docs(agent): file #137–#141 from the T14s round (three root-caused live)

Bernardo's 2026-07-16 observations, filed with the diagnosis done on the
machine rather than as symptoms:

- #137 plymouth: unindexed Window.Get* → whole splash placed on head 0's
  geometry; off-centre on any second monitor (LUKS prompt included).
- #138 [human] Meet claims no mic/speakers: audio graph + Flatpak sandbox
  ruled out live; needs the browser named and the failure in front of us.
- #139 terminal floats full-screen: % size rules silently no-op on
  Hyprland 0.55.4 (absolute px applies), and Kitty replays its remembered
  window size — the second half is the Ghostty regression from #95.
- #141 waybar updates click dead: whole-swaps use $TERMINAL, which is
  absent from Waybar's env; 4 jsonc lines, a parity break not a bug.
- #140 PROPOSED: Ask Claude → web chat instead of the claude-code REPL.

Also gives #127 its "if diagnosis fails" branch: a screensaver while
docked, which is the panel-protection consolation prize, not a new item.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

agent/BACKLOG.md

@@ -80,6 +80,15 @@ stays lit under hyprlock (journal: `dpms-off skipped: docked`); undocked
 still blanks at 10 min and wakes on input. Full original pass (wake after
 DPMS-while-docked) deferred to the LATER revisit.
 
+**Fallback if the cause resists (Bernardo 2026-07-16):** diagnosing stays the
+preference — DPMS when docked is the outcome we want. But the mitigation as
+shipped holds a near-static lock screen on the externals indefinitely, so if
+the cause cannot be pinned down, a **screensaver while docked** is the
+consolation prize: it protects the panels (burn-in — the same cost #135
+names) without re-entering the blanking path that bricked the session. Not a
+new item: it is the "if diagnosis fails" branch of this one, and #135's state
+key is where it would land.
+
 ## NEXT
 
 ### 135. Give the #127 DPMS mitigation a state key (and a way back)
@@ -181,6 +190,111 @@ migration (state.json created + git-staged, legacy `git rm`'d). Proven by
 hand 2026-07-15 (drvPaths identical across the rename) — this just keeps it
 proven. Delete together with the shim + the `nomarchy-theme-sync` alias.
 
+### 137. Plymouth splash draws on head 0's geometry, so it is off-centre when docked
+
+Bernardo 2026-07-16 (T14s, docked): the boot **and** shutdown logo is not
+centred on the external monitor. Cause is in the theme script, not the
+hardware: `modules/nixos/plymouth/nomarchy.script` calls `Window.GetWidth()`
+/ `Window.GetHeight()` with **no head index**, which returns head 0, while
+sprite coordinates live in a device-wide space spanning every head. So every
+element is placed at "head 0's centre" measured from the union origin — right
+on a single monitor, wrong the moment a second one exists, and wrong in a
+different direction depending on which head is 0 and how they are arranged.
+
+It is the whole splash, not just the logo: the same unindexed call places the
+logo (`:9`, `:15–16` — note the 15%-of-height *scale* also tracks the wrong
+head), the password entry (`:129`), the lock, the bullets, and the progress
+box/bar (`:210`, `:219`). Fix shape is the standard Plymouth one: loop
+`Window.GetX(i)` / `GetY(i)` / `GetWidth(i)` / `GetHeight(i)` over the heads
+and give each its own sprite set, so each monitor gets a centred, correctly
+scaled splash — the LUKS prompt especially, since a passphrase box half off
+the panel is the failure that matters. Pass = docked boot + shutdown centred
+on both heads; undocked unchanged.
+
+### 138. `[human]` Google Meet reports no mic or speakers; Zoom is fine
+
+Bernardo 2026-07-16 (T14s, docked): Meet insists there is no microphone or
+speaker even with the browser permission granted; Zoom calls work. What I
+could rule out from the live session, so the fix pass does not re-tread it:
+
+- The audio graph is healthy — PipeWire 1.6.5 with 4 sinks and 5 sources
+  (`wpctl status`), and a live `Chromium input` client was already attached.
+- Not a Flatpak sandbox: Chromium is the **native** nixpkgs build from the
+  template's `home.packages`. (Zoom, the app that works, is the Flatpak.
+  Brave — also Flatpak — holds `pulseaudio` + `xdg-run/pipewire-0` +
+  `devices=all`, so it would not fail this way either.)
+- One oddity worth checking first: the default source is **`OBSBOT Meet SE
+  Digital Stereo (IEC958)`** — an S/PDIF *passthrough* profile on the webcam,
+  a strange default for a mic. A device that enumerates but cannot open is a
+  plausible route to Meet's "no devices" wording.
+
+`[human]` because it needs the failure in front of us: **which browser** were
+you in (native Chromium, or the Brave Flatpak?), and with it open on Meet,
+`chrome://media-internals` plus the device list under `chrome://settings/content/microphone`
+would separate "Chromium enumerates nothing" from "Meet rejects what it is
+offered". Until then there is no way to tell a Nomarchy defect from a
+device-profile accident, and guessing costs more than asking.
+
+### 139. Terminal floats open full-screen — % size rules are ignored, and Kitty remembers
+
+Bernardo 2026-07-16: calcurse and nomarchy-doctor float but fill the screen —
+the reduced size we had under Ghostty is gone. Reproduced and root-caused live
+on the T14s (Hyprland 0.55.4, single 2560×1440 head), and it is **two** faults
+stacked, which is why it looks like the float rules are dead when they are not:
+
+1. **Percentage sizes silently no-op.** `com.nomarchy.calendar` came up
+   `floating=true` but `size=[2526,1360]` — the full work area, not the
+   `size 60% 65%` in `modules/home/hyprland.nix:1064`. A runtime rule with
+   **absolute** pixels (`size 1400 900`) applied perfectly and even centred
+   itself, so `size` works and the matcher works: it is the `%` form that this
+   Hyprland drops. No `hyprctl configerrors` — it fails silently, which is
+   how it survived the rule-syntax rewrite.
+2. **Kitty remembers its last window size.** With the rule not applying,
+   nothing overrides Kitty's own request, and Kitty defaults to
+   `remember_window_size yes` — `~/.cache/kitty/main.json` literally holds
+   `{"window-size": [...]}` and is replayed into every new OS window. So the
+   float inherits whatever the last Kitty was, which after any tiled window is
+   ~full-screen. **This is the Ghostty regression** (#95, Kitty-only): Ghostty
+   kept no such memory, so fault 1 was invisible — fault 2 exposed it.
+
+Both need fixing, and fixing either alone is a trap: absolute pixels alone
+stop being right on the 1366-wide Acer (#131's machine) or any scaled head,
+and pinning Kitty alone leaves the `%` rules rotting for the next window we
+float. Likely shape: `-o remember_window_size=no` (plus
+`initial_window_width/height`) in the `nomarchy-calendar` / doctor launchers
+so Kitty stops fighting us, and either the working syntax for `%` or a
+documented note that this Hyprland wants absolute px. Affects every rule using
+`%` — today calendar (`:1064`) and doctor (`:1070`). Pass = both sheets open
+centred at their intended fraction on the dev box **and** on the Acer.
+
+_(My probes wrote `{"window-size": [1400, 900]}` into your Kitty cache — it is
+a self-updating cache, harmless, and any resize overwrites it.)_
+
+### 141. Waybar's update module is dead in the whole-swap themes (`$TERMINAL` is unset)
+
+Bernardo 2026-07-16: clicking the Waybar updates module does nothing.
+Confirmed live and it is a **Waybar parity break** (CONVENTIONS § whole-swaps),
+not an updates bug. The generated module bakes the terminal in —
+`"${config.nomarchy.terminal} -e nomarchy-updates upgrade"`
+(`modules/home/waybar.nix:356`) — but the four whole-swap themes hand-write
+`"sh -c '$TERMINAL -e nomarchy-updates upgrade'"`, and **Waybar's environment
+has no `TERMINAL`**: its 55 vars contain zero matches (pid 1688, `/proc/*/environ`).
+`TERMINAL` is a `home.sessionVariables` entry (`modules/home/default.nix:104`),
+so login shells get it and the Hyprland-spawned bar does not. The click expands
+to `sh -c ' -e nomarchy-updates upgrade'` → nothing, no error, no toast. You
+are on **boreal**, a whole-swap, which is why you see it.
+
+Affected, one line each: `themes/{boreal,summer-day,summer-night,executive-slate}/waybar.jsonc`
+(`custom/updates.on-click` — grep says this is the *only* `$TERMINAL` use left
+in any whole-swap, so the blast radius is exactly these 4 lines). Fix = drop
+the env dependency the way calendar/doctor already do — name `kitty` outright,
+Kitty being sole supported since #95 — or, better and one place instead of
+five, give `nomarchy-updates` a subcommand that opens its own terminal so
+every caller stops needing `$TERMINAL`. Worth a check against the class of
+bug, not the instance: nothing today asserts that a whole-swap's on-click
+survives Waybar's actual environment, so the next hand-written jsonc can
+reintroduce it silently.
+
 ### 120. A netinstall ISO, next to the fat offline one
 
 Bernardo 2026-07-14, after seeing the measured size: **keep the current ISO
@@ -297,6 +411,36 @@ high-ROI, etc.) live in the journal + ROADMAP — not here.*
 
 ### Product / day-2
 
+### 140. Ask Claude should hand off to a web chat, not open a coding agent
+
+Bernardo 2026-07-16: "opening claude-code is too disruptive" — he would prefer
+the query go to a **web interface** (Gemini or Claude) instead. Today
+SUPER+CTRL+A → `nomarchy-menu ask` (`modules/home/rofi.nix:899–905`) takes the
+free-text line and does `${cfg.terminal} -e npx --yes @anthropic-ai/claude-code@latest "$q"`:
+a terminal window, an npm fetch on first run, and a REPL that stays open — a
+coding agent answering "what's the syntax for…". The prompt surface is right;
+the destination is wrong for the question the surface invites.
+
+Cheap version: URL-encode `$q` and `xdg-open` a prefill URL
+(`https://claude.ai/new?q=…`, `https://gemini.google.com/app?q=…`), landing in
+the default browser (Chromium, per Decisions). That is a handful of lines and
+deletes the `npx`/network/REPL cost — plus `pkgs.nodejs`, carried in
+`home.packages` (`:1685`) **only** for this feature, so the closure shrinks
+with it. To settle before building:
+- **Which destination, and is it a choice?** Bernardo named Claude *or*
+  Gemini. A `settings.ask.provider` key keeps it in-flake and menu-writable
+  (the convention) and stops us hardcoding a vendor into a vendor-neutral
+  distro; a fixed default is cheaper. Prefill-URL support is per-vendor and
+  can be changed by them unilaterally — worth verifying each still prefills,
+  since a silently-ignored `?q=` degrades to "opens a blank chat", which is
+  survivable but should be a known cost, not a surprise.
+- **Keep the CLI at all?** A second row (Ask ▸ web / Ask ▸ agent) preserves
+  today's behaviour for whoever wants it; deleting it is the simpler surface
+  and the one the complaint argues for. Bernardo's call.
+- Renaming: "Ask Claude" is the keybind desc (`keybinds.nix:57`) and shows in
+  the generated cheatsheet — if the destination becomes configurable, the
+  label should stop naming one vendor.
+
 ### 134. `unstable.<pkg>` in the downstream — a newer app without a second flake
 
 Bernardo 2026-07-15, wanting a newer LM Studio (pinned 0.4.15-2 vs unstable