@agent-native/core 0.109.1 → 0.109.2

package/corpus/core/src/deploy/build.ts

@@ -138,6 +138,14 @@ export const CLOUDFLARE_WORKER_STUB_MODULES: Record<string, string> = {
     "export default { Resvg };",
     "",
   ].join("\n"),
+  playwright: [
+    "const unavailable = async () => { throw new Error('playwright unavailable in Cloudflare Pages worker'); };",
+    "export const chromium = { launch: unavailable, connect: unavailable, connectOverCDP: unavailable };",
+    "export const firefox = { launch: unavailable, connect: unavailable };",
+    "export const webkit = { launch: unavailable, connect: unavailable };",
+    "export default { chromium, firefox, webkit };",
+    "",
+  ].join("\n"),
   "playwright-core": [
     "const unavailable = async () => { throw new Error('playwright-core unavailable in Cloudflare Pages worker'); };",
     "export const chromium = { launch: unavailable };",
@@ -187,6 +195,50 @@ export const CLOUDFLARE_WORKER_STUB_MODULES: Record<string, string> = {
     "export default async () => ({ elements: [], files: {} });\n",
 };
 
+export const CLOUDFLARE_WORKER_STUB_SUBPATH_MODULES: Record<string, string> = {
+  "pdf-parse/worker": [
+    "const unavailable = async () => { throw new Error('pdf-parse/worker unavailable in Cloudflare Pages worker'); };",
+    "export class CanvasFactory {}",
+    "export const getData = unavailable;",
+    "export default { CanvasFactory, getData };",
+    "",
+  ].join("\n"),
+};
+
+export function cloudflareWorkerStubAliasArgs(stubDir: string): string[] {
+  const subpathAliases = Object.keys(CLOUDFLARE_WORKER_STUB_SUBPATH_MODULES)
+    .sort((a, b) => b.length - a.length)
+    .map(
+      (mod) =>
+        `--alias:${mod}=${path.join(stubDir, `${mod.replace(/\//g, "__")}.js`)}`,
+    );
+  const packageAliases = Object.keys(CLOUDFLARE_WORKER_STUB_MODULES)
+    .sort((a, b) => b.length - a.length)
+    .map((mod) => `--alias:${mod}=${path.join(stubDir, mod, "index.js")}`);
+  return [...subpathAliases, ...packageAliases];
+}
+
+export function assertNoCloudflareWorkerStubDynamicImports(
+  code: string,
+  sourceName: string,
+): void {
+  const stubbedModules = [
+    ...Object.keys(CLOUDFLARE_WORKER_STUB_SUBPATH_MODULES),
+    ...Object.keys(CLOUDFLARE_WORKER_STUB_MODULES),
+  ];
+  const pattern = stubbedModules
+    .sort((a, b) => b.length - a.length)
+    .map((moduleName) => moduleName.replace(/[.*+?^${}()|[\]\\]/g, "\\$&"))
+    .join("|");
+  const unresolvedImport = code.match(
+    new RegExp(`\\bimport\\s*\\(\\s*(["'])(${pattern})\\1\\s*\\)`),
+  );
+  if (!unresolvedImport) return;
+  throw new Error(
+    `Cloudflare worker output ${sourceName} retained a dynamic import for stubbed module "${unresolvedImport[2]}". Use a literal import so the worker bundler can apply its fail-closed stub.`,
+  );
+}
+
 function cloudflareNodeBuiltinStubSource(
   moduleName: string,
   namedExports: string[],
@@ -1666,9 +1718,15 @@ async function buildCloudflarePages() {
       JSON.stringify({ name: mod, main: "index.js", type: "module" }),
     );
   }
-  const stubAliases = Object.keys(CLOUDFLARE_WORKER_STUB_MODULES).map(
-    (mod) => `--alias:${mod}=${path.join(stubDir, mod, "index.js")}`,
-  );
+  for (const [mod, source] of Object.entries(
+    CLOUDFLARE_WORKER_STUB_SUBPATH_MODULES,
+  )) {
+    fs.writeFileSync(
+      path.join(stubDir, `${mod.replace(/\//g, "__")}.js`),
+      source,
+    );
+  }
+  const stubAliases = cloudflareWorkerStubAliasArgs(stubDir);
   const nodeBuiltinStubDir = path.join(tmpDir, "node-builtin-stubs");
   fs.mkdirSync(nodeBuiltinStubDir, { recursive: true });
   const nodeBuiltinStubAliases: string[] = [];
@@ -1850,6 +1908,8 @@ async function buildCloudflarePages() {
       );
     }
 
+    assertNoCloudflareWorkerStubDynamicImports(code, jsFile);
+
     fs.writeFileSync(jsFile, code);
   }
 

package/corpus/templates/clips/.agents/skills/dictate/SKILL.md

@@ -1,9 +1,9 @@
 ---
 name: dictate
 description: >-
-  The Dictate tab in Clips — press-and-hold dictation history (Hold-Fn / Cmd+Shift+Space),
-  AI cleanup, and the desktop-app integration that captures the audio. Use when listing
-  past dictations, polishing dictation text, or wiring the desktop tray hand-off.
+  The Dictate tab in Clips — press-and-hold desktop dictation, mobile voice
+  dictation, history, AI cleanup, and native capture hand-offs. Use when
+  listing past dictations, polishing dictation text, or wiring native capture.
 ---
 
 # Dictate
@@ -25,7 +25,9 @@ The press-and-hold UX intentionally mirrors **Wispr Flow**. See `templates/clips
 
 ## Data model touched
 
-- **`dictations`** — `id`, `fullText`, `cleanedText`, `durationMs`, `audioUrl`, `source` (`hold-fn` | `hotkey` | `manual`), `createdAt`, `ownableColumns()`.
+- **`dictations`** — `id`, `fullText`, `cleanedText`, `durationMs`, `audioUrl`,
+  `source` (desktop shortcuts, `manual`, or `mobile`), `createdAt`,
+  `ownableColumns()`.
 
 `fullText` is the raw native transcript captured on-device (macOS Speech / Web Speech). `cleanedText` is the optional AI-polished version produced by the shared `cleanup-transcript` pipeline. UI shows `cleanedText ?? fullText`.
 
@@ -81,7 +83,28 @@ The press-and-hold flow is **owned by the desktop app** (`src-tauri/`). On Hold-
 3. The dictation row is created via the framework HTTP layer — the agent does not start/stop dictations.
 4. AI cleanup is a background pass via `cleanup-dictation`, not in the hot path — never block paste-on-release on a network round-trip.
 
-Agents must **never** wire dictation start/stop server-side. The native key listener is the only entry point.
+Agents must **never** wire dictation start/stop server-side. Desktop key
+listeners and the mobile capture UI own those user gestures.
+
+## Mobile dictation
+
+The Agent Native iOS/Android app exposes Dictate from native Home, deep links,
+and OS quick actions. Mobile is click-to-toggle rather than hold-to-talk:
+
+1. `expo-audio` records mic-only M4A and persists it under the app documents
+   directory before any network request.
+2. A durable capture-queue row protects the audio across interruptions, app
+   restarts, and upload failures.
+3. The named mobile voice client posts multipart audio to the authenticated
+   `/_agent-native/transcribe-voice` route. Provider keys never enter the app.
+4. The cleaned transcript is editable, copied to the OS clipboard, and saved
+   with `create-dictation --source=mobile`; edits use `update-dictation`.
+5. The audio also syncs to Clips through the resumable recording upload path so
+   a failed transcription never destroys the captured speech.
+
+Do not auto-send mobile dictation to an agent or another app. iOS/Android quick
+actions open capture; clipboard is the cross-app fallback until platform
+keyboard/accessibility insertion is enabled.
 
 ## UI conventions (don't break)
 

package/corpus/templates/clips/.agents/skills/meetings/SKILL.md

@@ -47,6 +47,13 @@ Meeting capture records **two streams** and tags transcript segments by source:
 
 Each transcript segment carries a `source: "mic" | "system"` tag, which we use to attribute action items to the right attendee. This is why **per-attendee action items only work reliably with mic + system capture** — mic-only recordings make remote attendees silent. Document this caveat whenever you surface action items.
 
+The iOS/Android companion is intentionally different: mobile background meeting
+capture is microphone-only because the operating systems do not expose another
+phone app's Zoom/Meet/Teams audio. It is appropriate for in-person rooms and for
+capturing audio played from a separate device. Mobile recordings still receive
+Clips transcription and summaries, but do not promise desktop-quality remote
+speaker attribution or per-attendee action items.
+
 ## Bidirectional recording ↔ meeting link
 
 A meeting can have an associated recording, and a recording can be linked back to a meeting:

package/corpus/templates/clips/.agents/skills/recording/SKILL.md

@@ -38,6 +38,27 @@ Some recordings are linked to a meeting — when `meeting_id` is non-null on the
    `processing` to `ready`. Do not wait for the upload/finalize response before
    opening the page.
 
+## Mobile companion lifecycle
+
+The Agent Native mobile app uses the same recording rows and binary upload
+routes with native capture primitives:
+
+1. Camera video/import uses `expo-camera` / the system photo picker; meeting
+   audio uses `expo-audio` with background recording enabled.
+2. The native file is copied into the documents directory and a typed
+   AsyncStorage capture job is written before upload starts.
+3. `create-recording` receives a stable client-generated id,
+   `sourceAppName: "Agent Native Mobile"`, and the container MIME type.
+4. Upload reads at most 3 MiB through an Expo FileHandle and persists the next
+   chunk index after every acknowledged POST. The 4 MiB server cap still
+   applies.
+5. Foreground/resume reconciliation polls `/api/uploads/:id/status`; retryable
+   failures back off and never discard the local file. Completion can post a
+   local notification and the recording appears in Clips everywhere.
+
+Mobile audio M4A is an MP4 container and uses the recording route's MP4 MIME
+alias. The bytes are not converted or loaded whole into JavaScript memory.
+
 ## Seekable playback (don't ship raw MediaRecorder output)
 
 Raw `MediaRecorder` files are not friendly to progressive HTTP playback, which
@@ -191,6 +212,8 @@ export function useRecorder() {
 - **Never** start a `MediaRecorder` without a user gesture (or a user-initiated `record-intent`).
 - **Never** re-prompt for permissions on pause/resume — reuse the stream.
 - **Never** fire the upload from the main thread if the chunks are large — prefer a web worker for anything longer than ~60s.
+- **Never** read a complete mobile recording into JavaScript memory. Use bounded
+  FileHandle reads and checkpoint every acknowledged chunk.
 - The `recordings` row must exist **before** the first chunk is sent.
 - On every lifecycle change, write `navigation` → `{ view: "record" }` → `{ view: "recording", recordingId }` so the agent can see what's happening.
 - All AI generated during/after recording goes through the agent chat — see `ai-video-tools`.

package/corpus/templates/clips/AGENTS.md

@@ -51,6 +51,16 @@ Detailed media, meeting, dictation, editing, and sharing rules live in
   `previewTruncated` is true, it may end mid-sentence and does not show where
   transcription ended. Call `get-recording-player-data` before judging
   completeness or quoting the full transcript.
+- Mobile captures originate in the Agent Native iOS/Android app. The phone
+  persists each audio/video file before network work, creates the recording
+  with `sourceAppName: "Agent Native Mobile"`, and resumes bounded chunk
+  uploads from its durable queue. Do not ask users to keep a capture screen
+  open or re-record after a transient upload failure; reconnect Clips and retry
+  the saved job from mobile Home.
+- Mobile meeting capture is microphone-only. It can record an in-person room or
+  audio playing from another device, but iOS and Android do not expose another
+  phone app's Zoom/Meet/Teams call audio to this flow. Do not claim mobile
+  attendee attribution has the desktop mic-plus-system-audio fidelity.
 - Dictation cleanup, Clip title/cleanup, and meeting summaries should pass
   bounded `voiceContext` to the shared cleanup/transcription path when active
   app context, learned vocabulary, user notes, or AGENTS.md preferences are

package/corpus/templates/clips/actions/create-dictation.ts

@@ -1,9 +1,10 @@
 /**
- * Create a dictation row from browser or desktop voice capture.
+ * Create a dictation row from browser, desktop, or mobile voice capture.
  */
 
 import { defineAction } from "@agent-native/core";
 import { writeAppState } from "@agent-native/core/application-state";
+import { and, eq } from "drizzle-orm";
 import { z } from "zod";
 
 import { getDb, schema } from "../server/db/index.js";
@@ -11,18 +12,28 @@ import {
   getActiveOrganizationId,
   getCurrentOwnerEmail,
   nanoid,
+  ownerEmailMatches,
 } from "../server/lib/recordings.js";
 
 export default defineAction({
   description:
-    "Create a dictation history row from captured voice text. Used by browser Dictate and desktop voice capture.",
+    "Create a dictation history row from captured voice text. Used by browser, desktop, and mobile voice capture.",
   schema: z.object({
+    id: z.string().min(1).optional().describe("Stable client-generated id"),
     fullText: z.string().min(1).describe("Raw dictation text"),
     cleanedText: z.string().nullable().optional(),
     durationMs: z.coerce.number().int().min(0).optional(),
     audioUrl: z.string().url().nullable().optional(),
     source: z
-      .enum(["fn-hold", "cmd-shift-space", "manual", "other", "fn", "custom"])
+      .enum([
+        "fn-hold",
+        "cmd-shift-space",
+        "manual",
+        "mobile",
+        "other",
+        "fn",
+        "custom",
+      ])
       .default("manual"),
     targetApp: z.string().nullable().optional(),
     startedAt: z.string().datetime().optional(),
@@ -32,10 +43,24 @@ export default defineAction({
     const ownerEmail = getCurrentOwnerEmail();
     const orgId = await getActiveOrganizationId().catch(() => null);
     const now = new Date().toISOString();
-    const id = nanoid();
+    const id = args.id ?? nanoid();
     const fullText = args.fullText.trim();
     if (!fullText) throw new Error("Dictation text is required");
 
+    if (args.id) {
+      const [existing] = await db
+        .select()
+        .from(schema.dictations)
+        .where(
+          and(
+            eq(schema.dictations.id, id),
+            ownerEmailMatches(schema.dictations.ownerEmail, ownerEmail),
+          ),
+        )
+        .limit(1);
+      if (existing) return existing;
+    }
+
     await db.insert(schema.dictations).values({
       id,
       fullText,

package/corpus/templates/clips/actions/finalize-recording.ts

@@ -204,9 +204,7 @@ async function verifyServedMediaUrl(videoUrl: string): Promise<number | null> {
       const statusOk = response.status === 200 || response.status === 206;
       if (statusOk) {
         const servedBytes = servedMediaSizeBytes(response);
-        if (servedBytes === null) {
-          lastFailure = "Stored media byte count could not be verified";
-        } else if (await responseHasReadableMediaBytes(response)) {
+        if (await responseHasReadableMediaBytes(response)) {
           // Builder stable video URLs intentionally replace the source object
           // with a smaller compressed generation at the same URL. A positive,
           // readable object is the invariant here; source-byte equality is
@@ -1670,7 +1668,7 @@ export default defineAction({
         servedBytes = await verifyServedMediaUrl(upload.url);
       } catch (err) {
         const failureReason = err instanceof Error ? err.message : String(err);
-        return leaveRecordingProcessingForMediaVerification({
+        return await leaveRecordingProcessingForMediaVerification({
           id,
           ownerEmail,
           failureReason,

package/corpus/templates/clips/actions/get-recording-player-data.ts

@@ -29,6 +29,7 @@ import { and, asc, eq, or, sql } from "drizzle-orm";
 import { z } from "zod";
 
 import { getDb, schema } from "../server/db/index.js";
+import { isMediaVerificationPending } from "../server/lib/media-verification-state.js";
 import { resolvePlayerVideoUrl } from "../server/lib/player-video-url.js";
 import {
   canOpenDirectRecordingPage,
@@ -178,12 +179,20 @@ export default defineAction({
       access.role === "owner" ||
       access.role === "admin" ||
       access.role === "editor";
-    const [cleanupStateRaw, builderCreditsRaw] = await Promise.all([
-      readAppState(`transcript-cleanup-${args.recordingId}`).catch(() => null),
-      canEditRecording
-        ? readAppState(CLIPS_BUILDER_CREDITS_STATE_KEY).catch(() => null)
-        : Promise.resolve(null),
-    ]);
+    const [cleanupStateRaw, builderCreditsRaw, verificationPending] =
+      await Promise.all([
+        readAppState(`transcript-cleanup-${args.recordingId}`).catch(
+          () => null,
+        ),
+        canEditRecording
+          ? readAppState(CLIPS_BUILDER_CREDITS_STATE_KEY).catch(() => null)
+          : Promise.resolve(null),
+        isMediaVerificationPending({
+          ownerEmail: rec.ownerEmail,
+          recordingId: args.recordingId,
+          recordingStatus: rec.status,
+        }),
+      ]);
     const cleanupState =
       cleanupStateRaw && typeof cleanupStateRaw === "object"
         ? (cleanupStateRaw as Record<string, unknown>)
@@ -326,6 +335,7 @@ export default defineAction({
         hasAudio: Boolean(rec.hasAudio),
         hasCamera: Boolean(rec.hasCamera),
         status: rec.status,
+        verificationPending,
         uploadProgress: rec.uploadProgress,
         failureReason: rec.failureReason,
         // Don't leak the password to clients (especially to MCP hosts that

package/corpus/templates/clips/actions/set-mobile-capture-state.ts

@@ -0,0 +1,37 @@
+import { defineAction } from "@agent-native/core/action";
+import { writeAppState } from "@agent-native/core/application-state";
+import { z } from "zod";
+
+export const MOBILE_CAPTURE_STATE_KEY = "mobile-capture-state";
+
+const mobileCaptureStateSchema = z
+  .object({
+    view: z.enum(["home", "dictate", "meeting", "video"]),
+    phase: z.enum([
+      "idle",
+      "ready",
+      "recording",
+      "paused",
+      "saving",
+      "processing",
+      "review",
+      "error",
+    ]),
+    captureId: z.string().trim().min(1).max(200).optional(),
+  })
+  .strict();
+
+export default defineAction({
+  description:
+    "Record the privacy-safe view and phase of the Agent Native mobile capture companion.",
+  agentTool: false,
+  schema: mobileCaptureStateSchema,
+  run: async (args) => {
+    const state = {
+      ...args,
+      updatedAt: new Date().toISOString(),
+    };
+    await writeAppState(MOBILE_CAPTURE_STATE_KEY, state);
+    return state;
+  },
+});

package/corpus/templates/clips/actions/update-dictation.ts

@@ -20,7 +20,15 @@ export default defineAction({
     fullText: z.string().optional(),
     cleanedText: z.string().nullable().optional(),
     source: z
-      .enum(["fn-hold", "cmd-shift-space", "manual", "other", "fn", "custom"])
+      .enum([
+        "fn-hold",
+        "cmd-shift-space",
+        "manual",
+        "mobile",
+        "other",
+        "fn",
+        "custom",
+      ])
       .optional(),
     targetApp: z.string().nullable().optional(),
   }),

package/corpus/templates/clips/app/routes/r.$recordingId.tsx

@@ -376,6 +376,7 @@ export default function RecordingPage() {
   }, [recordingId, playerDataForbidden, navigate]);
 
   const recording = playerDataQ.data?.recording;
+  const verificationPending = recording?.verificationPending === true;
   const role = playerDataQ.data?.role as
     | "owner"
     | "admin"
@@ -781,13 +782,22 @@ export default function RecordingPage() {
       setProcessingTimeout(false);
       return;
     }
+    if (verificationPending) {
+      setProcessingTimeout(false);
+      return;
+    }
     const timeoutMs =
       recording.status === "processing"
         ? PROCESSING_STUCK_TIMEOUT_MS
         : UPLOAD_STUCK_TIMEOUT_MS;
     const handle = setTimeout(() => setProcessingTimeout(true), timeoutMs);
     return () => clearTimeout(handle);
-  }, [recording?.status, recording?.videoUrl, recordingId]);
+  }, [
+    recording?.status,
+    recording?.videoUrl,
+    recordingId,
+    verificationPending,
+  ]);
 
   usePlayerShortcuts({ playerRef, chapters });
 
@@ -873,7 +883,8 @@ export default function RecordingPage() {
       isNativeSaveFailureReason(rawFailureReason);
     // Give a long-running desktop save an actionable recovery state without
     // claiming the upload failed while its bounded final request is still live.
-    const stuckFailure = !explicitFailure && processingTimeout;
+    const stuckFailure =
+      !explicitFailure && !verificationPending && processingTimeout;
     const isFailure = explicitFailure || waitingForStorage || nativeSaveFailed;
     const showRecoveryState = isFailure || stuckFailure;
     const displayReason = explicitFailure

package/corpus/templates/clips/app/routes/share.$shareId.tsx

@@ -478,6 +478,7 @@ export default function ShareRoute() {
   });
 
   const recording = dataQ.data?.data?.recording;
+  const verificationPending = recording?.verificationPending === true;
   const comments = dataQ.data?.data?.comments ?? [];
   const reactions = dataQ.data?.data?.reactions ?? [];
   const chapters = dataQ.data?.data?.chapters ?? [];
@@ -560,6 +561,10 @@ export default function ShareRoute() {
       setProcessingTimeout(false);
       return;
     }
+    if (verificationPending) {
+      setProcessingTimeout(false);
+      return;
+    }
 
     const timeoutMs =
       recording.status === "processing"
@@ -567,7 +572,12 @@ export default function ShareRoute() {
         : UPLOAD_STUCK_TIMEOUT_MS;
     const handle = setTimeout(() => setProcessingTimeout(true), timeoutMs);
     return () => clearTimeout(handle);
-  }, [recording?.id, recording?.status, recording?.videoUrl]);
+  }, [
+    recording?.id,
+    recording?.status,
+    recording?.videoUrl,
+    verificationPending,
+  ]);
 
   usePlayerShortcuts({ playerRef });
 
@@ -712,7 +722,8 @@ export default function ShareRoute() {
     const storageSetupFailure = isStorageSetupFailureReason(rawFailureReason);
     const loomStorageSetupFailure =
       storageSetupFailure && isLoomRecordingSource(recording);
-    const stuckFailure = !explicitFailure && processingTimeout;
+    const stuckFailure =
+      !explicitFailure && !verificationPending && processingTimeout;
     const isFailure = explicitFailure || storageSetupFailure || stuckFailure;
     const canManageStorage = viewerCanEdit;
     const signInHref = buildSignInHref(`/r/${recording.id}`);

package/corpus/templates/clips/changelog/2026-07-17-capture-dictation-background-meeting-audio-and-camera-videos.md

@@ -0,0 +1,6 @@
+---
+type: added
+date: 2026-07-17
+---
+
+Capture dictation, background meeting audio, and camera videos from the Agent Native iOS and Android app with automatic recovery and resumable Clips sync.

package/corpus/templates/clips/changelog/2026-07-17-delayed-upload-verification-keeps-local-backups-retryable.md

@@ -0,0 +1,6 @@
+---
+type: fixed
+date: 2026-07-17
+---
+
+Delayed upload verification now keeps local backups retryable and avoids showing a false processing failure.

package/corpus/templates/clips/desktop/src/app.tsx

@@ -1835,6 +1835,23 @@ export function App() {
     );
   }, []);
 
+  useEffect(() => {
+    let unlisten: (() => void) | null = null;
+    let cancelled = false;
+    listen("clips:pending-uploads-changed", () => {
+      void loadPendingUploads();
+    })
+      .then((stop) => {
+        if (cancelled) stop();
+        else unlisten = stop;
+      })
+      .catch(() => {});
+    return () => {
+      cancelled = true;
+      unlisten?.();
+    };
+  }, [loadPendingUploads]);
+
   useEffect(() => {
     if (popoverView === "meetings" && popoverVisible) {
       void fetchUpcomingMeetings();

package/corpus/templates/clips/desktop/src/lib/processing-backup-recovery.ts

@@ -0,0 +1,19 @@
+export async function reconcileProcessingBackup(args: {
+  waitForReady: () => Promise<{ status?: string } | null>;
+  onReady: () => Promise<unknown>;
+  onUnresolved: () => Promise<unknown>;
+  onPollError?: (error: unknown) => void;
+}): Promise<"ready" | "unresolved"> {
+  try {
+    const recovered = await args.waitForReady();
+    if (recovered?.status === "ready") {
+      await args.onReady();
+      return "ready";
+    }
+  } catch (error) {
+    args.onPollError?.(error);
+  }
+
+  await args.onUnresolved();
+  return "unresolved";
+}