Design: a verb's address is data, so the toolbar widget can stop existing

snaporca-7ih's remaining half. Both flyout factories registered their verbs INSIDE the
widget-building loop, so the ~40 retired tool buttons had to be constructed and then
Hide()n: skipping construction would have deleted 42 offer verbs (26 fly:<family>#<row>
+ 16 Shift+keys) while their rows still rendered and did nothing when picked.

Register first, build second. The addresses are pure data; the widget is one door onto
them, not their owner. A family absent from kBarKeep now returns before any wxWindow is
made. The keep-list stays a one-line data decision, not a structural one.

And close the class of bug for good: the constructor now verifies, once, that every verb
the atlas marks wired resolves to a real registration, logging each break and asserting in
debug. Rows that render and do nothing have shipped three times (edit_feature and sk_move
with action:null, then this) and are invisible from either side alone.

Verified on the snaporca rig by walking the offer, not by reading the code — all four
at-risk address kinds run with no widget behind them: fly:design_rect#2 drew an OBLIQUE
rectangle (the third variant, not the family's first), key:S+E opened Extrude with its
10 mm gizmo, fly:material#4 opened Thicken. Hover hints, icons and nesting intact. This
fork is code-identical here bar the two permitted DropDown divergences; it still owes a
build of its own (snaporca-5pl).

Two hints were wrong and are fixed: Cut said "Split the body with a plane", colliding with
the Split verb one row away and pointing at a card for a value the canvas already offers as
a draggable arrow; Split never said its plane comes from a picked face.

Also, because it blocked the verification and will block the next one: gui-session.sh
killed by full path while its own app_pid() matched by basename, so a differently-pathed
instance survived, held the single-instance lock, and got reported as a healthy session —
a Jul-30 binary nearly passed as this build. It now kills by basename and prints which
binary is actually on screen. Traps 6 and 7 documented.

docs/rig_build_traps.md

@@ -121,3 +121,31 @@ invisible and the session hunts a phantom. Pass `BIN` explicitly:
 Note also that the GUI containers do **not** mount `scripts/`: `/OrcaSlicer/scripts` inside them
 is the baked copy, so a local edit to `gui-session.sh` has no effect until you
 `docker cp scripts/gui-session.sh <container>:/OrcaSlicer/scripts/`.
+
+---
+
+## Trap 6 — `src/Release/` resolves resources to `build/resources`, which may not exist
+
+The binary derives `resources_dir()` from its own location, so the `src/Release/` one looks in
+`/OrcaSlicer/build/resources` while the packaged one looks inside `build/package/`. Only the
+packaging step creates the latter; nothing creates the former. Without it the app fails every
+`Failed to add custom font ".../build/resources/fonts/…"`, logs `Health check is not running`,
+and **exits 255 with nothing on stdout** — which reads exactly like a crash in whatever you just
+changed. Measured 2026-08-02: an hour was nearly spent bisecting a GUI change that was fine.
+
+`build/` is the shared cache volume, so one symlink fixes it permanently, and pointing it at the
+bind-mounted repo tree means the rig also picks up new `resources/images/*.svg` without a rebuild:
+
+    docker exec <fork>-gui ln -sfn /OrcaSlicer/resources /OrcaSlicer/build/resources
+
+Tell the two apart before debugging: a resource failure dies in the first second with no window;
+a real fault in your code gets past the version banner. Compare
+`~/.config/<App>/log/<newest>.log.0` against a known-good run — 47 lines versus 340 is the tell.
+
+## Trap 7 — a single-instance app plus a path-matched `pkill`
+
+`gui-session.sh` used to kill by `"$BIN"`, while its own `app_pid()` matched by BASENAME. Launch
+with a `BIN` that differs from the running instance's path and the old process survives, keeps
+the single-instance lock, and the new one exits seconds after loading fonts — then `status`
+reports the *stale* pid as a healthy session. Fixed by killing on the basename; `status` now also
+prints `binary : $(readlink -f /proc/<pid>/exe)`. **Read that line before trusting a screenshot.**

docs/ux/tool_atlas.json

@@ -567,7 +567,7 @@
       "gui": true,
       "action": "key:S+X",
       "icon": "design_cut",
-      "hint": "Split the body with a plane; set the offset in the card"
+      "hint": "Trim the body with a plane — drag the offset arrow; keep one half or both"
     },
     {
       "id": "split",
@@ -586,7 +586,7 @@
       "gui": false,
       "action": null,
       "icon": null,
-      "hint": "Split the body into separate solids"
+      "hint": "Split the body along a picked face into two solids"
     },
     {
       "id": "fillet",