inactive name rather than discarding custom content or leaving the action executable. A user config that depends on the same retired compatibility path may be repaired in that migration when doing so eliminates an overlapping migration, but only by matching and replacing the exact legacy path while preserving the rest of the file.
GROUP_DESCRIPTIONS[transcode]="Image and video transcoding"
GROUP_DESCRIPTIONS[tui]="Terminal UI launchers"
GROUP_DESCRIPTIONS[tz]="Timezone selection"
GROUP_DESCRIPTIONS[update]="Omarchy and system updates"
GROUP_DESCRIPTIONS[version]="Version and channel information"
GROUP_DESCRIPTIONS[voxtype]="Voxtype dictation"
@@ -122,6 +122,31 @@ append_pipe_value() {
fi
}
# A route can resolve to a binary before all args are consumed (`update aur` only matches the `update` binary, leaving `aur` and `--help` as leftovers)
# --help must never be lost among those leftovers and get forwarded into the real command, so this checks the whole remainder rather than just the first token.
# A `--` marks everything after it as belonging to the command itself, so scanning stops there.
Loaded 100 of 676 files, more files were not shown because too many files have changed in this diff.
Show more
Reference in new issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
ZacharyZhang-NY
Ryan Hughes
