Greptile Summary
\n\nThis PR wires up app catalog selections to open in managed Electrobun desktop windows, adds always-on-top controls for both managed surface windows and canvas/game windows, introduces concurrent-open coalescing for detached surfaces, and extends chat scope to cover Connectors, Plugins, and Settings views in detached windows. The previously flagged issues — silent no-op on `setAlwaysOnTop`, stale pin state on canvas window failure, and the `shouldUseHashNavigation` duplication — are all resolved in this patch.\nConfidence Score: 5/5
\n\nSafe to merge — all previously flagged issues are addressed and no new P0/P1 defects found.\n\nThe patch resolves every issue from the prior review thread: setAlwaysOnTop now returns {success:boolean} instead of silently no-opping, the alwaysOnTop state on canvasListWindows failure is handled in a try/catch, shouldUseHashNavigation duplication is eliminated by extracting shared utilities to navigation/index.ts, and the duplicate-surface-window guard is implemented via openSurfaceWindowOnce. New tests cover the concurrent-open coalescing, singleton guard, explicit webview loadURL, and the always-on-top toggle contract. No P1 or P0 issues found in the new code paths.\n\nNo files require special attention.\n\nImportant Files Changed
\n\n| Filename | Overview |\n|----------|----------|\n| packages/app-core/platforms/electrobun/src/surface-windows.ts | Adds openAppWindow, concurrent-open coalescing via pendingSurfaceWindows, setWindowAlwaysOnTop, and explicit webview loadURL. Singleton guard for detached surfaces is now present; loadURL deferred via setTimeout(0) to allow webview to initialize. |\n| packages/app-core/src/components/pages/AppsView.tsx | Adds AppWindowRecord tracking, heartbeat loop, desktop window event subscriptions, and the App Windows control strip. slugAutoLaunchDone.current is now set before the slug lookup to prevent re-launch after a manual catalog open updates the URL — intentional guard. |\n| packages/app-core/platforms/electrobun/src/rpc-handlers.ts | Adds desktopOpenAppWindow, desktopSetManagedWindowAlwaysOnTop, and canvasSetAlwaysOnTop RPC handlers. normalizeRendererRoutePath validates the incoming path is a renderer route before it reaches the URL builder. |\n| packages/app-core/platforms/electrobun/src/native/canvas.ts | Adds setAlwaysOnTop (returns {success: boolean}), exposes alwaysOnTop in listWindows output, and accepts alwaysOnTop in openGameWindow. Addresses the previously flagged silent no-op issue. |\n| packages/app-core/platforms/electrobun/src/native/desktop.ts | Adds openAppWindowCallback and managedWindowAlwaysOnTopCallback wiring. openSurfaceWindow now returns a ManagedWindowSnapshot rather than void, enabling callers to receive the created window ID. |\n| packages/app-core/platforms/electrobun/src/rpc-schema.ts | Schema-level additions: DesktopManagedWindowSnapshot, desktopOpenAppWindow, desktopSetManagedWindowAlwaysOnTop, canvasSetAlwaysOnTop, desktopManagedWindowsChanged push event, and alwaysOnTop on canvas/game types. |\n| packages/app-core/src/navigation/index.ts | Extracts shouldUseHashNavigation, getWindowNavigationPath, and isAppWindowRoute into shared utilities, eliminating the previously flagged duplication across AppsView, AppsPageView, AppContext, and useNavigationState. |\n| packages/agent/src/api/conversation-metadata.ts | Adds page-connectors, page-plugins, and page-settings to VALID_SCOPES. page-settings was already in the ConversationScope type union but was missing from the runtime validation set — this is a correct sync fix. |\n| packages/app-core/src/shell/DetachedShellRoot.tsx | Wraps connectors, plugins, triggers, and settings detached views in DetachedWorkspaceView to provide scoped chat context. Settings section gains overflow-y-auto for scroll. |\n| packages/app-core/src/components/apps/GameView.tsx | Adds always-on-top toggle button for game windows with try/catch guard on canvasListWindows fetch (addressing previous stale-state comment) and success-response check before updating local state. |\n| packages/app-core/platforms/electrobun/src/surface-windows.test.ts | New test file covering surface URL building, singleton guard, concurrent-open coalescing, and explicit webview loadURL. Good coverage of the new deduplication logic. |\n| packages/app-core/platforms/electrobun/src/native/canvas.test.ts | New test file covering openGameWindow with alwaysOnTop, setAlwaysOnTop toggle, and failure path for missing window IDs. Validates the { success: boolean } return contract. |\n\nSequence Diagram
\n\n```mermaid\nsequenceDiagram\n participant UI as AppsView (Renderer)\n participant Bridge as Desktop Bridge (RPC)\n participant SM as SurfaceWindowManager\n participant DM as DesktopManager\n participant CM as CanvasManager\n\n Note over UI: User clicks catalog app\n\n alt Internal tool app (plugins/connectors/browser/triggers)\n UI->>Bridge: desktopOpenSurfaceWindow(surface, alwaysOnTop)\n Bridge->>DM: openSurfaceWindow()\n DM->>SM: openSurfaceWindow(surface, browse, alwaysOnTop)\n SM->>SM: Check pendingSurfaceWindows (coalesce)\n SM->>SM: Check existing window (singleton guard)\n SM-->>UI: ManagedWindowSnapshot {id, alwaysOnTop}\n SM->>UI: desktopManagedWindowsChanged push event\n else External viewer app (game/canvas)\n UI->>Bridge: gameOpenWindow(url, title, alwaysOnTop)\n Bridge->>CM: openGameWindow(url, alwaysOnTop)\n CM-->>UI: {id}\n Note over UI: Heartbeat loop started\n else Generic app route\n UI->>Bridge: desktopOpenAppWindow(title, path, alwaysOnTop)\n Bridge->>SM: openAppWindow(title, path, alwaysOnTop)\n SM->>SM: buildAppWindowRendererUrl(rendererUrl, path)\n Note over SM: URL: host/?appWindow=1#/route\n SM-->>UI: ManagedWindowSnapshot {id, alwaysOnTop}\n SM->>UI: desktopManagedWindowsChanged push event\n end\n\n Note over UI: Toggle pin on open window\n alt Managed window\n UI->>Bridge: desktopSetManagedWindowAlwaysOnTop(id, flag)\n Bridge->>SM: setWindowAlwaysOnTop(id, flag)\n SM-->>UI: {success: boolean}\n else Canvas/game window\n UI->>Bridge: canvasSetAlwaysOnTop(id, flag)\n Bridge->>CM: setAlwaysOnTop(id, flag)\n CM-->>UI: {success: boolean}\n end\n```\nGreptile Summary
\n\nThis PR adds `tokenizeNameOccurrences` and wires it into `prepareDraftForSave` so that literal name text is re-tokenized to `{{name}}` on every save, making agent renames propagate without manual find-and-replace. The approach is sound, but there is a critical asymmetry between the save and load paths.\n\n- **Save path** now tokenizes `bio`, `system`, `topics`, `postExamples`, `style.all/chat/post`, and `messageExamples[].content.text`.\n- **Load path** (`loadCharacter` in `useCharacterState.ts`) applies `replaceNameTokens` only to `bio` and `system` — the four additional field groups are passed through raw. After the first save with this fix, users will see literal `{{name}}` tokens in the Topics, Style, Post Examples, and Message Examples editor fields.\n\nConfidence Score: 3/5
\n\nNot safe to merge — the save/load asymmetry will surface raw `{{name}}` tokens in the editor UI immediately after first save.\n\nA single clear P1 defect: the save path tokenizes six field groups but the load path only un-tokenizes two, creating an immediate and reproducible UX regression for every user who saves after this change ships.\n\n`packages/app-core/src/state/useCharacterState.ts` — `loadCharacter` must apply `replaceNameTokens` to `topics`, `style.*`, `postExamples`, and `messageExamples[].content.text` to match the expanded save-side coverage.\n\nImportant Files Changed
\n\n\n\n\n| Filename | Overview |\n|----------|----------|\n| packages/app-core/src/state/useCharacterState.ts | Threads `characterData?.name` as `previousName` into `prepareDraftForSave`, but the `loadCharacter` callback only applies `replaceNameTokens` to `bio` and `system` — not to the six field groups now tokenized on save — causing raw `{{name}}` tokens to appear in the editor UI after first save. |\n| packages/app-core/src/character/character-draft-helpers.ts | Adds `previousName` param and applies tokenization across all free-text fields on save; logic is sound, but correctness depends on the load path being symmetrically updated (which it isn't for several field groups). |\n| packages/app-core/src/utils/name-tokens.ts | Adds `tokenizeNameOccurrences` — case-sensitive, whole-word, idempotent, with guardrails for short/empty names and regex-special chars. Word boundary behaviour is limited to ASCII names; non-ASCII names may not tokenize reliably. |\n| packages/app-core/src/utils/name-tokens.test.ts | Good unit coverage of `tokenizeNameOccurrences` and `prepareDraftForSave` field tokenization; missing round-trip tests verifying load-side expansion for `topics`, `style`, `postExamples`, and `messageExamples`. |\n\n\n\n\n\nSequence Diagram
\n\n```mermaid\nsequenceDiagram\n participant UI as Editor UI\n participant State as useCharacterState\n participant Helpers as character-draft-helpers\n participant API as Backend API\n\n Note over UI,API: Save flow (this PR)\n UI->>State: handleSaveCharacter()\n State->>Helpers: prepareDraftForSave(draft, previousName)\n Note over Helpers: tokenizes bio, system, topics,postExamples, style.*, messageExamples\n Helpers-->>State: tokenized draft\n State->>API: updateCharacter(tokenizedDraft)\n API-->>State: success\n State->>State: loadCharacter()\n\n Note over UI,API: Load flow (this PR — unchanged)\n State->>API: getCharacter()\n API-->>State: character data (with {{name}} tokens)\n Note over State: clean() applied to bio + system only\n Note over State: topics / style / postExamples /
messageExamples NOT cleaned ⚠️\n State->>UI: setCharacterDraft (raw {{name}} tokens visible in editor)\n```\n\n\n
Comments Outside Diff (1)
\n\n1. `packages/app-core/src/state/useCharacterState.ts`, line 86-101 ([link](https://github.com/elizaos/eliza/blob/7cb0d48b06af3b7ab64803113994ac619151d18b/packages/app-core/src/state/useCharacterState.ts#L86-L101)) \n\n ![\"P1\"](\"https://greptile-static-assets.s3.amazonaws.com/badges/p1.svg?v=7\")
**Load path doesn't un-tokenize `topics`, `style`, `postExamples`, `messageExamples`**\n\n `loadCharacter` applies `clean` (i.e. `replaceNameTokens`) to `bio` and `system` only, but `prepareDraftForSave` now tokenizes six additional field groups — `topics`, `postExamples`, `style.all/chat/post`, and `messageExamples[].content.text`. After the first save with this fix, those fields will be persisted with `{{name}}` tokens. On reload, the draft will contain raw token strings such as `\"how {{name}} thinks\"`, which the editor will display verbatim to the user.\n\n ```ts\n // before setting the draft, add:\n const cleanArr = (arr: string[]) => arr.map(clean);\n\n setCharacterDraft({\n name: savedName,\n username: character.username ?? \"\",\n bio: Array.isArray(character.bio)\n ? character.bio.map(clean).join(\"\\n\")\n : clean(character.bio ?? \"\"),\n system: clean(character.system ?? \"\"),\n adjectives: character.adjectives ?? [],\n topics: cleanArr(character.topics ?? []),\n style: {\n all: cleanArr(character.style?.all ?? []),\n chat: cleanArr(character.style?.chat ?? []),\n post: cleanArr(character.style?.post ?? []),\n },\n messageExamples: (character.messageExamples ?? []).map((group) => ({\n ...group,\n examples: group.examples.map((ex) => ({\n ...ex,\n content: { ...ex.content, text: clean(ex.content.text) },\n })),\n })),\n postExamples: cleanArr(character.postExamples ?? []),\n });\n ```\n\nGreptile Summary
\n\nThis PR adds `tokenizeNameOccurrences` to reverse the load-side `replaceNameTokens` expansion on save, so character renames propagate through all free-text fields. The core logic and tests are solid, but the save-side tokenization is not symmetric with the load-side de-tokenization.\n\n- **P1 — Editor shows raw `{{name}}` after save**: `prepareDraftForSave` now tokenizes `topics`, `postExamples`, `style.*`, and `messageExamples[].content.text`, but `loadCharacter` in `useCharacterState.ts` only calls `replaceNameTokens` on `bio` and `system`. Because `handleSaveCharacter` calls `loadCharacter()` on success, these fields immediately reload with raw `{{name}}` template syntax visible in the editor UI.\n\nConfidence Score: 3/5
\n\nNot safe to merge — the load/save asymmetry will cause raw {{name}} tokens to appear in the editor immediately after any save.\n\nOne P1 defect: loadCharacter in useCharacterState.ts does not de-tokenize the four field groups that prepareDraftForSave now tokenizes. Because handleSaveCharacter calls loadCharacter() on success, users will see {{name}} as literal text in the topics, post-examples, style, and message-example editors after every save.\n\npackages/app-core/src/state/useCharacterState.ts — the loadCharacter callback (lines 86–102) must be extended to apply replaceNameTokens to all newly-tokenized fields.\n\nImportant Files Changed
\n\n\n\n\n| Filename | Overview |\n|----------|----------|\n| packages/app-core/src/state/useCharacterState.ts | Adds characterData to the dependency array and passes it as previousName to prepareDraftForSave, but loadCharacter only de-tokenizes bio and system — leaving topics, postExamples, style.*, and messageExamples with raw {{name}} tokens visible in the editor after each save. |\n| packages/app-core/src/character/character-draft-helpers.ts | Correctly introduces a tokenize closure that covers all free-text fields on save; adjectives are intentionally excluded. Implementation is well-structured. |\n| packages/app-core/src/utils/name-tokens.ts | New tokenizeNameOccurrences is case-sensitive, whole-word, idempotent, and safely escapes regex metacharacters; single-char guard is correct. |\n| packages/app-core/src/utils/name-tokens.test.ts | Comprehensive 15-case test suite covering tokenization, idempotency, edge guards, and round-trip with replaceNameTokens; no issues found. |\n\n\n\n\n\nSequence Diagram
\n\n```mermaid\nsequenceDiagram\n participant Editor\n participant useCharacterState\n participant prepareDraftForSave\n participant Server\n\n Editor->>useCharacterState: handleSaveCharacter()\n useCharacterState->>prepareDraftForSave: draft + previousName\n prepareDraftForSave->>prepareDraftForSave: tokenize(bio, system, topics, postExamples, style.*, messageExamples)\n prepareDraftForSave-->>useCharacterState: tokenized draft\n useCharacterState->>Server: updateCharacter(draft)\n Server-->>useCharacterState: success\n useCharacterState->>useCharacterState: loadCharacter()\n useCharacterState->>Server: getCharacter()\n Server-->>useCharacterState: character with {{name}} in all fields\n useCharacterState->>useCharacterState: clean(bio) ✅ clean(system) ✅ topics as-is ❌ postExamples as-is ❌ style.* as-is ❌ messageExamples as-is ❌\n useCharacterState-->>Editor: draft with raw {{name}} visible in topics/style/etc\n```\n\n\nComments Outside Diff (1)
\n\n1. `packages/app-core/src/state/useCharacterState.ts`, line 86-102 ([link](https://github.com/elizaos/eliza/blob/406f30fea4094536e09fce7c72935c8b668edf2b/packages/app-core/src/state/useCharacterState.ts#L86-L102)) \n\n **Load-side de-tokenization is incomplete — users will see raw `{{name}}` in the editor**\n\n `prepareDraftForSave` now tokenizes `topics`, `postExamples`, `style.*`, and `messageExamples[].content.text`, but `loadCharacter` (lines 86–102) only calls `clean` on `bio` and `system`. After a successful save, `handleSaveCharacter` calls `loadCharacter()` (line 137), which reloads the character without de-tokenizing those four field groups — so the user immediately sees `{{name}}` as raw text in the topics, style, post-examples, and message-example editors.\n\n Concrete path: save character \"Momo\" with topic `\"how Momo thinks\"` → stored as `\"how {{name}} thinks\"` → `loadCharacter` puts it back into the draft untouched → editor renders `\"how {{name}} thinks\"`.\n\n Fix: extend the `clean` pass in `loadCharacter` to cover all fields that `prepareDraftForSave` now tokenizes:\n\n ```\n topics: (character.topics ?? []).map(clean),\n postExamples: (character.postExamples ?? []).map(clean),\n style: {\n all: (character.style?.all ?? []).map(clean),\n chat: (character.style?.chat ?? []).map(clean),\n post: (character.style?.post ?? []).map(clean),\n },\n messageExamples: (character.messageExamples ?? []).map((group) => ({\n ...group,\n examples: group.examples.map((ex) => ({\n ...ex,\n content: { ...ex.content, text: clean(ex.content.text) },\n })),\n })),\n ```\n\n