Purpose: convert the current user repro report into issue-ready stubs and lane-aligned slices without losing context.

Source: in-session user field report (Windows 11 primary runtime, plus WSL/WSLg environment observations).

Lane	Priority	Why
lane:stabilization (#88)	Immediate / blocking	Graph canvas pan/zoom/fit controls still fail globally; tab/pane focus/render activation is inconsistent.
lane:control-ui-settings (#89)	Immediate planning / next execution	Command palette semantics/context behavior and theme toggle are user-facing IA gaps with clear direction.
lane:embedder-debt (#90)	Parallel root-cause lane	Legacy context menu/new-tab paths and render/focus activation races smell like servoshell inheritance paths.
lane:layout-semantics	Design follow-on (queue now)	Tile/pane/workspace/workbench semantic distinctions and overview UX need an explicit model.
lane:viewer-platform (#92) / lane:spec-code-parity (#99)	Secondary	Legacy webview context menu behavior does not match Graphshell command/pane semantics.

Title:

• Stabilization: graph canvas pan/wheel zoom/zoom commands no-op across contexts

Summary:

• Graph canvas interaction is partially alive (select node, lasso), but camera/navigation controls fail broadly:
• pan drag, wheel zoom, zoom in/out/reset, and zoom to fit fail anywhere the user can currently interact with the graph.

Repro (current report):

• Launch app on Windows 11.
• Open default graph pane.
• Try drag-pan, mouse wheel zoom, command/shortcut zoom actions, and zoom-to-fit.
• Result: node creation and some shortcuts work, but graph camera controls do not.

Likely hotspots:

• render/mod.rs
• app.rs
• input/mod.rs
• shell/desktop/ui/gui.rs
• graph metadata/input ownership paths (egui_graphs integration)

Notes / hypotheses:

• Recent fixes addressed targeted camera command ownership and lasso metadata keying.
• Remaining failure likely sits in input gating/consumption, graph metadata availability, or a stuck UI-state gate (radial/palette/focus ownership).

Non-goals:

• Lasso UX improvements (track separately)
• Command palette feature redesign

Done gate:

• Drag-pan, wheel zoom, zoom in/out/reset, and zoom-to-fit all work in the default graph pane.
• Behavior remains correct after opening/closing node panes and after focus changes.
• Add targeted regression tests and diagnostics receipt (camera command application + input path visibility).

Title:

• Stabilization: new tab/pane focus activation race leaves viewport blank until follow-up focus changes

Summary:

• Newly opened tabs/panes sometimes do not render content immediately.
• Behavior varies with focus state: sometimes content appears after click/switch, sometimes not on first click.

Repro (current report):

• Open tabs from node/link flows.
• Observe that some tabs spawn unfocused or visually blank.
• Switching tabs or repeated clicks can cause render to appear later.
• Deleting another pane can leave the graph pane unfocused with an empty-looking workspace.

Likely hotspots:

• shell/desktop/ui/gui.rs
• shell/desktop/ui/gui_frame.rs
• shell/desktop/workbench/* focus/pane activation paths
• shell/desktop/lifecycle/webview_controller.rs
• shell/desktop/workbench/tile_runtime.rs

Architectural context:

• This looks like focus ownership + render activation ordering debt, not just a local paint bug.
• Likely overlaps with servoshell-derived host/frame assumptions (lane:embedder-debt).

Done gate:

• New tabs/panes consistently render on first spawn when intended to be focused.
• Focus transitions after pane deletion promote a sensible next pane (graph pane included) and render immediately.
• Repro captured in scenario test or diagnostics receipt.

Title:

• Stabilization: graph selection deselect-on-background-click is inconsistent

Summary:

• Node selection works, but deselecting by clicking away feels inconsistent/funky and may hide deeper selection-state logic bugs.

Scope:

• Audit selection state transitions for:
• background click deselect
• multi-select interactions
• lasso + click-away transitions
• focus changes between graph pane and node pane

Likely hotspots:

• render/mod.rs
• input/mod.rs
• selection state in app.rs

Done gate:

• Background click deselect behavior is deterministic and documented.
• Selection state transitions are covered by targeted tests for single-select, multi-select, and lasso-to-click sequences.

Title:

• Stabilization: lasso boundary hit-testing misses nodes at selection edge

Summary:

• Lasso works and is visually strong, but it sometimes misses nodes near the edge of the lasso box.
• User expectation: node center inside lasso should count as a hit, even if visual radius crosses boundary.

Likely hotspots:

• render/mod.rs
• render/spatial_index.rs
• graph node visual bounds / hit proxy calculations

Done gate:

• Define and document lasso inclusion semantics (center-point inclusive minimum).
• Boundary/edge-node regressions covered by tests.

Title:

• Control UI: lasso drag preview highlights nodes entering/leaving capture set in real time

Summary:

• Add live visual feedback during lasso drag so nodes highlight when captured and unhighlight when released before mouse-up.

Why now:

• This improves confidence and makes lasso hit-testing problems more visible.
• It is a UX improvement, not a blocker for lasso correctness.

Likely hotspots:

• render/mod.rs
• graph styling/selection preview state

Done gate:

• During lasso drag, preview highlight updates continuously and matches final selection result.
• Reduced-motion / visual clarity considered.

Title:

• Control UI: unify F2 command palette and right-click context command surface across canvas/panes/workbench

Summary:

• Current command surfaces are fragmented:
• F2 summons a menu labeled Edge Commands
• right-click often shows a short legacy context menu
• context availability varies by hover target (too node-biased)

Desired direction (from report):

• F2: larger global command palette
• Right-click: compact contextual command palette
• Both backed by the same action/registry-driven command surface model

Likely hotspots:

• render/command_palette.rs
• render/mod.rs
• input/mod.rs
• action registry / runtime registries integration
• webview context menu bridge paths

Done gate:

• One command surface model supports both global and contextual invocation variants.
• Labeling no longer says Edge Commands unless truly edge-specific.
• Palette can be invoked from canvas, nodes, edges, panes, and workbench/workspace chrome (with context-appropriate command enablement).

Title:

• Control UI: contextual command palette categories (node/edge/tile/pane/workbench/canvas) with disabled-state policy

Summary:

• Formalize the contextual command palette information architecture:
• categories represent actionable entities in the current UI context
• unavailable commands/categories are shown disabled or deprioritized instead of disappearing unpredictably

Architectural value:

• Serves as a concrete manifestation of action/layout/mod registries in UI.
• Reduces context-menu inconsistency and discoverability debt.

Done gate:

• Category model documented and implemented for core contexts.
• Disabled/hidden policy is consistent and testable.

Title:

• Control UI / Settings: add theme mode toggle (dark, light, system)

Summary:

• Add a user-facing settings toggle for theme mode selection:
• System, Light, Dark

Likely hotspots:

• settings UI surface
• theme registry / theme application path
• persistence/preferences

Done gate:

• Theme mode can be changed in settings and persists across restart.
• System mode follows OS preference on supported platforms.

Title:

• Control UI: radial menu option spacing/readability pass

Summary:

• Radial menu works but options are visually crowded.
• Needs spacing/legibility polish before it becomes a primary interaction surface.

Done gate:

• Option spacing and hit targets are improved on desktop DPI ranges.
• No regression to command dispatch behavior.

Title:

• Control UI: omnibar node-search Enter action retains focus for result iteration

Summary:

• Ctrl+F search is useful but loses omnibar text focus after Enter when iterating node search results, forcing repeated re-clicks.

Likely hotspots:

• shell/desktop/ui/toolbar/toolbar_ui/toolbar_omnibar.rs
• toolbar focus routing

Done gate:

• Enter-based result iteration keeps focus in omnibar when in search mode (unless explicitly committing navigation).
• Behavior is documented/tested for node-search vs navigation submit modes.

Title:

• Design slice: scoped undo/redo model (pane history vs graph/workbench action history)

Summary:

• User reports current undo/redo behavior as interesting but buggy and semantically ambiguous.
• Need a design decision on whether to separate:
• pane-local back/forward navigation
• graph/workbench edit undo/redo
• potentially workspace layout history

Deliverable:

• Decision note defining history scopes, UI affordances, and command bindings.

Done gate:

• Chosen model documented with examples and migration path from current behavior.
• Follow-on implementation issues seeded.

Title:

• Embedder debt: web content new-tab/context-menu path bypasses Graphshell node creation and pane semantics

Summary:

• Right-click/ctrl-clicking links in webpage content (e.g., servo.org) can open a tile/tab path that appears to bypass node creation and may produce panes without a mapped node.
• A short legacy context menu still appears instead of the Graphshell command surface.

Why this matters:

• Breaks the graph-as-authoritative model.
• Makes split/workspace testing harder because behavior depends on legacy webview paths.

Likely hotspots:

• shell/desktop/ui/gui.rs
• shell/desktop/host/*
• shell/desktop/lifecycle/webview_controller.rs
• shell/desktop/workbench/tile_runtime.rs
• webview context menu / open-new-view handlers

Done gate:

• Link open actions from web content route through Graphshell node/pane creation semantics.
• Legacy fallback context menu path is either removed, bridged into command palette, or explicitly constrained/deferred.

Title:

• Embedder/platform tracker: WSLg window drop shadow remains at initial viewport bounds after resize/fullscreen

Summary:

• In VS Code WSL session (cargo run via bash), app window shadow/decorations remain stuck to original launch bounds and position.
• Repro persists across window operations and fullscreen/maximize.
• Does not repro in native Windows run.

Notes:

• Likely WSLg/Xwayland/Wayland window-decoration/compositor interaction, not core Graphshell logic.
• Track separately so it does not block cross-platform app logic stabilization.

Done gate:

• Repro is documented with environment details.
• Marked as upstream/toolkit/platform issue or mitigated if app-side knob exists.

The report gives a concrete product requirement that aligns with current architecture:

• action/mod/layout registries should become the source of command availability and grouping
• command palette is not just a search box; it is a contextual command surface

Action:

• Fold this into lane:control-ui-settings issue scopes and 2026-02-24_control_ui_ux_plan.md.

The report describes a good split:

• F2: larger global command palette
• right-click: compact contextual variant near pointer

This reduces the false choice between “palette” and “context menu” and lets both share one backend model.

Questions raised in the report are not polish questions; they are model questions:

• what is a new tile vs a new workspace
• how does a user get back to the root workbench overview
• how should workspaces map to the tile tree and graph

Action:

• Track as lane:layout-semantics design/UX slice (with lane:control-ui-settings overlap for command surfacing).

The report points toward a multi-scope model:

• pane-local navigation history (Back/Forward)
• graph/workbench structural edits (Undo/Redo)

Action:

• Do a design slice first before more bugfix patching in this area.

The report explicitly notes difficulty testing split-view behavior because focus and legacy paths interfere.

Action:

• Add stabilization diagnostics/receipts for focus ownership and pane activation to reduce manual repro ambiguity.

1. lane:stabilization: graph canvas camera controls fail globally
2. lane:stabilization: tab/pane focus activation race (blank viewport)
3. lane:embedder-debt: web content new-tab/context-menu bypasses node creation
4. lane:control-ui-settings: command palette/context menu unification
5. lane:control-ui-settings: theme mode toggle + omnibar focus retention (can be split)

• Prefer creating these as child issues under #88, #89, and #90.
• Cross-link the lasso preview issue to the lasso hit-testing stabilization issue so UX work does not hide correctness regressions.
• Keep the WSLg shadow artifact as a platform tracker, not a core stabilization blocker.