@agent-native/core 0.0.0-beta-20260820155159 → 0.0.0-beta-20260820171000

package/corpus/README.md

@@ -31,4 +31,4 @@ rg -n "defineAction|useActionQuery" node_modules/@agent-native/core/corpus
 
 ## Generated Counts
 
-- template files: 8485
+- template files: 8490

package/corpus/templates/analytics/changelog/2026-08-20-staged-data-limit-recovery.md

@@ -0,0 +1,5 @@
+---
+type: improved
+date: 2026-08-20
+---
+Analytics now explains how to recover when staged data reaches its size limit.

package/corpus/templates/clips/changelog/2026-08-20-cross-tab-recording-controls.md

@@ -0,0 +1,5 @@
+---
+type: improved
+date: 2026-08-20
+---
+Recording controls now follow you across browser tabs and page navigations.

package/corpus/templates/clips/changelog/2026-08-20-overlay-follow-review-fixes.md

@@ -0,0 +1,5 @@
+---
+type: fixed
+date: 2026-08-20
+---
+Clips overlays now follow tab switches during the recording countdown and recover cleanly on tabs opened before Clips.

package/corpus/templates/clips/changelog/2026-08-20-provider-key-recovery.md

@@ -0,0 +1,5 @@
+---
+type: fixed
+date: 2026-08-20
+---
+Clips now opens provider setup instead of retrying a rejected key in a loop.

package/corpus/templates/clips/chrome-extension/PERMISSIONS.md

@@ -1,41 +1,45 @@
 # Permissions model
 
-This extension ships **activeTab-only for recording overlays**: it deliberately
-does **not** request the broad `<all_urls>` host permission for recording UI.
+This extension ships **cross-tab follow for recording overlays**: it declares
+the broad `<all_urls>` host permission and a matching declarative content script
+so the camera bubble and controls can survive tab switches and navigations.
 
 ## Why
 
 A declarative content script on `<all_urls>` (and the matching `<all_urls>` host
 permission) triggers Chrome Web Store's **broad host permission in-depth review**,
-which significantly delays publishing and updates. We avoid that.
+which significantly delays publishing and updates. We take that review path only
+for the release that re-enables cross-tab follow.
 
-The only declarative content script is scoped to `https://github.com/*`. It
-looks for Clips links in GitHub issue/PR markdown and replaces them with a
-playable preview iframe owned by the extension. That adds a narrow GitHub host
-disclosure, but does not grant broad access to every page.
+The declarative content scripts are scoped to:
+
+- `https://github.com/*` for Clips link previews in GitHub issue/PR markdown.
+- `<all_urls>` for the recording overlay host that keeps the face bubble and
+  controls mounted across pages while a recording is active.
+
+The GitHub preview script adds a narrow host disclosure, while the overlay host
+is the broad review boundary.
 
 ## How the overlay gets on the page
 
 When the user clicks the extension and starts a recording, the background service
-worker injects the content script into the **launch tab** with
-`chrome.scripting.executeScript({ target: { tabId }, files: ["assets/content-script.js"] })`.
-That call is authorized by the **`activeTab`** permission, which Chrome grants for
-the tab that was active when the user invoked the extension — no broad host access
-needed. The content script then mounts the overlay iframes (countdown, camera
-bubble, controls). `web_accessible_resources` stays `<all_urls>` — that is **not**
-a host permission and does not trigger the review.
+worker mounts the overlay on the active tab and keeps rebroadcasting it as the
+active tab changes. The content script then mounts the overlay iframes
+(countdown, camera bubble, controls). `web_accessible_resources` stays
+`<all_urls>` — that is **not** a host permission and does not trigger the review.
 
 Declared permissions: `activeTab`, `debugger`, `offscreen`, `scripting`,
-`storage`. Host permissions: only the configured Clips app + `forms.agent-native.com`
+`storage`. Host permissions: the configured Clips app + `forms.agent-native.com`
 
 - `localhost`/`127.0.0.1`.
-- `https://github.com/*` is content-script scoped for link previews only.
+- `https://github.com/*` is content-script scoped for link previews.
+- `<all_urls>` is the overlay host required for cross-tab follow.
 
 ## What this costs
 
-- The overlay (countdown, camera bubble, recording controls) lives **only on the
-  tab the recording was launched from**. It does **not** follow the user to other
-  tabs during a full-screen or multi-tab recording.
+- The overlay (countdown, camera bubble, recording controls) follows the user
+  across tabs while a recording is active. Navigating to a new page or tab keeps
+  the face bubble and controls mounted once the page loads.
 - **Recording itself is unaffected.** Capture runs in the offscreen document via
   `getDisplayMedia`, independent of any tab's content script — full-screen,
   window, and other-tab content are all still captured normally. Only the on-page
@@ -43,36 +47,15 @@ Declared permissions: `activeTab`, `debugger`, `offscreen`, `scripting`,
 
 ## Re-enabling cross-tab follow
 
-The full cross-tab behavior is gated behind a single flag in
-`src/background.ts`:
+The full cross-tab behavior is gated behind a single flag in `src/background.ts`:
 
 ```ts
-const CROSS_TAB_FOLLOW = false; // flip to true
+const CROSS_TAB_FOLLOW = true;
 ```
 
-Flipping it to `true` restores the all-tabs broadcast in `broadcastMount()` /
-`broadcastUnmount()` and the `chrome.tabs.onActivated` follow listener. To make
-that actually work you must **also** restore broad host access so the worker can
-inject into arbitrary tabs. Two options:
-
-1. **Static (simplest, but in-depth review):** re-add to `public/manifest.json`:
-   - `"<all_urls>"` in `host_permissions`, and
-   - the `content_scripts` block:
-     ```json
-     "content_scripts": [
-       {
-         "matches": ["<all_urls>"],
-         "js": ["assets/content-script.js"],
-         "run_at": "document_idle",
-         "all_frames": false
-       }
-     ],
-     ```
-
-2. **Optional (preferred — keeps the default install lean):** declare
-   `"optional_host_permissions": ["<all_urls>"]` in the manifest and call
-   `chrome.permissions.request({ origins: ["<all_urls>"] })` at runtime (e.g. from
-   a settings toggle) before enabling follow. The user grants the broad access
-   explicitly, only if they want cross-tab overlays.
+That keeps the all-tabs broadcast in `broadcastMount()` / `broadcastUnmount()`
+and the `chrome.tabs.onActivated` follow listener live. The manifest must keep
+the broad host permission and the `<all_urls>` declarative content script in
+place so the worker can inject into arbitrary tabs.
 
 After any change: `pnpm build`, then re-zip `dist/`.

package/corpus/templates/clips/chrome-extension/STORE_DRAFT.md

@@ -56,7 +56,7 @@ Upload the generated artifact (the version matches `public/manifest.json`; the
 Chrome Web Store requires each upload to use a higher version than the last):
 
 ```txt
-templates/clips/chrome-extension/releases/clips-chrome-extension-0.1.18.zip
+templates/clips/chrome-extension/releases/clips-chrome-extension-0.1.19.zip
 ```
 
 ## Web App Rollout Gate

package/corpus/templates/clips/chrome-extension/public/manifest.json

@@ -2,7 +2,7 @@
   "manifest_version": 3,
   "name": "Agent-Native Clips",
   "short_name": "Clips",
-  "version": "0.1.18",
+  "version": "0.1.19",
   "description": "Start Clips recordings from Chrome, capture optional diagnostics, and preview Clips links on GitHub and Jira.",
   "minimum_chrome_version": "116",
   "action": {
@@ -31,7 +31,8 @@
     "https://clips.agent-native.com/*",
     "https://forms.agent-native.com/*",
     "http://localhost/*",
-    "http://127.0.0.1/*"
+    "http://127.0.0.1/*",
+    "<all_urls>"
   ],
   "content_scripts": [
     {
@@ -43,6 +44,12 @@
       "js": ["assets/github-preview-content.js"],
       "run_at": "document_idle",
       "all_frames": false
+    },
+    {
+      "matches": ["<all_urls>"],
+      "js": ["assets/content-script.js"],
+      "run_at": "document_idle",
+      "all_frames": false
     }
   ],
   "web_accessible_resources": [

package/corpus/templates/clips/chrome-extension/src/background.ts

@@ -6,6 +6,10 @@ import {
   shouldReconcilePersistedRecording,
   type OffscreenRecordingState,
 } from "./native-recording-state";
+import {
+  sendWithInjectionFallback,
+  shouldFollowOverlay,
+} from "./overlay-follow";
 import { captureExtensionError, initExtensionSentry } from "./sentry";
 
 initExtensionSentry("background");
@@ -276,16 +280,11 @@ let overlayBaseEpochMs = 0;
 // up, for both camera-only AND screen+camera recording (recordingShowsBubble).
 let overlayShowsBubble = false;
 let recordingShowsBubble = false;
-// Cross-tab follow: when true, the overlay is pushed to whatever tab the user
-// switches to mid-recording. That requires "<all_urls>" + a declarative content
-// script, which triggers Chrome's broad-host in-depth review. Shipped OFF
-// (activeTab-only: the overlay lives on the launch tab the user invoked the
-// extension from). Capture is unaffected — it runs in the offscreen document via
-// getDisplayMedia regardless. See PERMISSIONS.md to re-enable.
-// Typed `boolean` (not inferred literal `false`) so the cross-tab branches below
-// don't read as unreachable code; flip the value to re-enable. See PERMISSIONS.md.
-const CROSS_TAB_FOLLOW: boolean = false;
-// The tab the recording was launched from — the only tab activeTab lets us touch.
+// Cross-tab follow stays enabled for the next release. It needs the manifest's
+// broad-host review path (`<all_urls>` + declarative content script coverage) so
+// the face bubble and controls keep following the user across tabs and reloads.
+const CROSS_TAB_FOLLOW: boolean = true;
+// The launch tab stays the anchor for the explicit reload recovery path.
 let overlayTabId: number | null = null;
 let countdownEndsAtMs = 0;
 let armingNativeRecordingSessionId: string | null = null;
@@ -438,20 +437,19 @@ async function restoreRuntimeState(): Promise<void> {
 function sendTabMessage(
   tabId: number,
   message: Record<string, unknown>,
-): Promise<void> {
+): Promise<boolean> {
   return new Promise((resolve) => {
     try {
       chrome.tabs.sendMessage(tabId, message, () => {
-        void chrome.runtime.lastError;
-        resolve();
+        resolve(!chrome.runtime.lastError);
       });
     } catch {
-      resolve();
+      resolve(false);
     }
   });
 }
 
-async function ensureContentScript(tabId: number): Promise<void> {
+async function injectContentScript(tabId: number): Promise<boolean> {
   const scripting = (
     chrome as typeof chrome & {
       scripting?: {
@@ -462,18 +460,32 @@ async function ensureContentScript(tabId: number): Promise<void> {
       };
     }
   ).scripting;
-  if (!scripting) return;
-  await scripting
-    .executeScript({ target: { tabId }, files: ["assets/content-script.js"] })
-    .catch(() => undefined);
+  if (!scripting) return false;
+  try {
+    await scripting.executeScript({
+      target: { tabId },
+      files: ["assets/content-script.js"],
+    });
+    return true;
+  } catch {
+    // Unsupported pages (chrome://, the Chrome Web Store, and similar) reject
+    // injection. They still need to record without an in-page overlay.
+    return false;
+  }
 }
 
-async function mountOverlayOnTab(tabId: number): Promise<void> {
-  await ensureContentScript(tabId);
-  await sendTabMessage(tabId, {
-    type: "CLIPS_OVERLAY_MOUNT",
-    parts: desiredParts(),
-  });
+async function mountOverlayOnTab(
+  tabId: number,
+  parts: OverlayPart[] = desiredParts(),
+): Promise<boolean> {
+  return sendWithInjectionFallback(
+    () =>
+      sendTabMessage(tabId, {
+        type: "CLIPS_OVERLAY_MOUNT",
+        parts,
+      }),
+    () => injectContentScript(tabId),
+  );
 }
 
 function allTabs(): Promise<chrome.tabs.Tab[]> {
@@ -483,9 +495,8 @@ function allTabs(): Promise<chrome.tabs.Tab[]> {
 }
 
 async function broadcastMount(): Promise<void> {
-  // activeTab-only: keep the overlay on the launch tab (re-ensures the injected
-  // content script too, so it survives a worker suspend). Cross-tab broadcast is
-  // gated behind CROSS_TAB_FOLLOW because it needs broad host access.
+  // Cross-tab follow uses the declarative content script path, so every mounted
+  // tab can receive the current overlay parts and survive reloads.
   if (!CROSS_TAB_FOLLOW) {
     if (overlayTabId !== null) await mountOverlayOnTab(overlayTabId);
     return;
@@ -548,12 +559,9 @@ async function startPreview(wantsCamera: boolean): Promise<void> {
   if (!tab || typeof tab.id !== "number") return;
   if (previewTabId !== null && previewTabId !== tab.id) await stopPreview();
   previewTabId = tab.id;
-  await ensureContentScript(tab.id);
-  // Injection / send fail silently on unsupported pages (chrome://, etc.).
-  await sendTabMessage(tab.id, {
-    type: "CLIPS_OVERLAY_MOUNT",
-    parts: ["bubble"],
-  });
+  // Message delivery reuses the declarative script; fallback injection fails
+  // silently on unsupported pages (chrome://, the Web Store, and similar).
+  await mountOverlayOnTab(tab.id, ["bubble"]);
 }
 
 async function stopPreview(): Promise<void> {
@@ -2560,29 +2568,40 @@ async function dispatchRuntimeMessage(message: unknown): Promise<unknown> {
   }
 }
 
-// While a recording is active, keep the overlay following the user as they
-// switch tabs (programmatic injection covers tabs opened before the extension
-// loaded; declared content scripts cover navigations).
+// While a recording is active or arming, keep the overlay following the user as
+// they switch tabs. The message-first path reuses declarative scripts and only
+// injects for tabs that predate the extension or otherwise have no receiver.
 chrome.tabs.onActivated.addListener((info) => {
-  if (!CROSS_TAB_FOLLOW) return; // activeTab-only: overlay stays on the launch tab
+  if (!CROSS_TAB_FOLLOW) return;
   void (async () => {
     await ensureRestored();
-    if (overlayPhase === "idle" || !activeNativeRecording) return;
+    if (
+      !shouldFollowOverlay(
+        overlayPhase,
+        Boolean(activeNativeRecording),
+        Boolean(armingNativeRecordingSessionId),
+      )
+    )
+      return;
     await mountOverlayOnTab(info.tabId);
   })();
 });
 
-// The overlay content script is injected on demand (activeTab), so reloading the
-// launch tab wipes it — the camera bubble and toolbar would vanish for the rest
-// of the recording even though capture keeps running in the offscreen document.
-// Re-inject and re-mount once the launch tab finishes reloading while a recording
-// is active. Scoped to the launch tab (the only tab activeTab lets us touch); a
-// same-URL reload keeps that grant.
+// Keep the launch tab in sync after a reload. Declarative follow covers the
+// other tabs; this is the explicit recovery path for the tab that originally
+// started the recording.
 chrome.tabs.onUpdated.addListener((tabId, changeInfo) => {
   if (changeInfo.status !== "complete") return;
   void (async () => {
     await ensureRestored();
-    if (overlayPhase === "idle" || !activeNativeRecording) return;
+    if (
+      !shouldFollowOverlay(
+        overlayPhase,
+        Boolean(activeNativeRecording),
+        Boolean(armingNativeRecordingSessionId),
+      )
+    )
+      return;
     if (tabId !== overlayTabId) return;
     await mountOverlayOnTab(tabId);
     broadcastOverlayState();

package/corpus/templates/clips/chrome-extension/src/content-script.ts

@@ -1,12 +1,10 @@
-// Loom-style in-page overlay host. The background service worker injects this
-// into the LAUNCH TAB on demand via chrome.scripting.executeScript (covered by
-// the activeTab permission the user grants when they click the extension), NOT
-// declaratively on every page — that would need broad "<all_urls>" host access
-// and Chrome's in-depth review. So the overlay lives on the tab the recording
-// was started from; it does not follow across tabs unless CROSS_TAB_FOLLOW is
-// re-enabled in background.ts (see PERMISSIONS.md). Wrapped in an IIFE so it
-// emits a single self-contained classic script with no module imports/exports
-// and leaks no names into the shared global scope. Its only job is to
+// Loom-style in-page overlay host. The next release declares this on every page
+// so the face bubble and controls can survive tab switches and navigations; the
+// background worker can also inject it into the launch tab for immediate and
+// reload recovery. That broad "<all_urls>" host access is the Chrome Web Store
+// in-depth review boundary. Wrapped in an IIFE so it emits a single self-contained
+// classic script with no module imports/exports and leaks no names into the shared
+// global scope. Its only job is to
 // mount/unmount the overlay iframes; all UI and control logic lives inside the
 // extension-origin overlay pages (src/overlay.html). The worker is the source of
 // truth for which "parts" are visible and pushes them here.
@@ -18,6 +16,12 @@
   const ALL_PARTS: OverlayPart[] = ["bubble", "countdown", "toolbar", "saving"];
   const flags = window as unknown as { __clipsOverlayHostReady?: boolean };
 
+  // Declarative and recovery injection can target the same document. Return
+  // before any side effects so global listeners remain one-per-page.
+  if (flags.__clipsOverlayHostReady) return;
+  flags.__clipsOverlayHostReady = true;
+  let recordingActive = false;
+
   function errorPayload(error: unknown): {
     name: string;
     message: string;
@@ -40,6 +44,7 @@
     error: unknown,
     context: Record<string, unknown> = {},
   ): void {
+    if (!recordingActive) return;
     try {
       chrome.runtime.sendMessage(
         {
@@ -73,6 +78,15 @@
     });
   });
 
+  try {
+    chrome.storage.onChanged.addListener((changes, areaName) => {
+      if (areaName !== "local" || !changes.clipsRecordingActive) return;
+      recordingActive = changes.clipsRecordingActive.newValue === true;
+    });
+  } catch {
+    /* storage unavailable */
+  }
+
   // ----- Draggable, resizable camera bubble ---------------------------------
   // Size + position persist in storage so the bubble stays where the user put it
   // across pages and recordings (like the desktop app). The iframe can't move or
@@ -339,7 +353,8 @@
     try {
       chrome.storage.local.get("clipsRecordingActive", (value) => {
         if (chrome.runtime.lastError) return;
-        if (value && value.clipsRecordingActive) requestState();
+        recordingActive = value?.clipsRecordingActive === true;
+        if (recordingActive) requestState();
       });
     } catch {
       /* ignore */
@@ -593,13 +608,6 @@
     lastWantedParts = wanted;
   }
 
-  // Guard against rare double-injection (SPA soft-reloads re-running the script).
-  if (flags.__clipsOverlayHostReady) {
-    syncIfRecording();
-    return;
-  }
-  flags.__clipsOverlayHostReady = true;
-
   chrome.runtime.onMessage.addListener((message) => {
     if (!message || typeof message !== "object") return;
     const type = (message as { type?: unknown }).type;

package/corpus/templates/clips/chrome-extension/src/overlay-follow.ts

@@ -0,0 +1,25 @@
+export type OverlayFollowPhase =
+  | "idle"
+  | "countdown"
+  | "recording"
+  | "paused"
+  | "saving";
+
+export function shouldFollowOverlay(
+  phase: OverlayFollowPhase,
+  hasActiveNativeRecording: boolean,
+  hasArmingNativeRecording: boolean,
+): boolean {
+  return (
+    phase !== "idle" && (hasActiveNativeRecording || hasArmingNativeRecording)
+  );
+}
+
+export async function sendWithInjectionFallback(
+  sendMessage: () => Promise<boolean>,
+  injectContentScript: () => Promise<boolean>,
+): Promise<boolean> {
+  if (await sendMessage()) return true;
+  if (!(await injectContentScript())) return false;
+  return sendMessage();
+}

package/dist/client/chat/run-recovery.js

@@ -213,10 +213,15 @@ export function RunErrorRecoveryCard({ info, onContinue, onRetry, onFork, onDism
     const shouldShowMissingProviderSetup = isMissingLlmProviderRunError(info);
     const isProviderAuthError = isProviderAuthenticationError([info.message, info.details].filter(Boolean).join("\n"), info.errorCode);
     // Blocked on something the reader goes and fixes elsewhere, then comes back
-    // to. Without a retry the card is a dead end and its own copy ("then retry")
-    // points at a button that isn't there.
+    // to. Recoverable runs and email verification keep a retry path; rejected
+    // provider credentials use the setup flow below so the same bad key is not
+    // replayed.
     const isUnblockableExternally = info.errorCode === "email_verification_required";
-    const canRetry = canRecover || isProviderAuthError || isUnblockableExternally;
+    // Rejected provider keys already have a recovery path: update/connect the
+    // credential, then let the setup callback re-run the turn. Exposing a
+    // separate retry button here just replays the same rejected credential and
+    // turns a permanent auth failure into a loop.
+    const canRetry = canRecover || isUnblockableExternally;
     const builderReconnectResolved = shouldShowBuilderReconnect &&
         builderReconnect.hasFetchedStatus &&
         builderReconnect.configured;

package/dist/collab/awareness.d.ts

@@ -62,11 +62,11 @@ export declare const postAwareness: import("h3").EventHandlerWithFetch<import("h
     error: string;
     states?: undefined;
 } | {
-    error?: undefined;
     states: {
         clientId: number;
         state: string;
     }[];
+    error?: undefined;
 }>>;
 /**
  * GET /_agent-native/collab/:docId/users
@@ -77,9 +77,9 @@ export declare const getActiveUsers: import("h3").EventHandlerWithFetch<import("
     error: string;
     users?: undefined;
 } | {
-    error?: undefined;
     users: {
         clientId: number;
         lastSeen: number;
     }[];
+    error?: undefined;
 }>>;

package/dist/collab/routes.d.ts

@@ -26,8 +26,8 @@ export declare const getCollabState: import("h3").EventHandlerWithFetch<import("
  * Body: { update: string (base64), requestSource?: string }
  */
 export declare const postCollabUpdate: import("h3").EventHandlerWithFetch<import("h3").EventHandlerRequest, Promise<{
-    ok?: undefined;
     error: string;
+    ok?: undefined;
 } | {
     error?: undefined;
     ok: boolean;

package/dist/collab/struct-routes.d.ts

@@ -13,8 +13,8 @@
  * Body: { json: any, fieldName?: string, type?: "map"|"array", requestSource?: string }
  */
 export declare const postCollabJson: import("h3").EventHandlerWithFetch<import("h3").EventHandlerRequest, Promise<{
-    ok?: undefined;
     error: string;
+    ok?: undefined;
 } | {
     error?: undefined;
     ok: boolean;

package/dist/notifications/routes.d.ts

@@ -11,14 +11,14 @@
  *   DELETE /_agent-native/notifications/:id         — delete
  */
 export declare function createNotificationsHandler(): import("h3").EventHandlerWithFetch<import("h3").EventHandlerRequest, Promise<"" | import("./types.js").Notification[] | {
-    error?: undefined;
     count: number;
     updated?: undefined;
     ok?: undefined;
-} | {
     error?: undefined;
+} | {
     updated: number;
     ok?: undefined;
+    error?: undefined;
     count?: undefined;
 } | {
     updated?: undefined;
@@ -26,8 +26,8 @@ export declare function createNotificationsHandler(): import("h3").EventHandlerW
     ok?: undefined;
     count?: undefined;
 } | {
-    error?: undefined;
     updated?: undefined;
     ok: boolean;
+    error?: undefined;
     count?: undefined;
 }>>;

package/dist/provider-api/actions/provider-api.js

@@ -193,6 +193,13 @@ export function createProviderApiDocsSchema(providerSchema = z.string().min(1))
         search: WebContentSearchSchema.optional().describe("Optional post-fetch search over extracted content by default. Supports query, queries, terms, regex, source, maxMatches, and contextChars."),
     });
 }
+const STAGED_DATASET_CAP_ERROR = /Staged dataset (?:byte )?cap exceeded/i;
+function rethrowStagingError(error) {
+    if (error instanceof Error && STAGED_DATASET_CAP_ERROR.test(error.message)) {
+        throw new Error(`${error.message} Recover by deleting older staged datasets with list-staged-datasets/delete-staged-dataset, or switch this request to saveToFile / a smaller staged result before trying again.`, { cause: error });
+    }
+    throw error;
+}
 export function createProviderApiRequestAction(runtime, options = {}) {
     const schema = options.schema ?? createProviderApiRequestSchema();
     return defineAction({
@@ -244,10 +251,15 @@ export function createProviderApiRequestAction(runtime, options = {}) {
                 if (!ownerEmail) {
                     throw new Error("No authenticated context for provider API staging.");
                 }
-                return stagingExecuteRequest(args, runtime.executeRequest, {
-                    appId: options.appId,
-                    ownerEmail,
-                });
+                try {
+                    return await stagingExecuteRequest(args, runtime.executeRequest, {
+                        appId: options.appId,
+                        ownerEmail,
+                    });
+                }
+                catch (error) {
+                    rethrowStagingError(error);
+                }
             }
             return runtime.executeRequest(args);
         },

package/dist/resources/handlers.d.ts

@@ -51,8 +51,8 @@ export declare function handleDeleteResource(event: any): Promise<{
     error: string;
     ok?: undefined;
 } | {
-    error?: undefined;
     ok: boolean;
+    error?: undefined;
 }>;
 /** POST /_agent-native/resources/upload — upload a file as a resource */
 export declare function handleUploadResource(event: any): Promise<import("./store.js").Resource | {
@@ -73,10 +73,10 @@ export declare function handleUploadResource(event: any): Promise<import("./stor
     runId: string | null;
     expiresAt: number | null;
     metadata: string | null;
-    error?: undefined;
     url: string;
     provider: string;
     storageSetupRequired?: undefined;
+    error?: undefined;
 } | {
     error: string;
     storageSetupRequired: boolean;

package/dist/secrets/routes.d.ts

@@ -34,16 +34,16 @@ export declare function createListSecretsHandler(): import("h3").EventHandlerWit
 /** POST /_agent-native/secrets/:key — write a secret. */
 export declare function createWriteSecretHandler(): import("h3").EventHandlerWithFetch<import("h3").EventHandlerRequest, Promise<{
     error: string;
-    ok?: undefined;
     status?: undefined;
+    ok?: undefined;
 } | {
     ok: boolean;
     status: string;
     error?: undefined;
 } | {
-    ok?: undefined;
     error: string;
     removed?: undefined;
+    ok?: undefined;
 } | {
     ok: boolean;
     removed: boolean;
@@ -54,9 +54,9 @@ export declare function createWriteSecretHandler(): import("h3").EventHandlerWit
  * or the current stored value without changing anything.
  */
 export declare function createTestSecretHandler(): import("h3").EventHandlerWithFetch<import("h3").EventHandlerRequest, Promise<{
-    ok?: undefined;
     error: string;
     note?: undefined;
+    ok?: undefined;
 } | {
     ok: boolean;
     note?: undefined;

package/dist/server/realtime-token.d.ts

@@ -26,8 +26,8 @@ export declare function createRealtimeTokenHandler(): import("h3").EventHandlerW
     expiresAt?: undefined;
     ttlSeconds?: undefined;
 } | {
+    error?: undefined;
     token: string;
     expiresAt: string;
     ttlSeconds: number;
-    error?: undefined;
 }>>;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agent-native/core",
-  "version": "0.0.0-beta-20260820155159",
+  "version": "0.0.0-beta-20260820171000",
   "description": "Framework for agent-native application development — where AI agents and UI share SQL state, actions, and context",
   "homepage": "https://github.com/BuilderIO/agent-native#readme",
   "bugs": {