docs(fork): apply review nits from the PR 26/28/29 review pass

.fork/customizations.yaml

@@ -611,11 +611,9 @@
     it shares with these rows spends 8px of air on the left and 8+6 on the
     right: every card sits visibly off-centre in its own column, and the chrome
     rows above end 6px past where the cards do. The list's end padding gives
-    that 6px back — reading --app-scrollbar-width rather than restating 6px, so
-    the symmetry follows upstream if the scrollbar is ever retuned. The gutter
-    itself stays: the scrollbar keeps its lane, the cards are 8px from both
-    edges, and the compensation is meaningless without the gutter that causes
-    it, so the two must move together.
+    that 6px back. The gutter itself stays: the scrollbar keeps its lane, the
+    cards are 8px from both edges, and the compensation is meaningless without
+    the gutter that causes it, so the two must move together.
 
     The width given back is measured, not assumed. --app-scrollbar-width is only
     true where ::-webkit-scrollbar applies: Firefox sets no scrollbar-width, so
@@ -682,9 +680,12 @@
     - apps/web/src/components/SidebarStageBackdrop.test.tsx
     - apps/web/src/components/SidebarV2.tsx
     - apps/web/src/components/sidebar/SidebarChrome.tsx
-    # Defines --app-scrollbar-width, which the thread list's end padding reads
-    # to cancel its own scroll gutter. Retiring or renaming that token silently
-    # returns the list to 8px of air on one side and 14px on the other.
+    # Styles ::-webkit-scrollbar { width: var(--app-scrollbar-width) }, which
+    # is what makes the thread list's reserved lane a styled 6px at all — and
+    # what the intent's cross-engine argument is anchored to. A retune the
+    # measurement absorbs; the failure is this rule disappearing or gaining
+    # scrollbar-width/appearance handling, which hands the lane to the UA
+    # default width the drawn 6px gutter does not assume.
     - apps/web/src/index.css
   verify:
     - apps/web/src/__fork_guards__/forkSidebarChrome.test.ts
@@ -1021,8 +1022,10 @@
     # customization switches off. An upstream rework of that paint would leave
     # the fork disabling rules that no longer exist.
     - apps/web/src/index.css
-    # min-h-17.5 on the prompt editor, and the placeholder that has to track the
-    # prompt's type ramp.
+    # min-h-17.5 on the prompt editor, the placeholder that has to track the
+    # prompt's type ramp, and the max-h-50 / overflow-y-auto cap the fork
+    # re-homes on the prompt wrapper at >=40rem by overriding — the classes
+    # themselves stay on the editor here.
     - apps/web/src/components/ComposerPromptEditor.tsx
   verify:
     - apps/web/src/__fork_guards__/forkComposerShell.test.ts

apps/web/src/__fork_guards__/forkSidebarChrome.test.ts

@@ -182,7 +182,16 @@ describe("fork guard: fork-sidebar-chrome", () => {
     const gutter = readSibling("../custom/useScrollGutterWidth.ts");
     expect(gutter).toContain("offsetWidth - node.clientWidth");
     expect(sidebarV2).toContain("ref={listScrollGutterRef}");
-    expect(sidebarV2).not.toContain("calc(0.5rem-var(--app-scrollbar-width))");
+    // The token-based giveback is this file's own prior implementation —
+    // pe-[calc(0.5rem-var(--app-scrollbar-width))], replaced by the measured
+    // gutter in 56954bc8 — which makes it the likeliest thing a revert or a
+    // wrong-side conflict resolution puts back, including *alongside* the
+    // measured term, where the containment checks above stay green and the
+    // last pe-* wins. Bounded to a pe-* utility so prose about the token
+    // cannot trip it, and looser than the old exact literal so a re-spelling
+    // like pe-[calc(var(--sidebar-list-pad)-var(--app-scrollbar-width))]
+    // cannot sail past.
+    expect(sidebarV2).not.toMatch(/pe-\[[^\]]*--app-scrollbar-width/u);
   });
 
   it("puts the brand on the header's trailing edge", () => {