meta: 2026-07-30 residual-work sweep — index (#647-#660)

[vibechoom]

Index for a residual-work sweep run on 2026-07-30, in the same shape as #456's code-climate sweep. Filing the index so the cluster is skippable as a unit rather than reading as fourteen unrelated tickets.

Nothing here is a new feature request. Every item is work that was already started, promised, or documented and then left behind — a deferral whose follow-up never happened, a doc that describes a tree that no longer exists, or plumbing that was built and never wired.

Method, and what it is worth

Four blind sweeps ran independently over the workspace, each with a different modality so none could see what the others surfaced:

1. in-tree TODO/FIXME/"for now"/"stand-in" markers and their surrounding code
2. deferred scope announced in the last 60 merged PR bodies, then checked for a successor that never came
3. documentation drift — CLAUDE.md, etc/*/README.md, docs/superpowers/, docs/live-verify.md — every checkable factual claim verified against the current tree
4. built-but-unwired plumbing: public accessors, effect variants, and config options with zero consumers

Each sweep's output then went through a hostile stage that had to dedup against all ~400 open and closed issues, confirm the work had not since been done, and actively try to refute the finding. 44 candidates went in; 20 were killed — as duplicates of existing issues, as work already quietly shipped, or as false positives from the search technique. The 24 survivors are clustered here into 14 issues.

I then re-verified the load-bearing citation of every cluster against HEAD myself before filing. That pass corrected at least one earlier finding in the same family (a sleep(2s) that looked like a retry loop and is actually a one-shot debounce), so treat the surviving citations as checked rather than asserted — but the severity ratings are judgement, not measurement.

The issues

Wiring that was built and never connected

• feat(plugins): Effect::{Niri,Media,Audio} are declared, cap-gated and audited, but the host broker silently drops all three #648 — Effect::{Niri,Media,Audio} are declared in the proto, capability-gated, and named in the audit log, but the host broker drops all three with a warn! the plugin author never sees
• fix(bus): nothing consumes OwnState — a contested bus name is silent and retries RequestName at 4 Hz forever #653 — nothing consumes OwnState; a contested bus name is silent, and the already-owned path retries RequestName at 4 Hz forever with no log output
• fix(mpris): the Media panel's "Auto" chip can never stay pressed, and bare-string xesam:artist parses to empty #651 — mpris::selected_player() has zero subscribers, so the Media panel's "Auto" chip can never stay pressed
• feat(reactive): add spawn_supervised_blocking and a panic hook — niri's IPC loop is still unsupervised #654 — no spawn_supervised_blocking and no process panic hook; niri's IPC loop, the source of workspaces and fullscreen detection, is still unsupervised
• fix(widgets): battery emergency pulse uses a hardcoded 3 % instead of the UPower WarningLevel the toasts already ride #656 — the battery emergency pulse is keyed off a hardcoded 3 % rather than the UPower WarningLevel the toasts already ride
• feat(wifi): the secret prompt has no in-flight state and can only return one VPN secret per GetSecrets round #652 — the Wi-Fi/VPN secret prompt still has no in-flight state, and returns only one VPN secret per GetSecrets round

Stubs and duplication left behind

• fix(control-center): drop the "Display — Coming soon (#393)" placeholder tab; #393 was re-scoped and delivered elsewhere #647 — the control-center still ships a "Display — Coming soon (feat(display): org.gnome.Mutter.DisplayConfig shim over niri-ipc — g-c-c drives display (#375) #393)" tab for a feature feat(display): org.gnome.Mutter.DisplayConfig shim over niri-ipc — g-c-c drives display (#375) #393 explicitly re-scoped away and delivered elsewhere
• refactor(services): two independent 2 s niri Outputs pollers — displays.rs and display_config.rs never shared a source #655 — displays.rs and display_config.rs each open their own socket to poll niri Outputs every 2 s and never shared a source
• refactor(plugin-pet): migrate face.rs onto preem::Frame — the sibling #378 named and skipped #650 — hytte-plugin-pet's face.rs is the last hand-rolled raster buffer in the fleet, never migrated onto preem::Frame
• fix(nix): night light is home-manager-only, but its four options are declared — and documented as working — in the shared module base #657 — night light is home-manager-only, yet all four programs.trollshell.nightlight.* options are declared in the shared module base and document themselves as working

Documentation that is actively wrong

• chore(plugins,bus): three in-source deferrals describe a tree that no longer exists (caw, departures, hytte-bus tests) #649 — three in-source deferrals describe a tree that no longer exists (caw's briefing, departures, a hytte-bus test)
• docs: CHOOM-INIT.md and CLAUDE.md assert facts the tree contradicts #658 — docs/CHOOM-INIT.md and CLAUDE.md assert facts the tree contradicts
• docs(etc): four deployment READMEs describe shipped features as unshipped, and #219's GAction verbs are bound nowhere #659 — four deployment READMEs under etc/ describe shipped features as unshipped, and feat(shell): d-bus command surface so niri keybinds can drive sidebar, drawer pages, and power menu #219's GAction verbs are bound nowhere in the shipped niri config
• docs(live-verify): fold in #637/#639/#642 and correct the #630 entry's close() → destroy() #660 — docs/live-verify.md stops at fix(wifi): retry an inconclusive backend probe instead of latching "no backend" (#613) #634 while fix(overlays): destroy unrealized layer windows on teardown, not close() (#632) #637/fix(hytte-ui): destroy() the three remaining close()-teardown sites (#638) #639/fix(places): correct the shipped default station id (#641) #642 merged behind it

Sequencing notes

• docs(live-verify): fold in #637/#639/#642 and correct the #630 entry's close() → destroy() #660 should wait. fix(overlays): release RefCell borrows before the GTK call that can re-enter them (#631) #644 and fix(networkd): retry a transient startup refresh failure instead of going inert (#621) #645 are open and will each add live-verify entries; folding in fix(overlays): destroy unrealized layer windows on teardown, not close() (#632) #637/fix(hytte-ui): destroy() the three remaining close()-teardown sites (#638) #639/fix(places): correct the shipped default station id (#641) #642 now means touching the file twice. Do it once the current wave merges.
• fix(control-center): drop the "Display — Coming soon (#393)" placeholder tab; #393 was re-scoped and delivered elsewhere #647 is lane-blocked behind the in-flight fix: the RefCell-across-a-GTK-call pattern also lives outside overlays/ — reactive_list.rs, calendar/tasks/connections rows, control-center, hytte-ui popup (#631 blast radius) #643 work, which also edits crates/trollshell-control-center/src/main.rs.
• The documentation cluster (docs: CHOOM-INIT.md and CLAUDE.md assert facts the tree contradicts #658, docs(etc): four deployment READMEs describe shipped features as unshipped, and #219's GAction verbs are bound nowhere #659) is independent of all open PRs and can go first.

Severity, honestly

None of these is a crash or a data-loss bug. The medium ratings are mostly misleading-to-a-reader rather than broken-for-a-user: a plugin author who declares Capability::Niri and gets silence, or a contributor who reads CLAUDE.md and looks for a file that is now a directory. The documentation items are rated the way they are because this repo's docs are load-bearing for both humans and agents working in it.

Pick from this rather than working it top to bottom — it is a menu, not a queue.