The page shows which keys exist, what the agent is holding, and the hosts this machine has met, with a two-press forget for a host whose key has changed. Nothing here reads private key material. Fingerprints and comments come from the .pub file, and "does this key need a passphrase" is answered by asking ssh-keygen to derive the public half with an empty one -- it succeeds for an unencrypted key and fails for an encrypted one, and either way the only thing it can emit is public. The contract checks that against the payload that actually reaches the page rather than against the source, because what the code intends and what it ships are different claims. Unloading a key from the agent is refused, with its reason. On this desktop `ssh-add -d` prints "Identity removed" and the key is still offered a second later: gnome-keyring's agent lists every key it finds in ~/.ssh, so a removed one comes straight back off disk. That was measured rather than assumed -- a plain ssh-agent removes durably, this one does not -- and a button reporting success while changing nothing is worse than no button. The page says so and names the thing that does work: move the file out of ~/.ssh. SSH_AUTH_SOCK is not set in a normal shell here, so a naive check reports "no agent" while one is plainly running. The helper falls back to the keyring socket, and an agent started by hand still wins. That gap is the same one that made reaching these servers awkward in the first place. Generating a key is deliberately absent. A passphrase cannot reach ssh-keygen without going somewhere it should not -- -N puts it in argv, which every process on the machine can read -- and driving the prompt over a pty did not work. Offering to generate an unencrypted key instead would be a downgrade dressed as a feature, so the page does not offer to generate at all. Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Settings
The control center for everything Panama owns. Anything the system owns — hardware, accounts, printers — is delegated to GNOME Settings and labeled as such rather than half-reimplemented.
System Health
The stable internal services route renders System Health. It is reachable
from the Settings sidebar and its live 54px footer, the degraded-only bar
indicator, and Vicinae's Check System Health command. Healthy scans
reserve no bar space and produce no notification.
services/Health.qml owns the last accepted redacted snapshot. It invokes
scripts/panama-doctor for scans and bounded repairs, wl-copy only for an
explicit Copy Report, and bounded notify-send only when an external repair
fails. For a concise terminal view, run:
~/.config/quickshell/scripts/panama-doctor --summary
The helper diagnoses Panama-owned desktop services, dependencies, links, and
configured integrations. It does not read secret values, clipboard or
notification contents, calendar events, SSIDs, addresses, or arbitrary command
output. Its repair interface is an authored allow-list: it never installs a
package, runs sudo, deletes user data, or repairs a service Panama does not
own. A repair remains degraded until a fresh scan observes recovery.
The final card is the ownership boundary. Network configuration and the exact Users, Sharing, Color profiles, and Digital wellbeing handoffs open GNOME Settings because Fedora's system services own those areas.
Adding a setting
One schema entry. That is the whole job.
// config/PreferenceSchema.qml
{
key: "blurSize", type: "int", def: 8, min: 1, max: 20, step: 1,
unit: "px", group: "effects",
label: "Blur radius",
detail: "Larger is softer and costs more frame time",
hypr: { path: ["decoration", "blur", "size"], option: "decoration:blur:size", readAs: "int" }
}
// the page
SliderRow { setting: "blurSize" }
Persistence, validation, clamping, reset, search indexing, and — with a hypr
block — live application to the compositor and the startup replay all derive
from that entry. There is nothing else to register.
If it is compositor-backed, add the matching prefs.get("blurSize", 8) in
hypr/looks.lua so the Hyprland config still stands alone with no settings
file.
Setting ownership
Every preference has one primary page, derived from its schema group and
the route in services/SettingsSearch.qml. Search results always open that
owner. A control may appear on a second page only when the same adaptation is
part of another established mental model; otherwise use a labeled handoff to
the owner instead of duplicating it.
Intentional mirrors
| Setting | Primary page | Mirror | Why the mirror earns its place |
|---|---|---|---|
animationsEnabled |
Appearance | Accessibility | Reduced motion belongs both to visual polish and motion accessibility. |
cursorInactiveTimeout |
Mouse | Accessibility | Pointer visibility is configured with pointer behavior but affects motor and visual access. |
cursorSize |
Accessibility | Mouse | Large cursors are an accessibility adaptation that users also look for beside pointer controls. |
inactiveOpacity |
Appearance | Accessibility | Window translucency is an appearance choice with a direct readability impact. |
lockMinutes |
Power | Privacy | Idle timing owns the mechanism; privacy owns the expectation that the unattended desktop locks. |
lockOnSleep |
Power | Privacy | Suspend owns the transition; privacy owns whether waking requires authentication. |
Lock-screen visuals belong only to Appearance: background source, blur, clock, date, user name, and password-field presentation. Power owns when the session locks, while Privacy keeps only the established timing mirrors above. Visual controls must not be copied onto either page.
Displays is the sole owner of mode, scale, rotation, arrangement, and primary role. Those values form one safety transaction: every connected output is applied, verified, confirmed, or restored together. Other pages may link to Displays, but must never expose a second geometry control or persist a partial layout.
Mirrors must remain the same schema-backed control, never a second preference
or a copied default. Additions to this table require a concrete discoverability
reason and an update to tests/quickshell/settings-ownership-contract.sh.
Window border color follows the same ownership rule. The inactive border is a
scheme-relative role owned by ColorScheme.qml: it changes only to retain
neutral contrast in light and dark modes. The focused Prism border is the
accent role, driven by the chosen accentName and also owned by
ColorScheme.qml: each accent carries a separate pair for light and dark, so
ColorScheme.qml restates the focused border alongside the inactive one on
every scheme change, rather than leaving a scheme flip to erase a
user-selected accent.
The rows
| Component | For |
|---|---|
SettingsPage |
The page scaffold: title, lede, optional pinned header |
ToggleRow { setting } |
A boolean |
SliderRow { setting } |
A number; zeroLabel renders 0 as "Never"/"Instant"/"None" |
ChoiceRow { setting } |
An enum, as a segmented control |
ActionRow |
A button: opens a GNOME panel, runs a one-shot |
TextRow |
A genuinely read-only fact |
TextRow is for facts, not for settings that were merely expensive to wire.
Before Stage 3 more than half of all rows were static text standing in for
controls; that is the failure this vocabulary exists to prevent.
Rows write through SystemSettings.commitPreference(key, value), which routes
compositor-backed keys through apply-and-verify and local keys straight to the
store. A row never needs to know which kind it holds.
Things that will bite you
readAs describes the answer, not the setting. hyprctl getoption returns
the value in a different JSON field per type — int, bool, float, str,
and css for gaps (a four-value box). Declaring the wrong one does not fail
loudly: it makes every write to that key look rejected, and the user sees an
error for a change that worked. tests/quickshell/schema-hypr-shape-contract.sh
asks the compositor for the real shape of every mapped option.
Never trust an exit code from hyprctl. keyword refuses to work on a
Lua-configured Hyprland, prints the refusal to stdout, and exits 0. eval exits
0 on syntax and runtime errors too. The only trustworthy signal that a write
landed is reading the value back.
The Settings window is tiled. implicitWidth is a hint; the layout decides,
and it ranges from a half-screen split to the full display. SliderRow stacks
its control under the label below 520px. Test narrow.
Binding an anchor to undefined does not reliably release it. Switching
layouts that way left a slider anchored to both edges with the label squeezed
into what was left. Position explicitly instead.
Inside a SettingsCard, parent is the card's internal Column. So
parent.modelData in a nested Repeater is undefined and the rows silently
never appear — you get a card with a heading and nothing under it. Address the
outer model through an explicit id.
A TapHandler declared as a child of SettingRow lands in the trailing
slot, because that is the row's default property, so only the right-hand edge
becomes clickable. Use activatable: true with onActivated for a whole-row
target.
A content-identical Quickshell entry can share the live shell's ID.
Quickshell derives the Shell ID from config content, not path. Runtime
harnesses therefore create a distinct semantic entry file, address that exact
file with qs -p, and discover its PID from the exact Config path in
qs list --all. They terminate only that recorded PID with kill; never use
qs kill from a copied configuration.
Where state lives
| File | Holds |
|---|---|
~/.config/panama/settings.json |
Everything in the schema. Read by the shell and by hypr/prefs.lua |
$XDG_STATE_HOME/panama/panama-home.json |
Home accessory favorites and aliases |
$XDG_STATE_HOME/panama/backups/ |
Settings snapshots |
$XDG_STATE_HOME/panama/hypridle.conf |
Generated idle config |
$XDG_STATE_HOME/panama/hyprlock.conf |
Generated lock-screen appearance |
SystemSettings.restoreDefaults() spans all of them. A reset that silently
skipped one would be worse than having no reset, because nothing would say so.
Not stored by Panama
Timezone and network time are read from and written to timedatectl directly.
They belong to the machine and are shared with sessions that never see Panama's
file; storing a copy would create a second answer to a question the system
already answers.