### Fixes

- Agents/TTS: preserve legacy `[[audio_as_voice]]` hints on trusted tool-result

`MEDIA:` payloads so generated audio still delivers as a voice note. (#46535)

Thanks @azade-c.

- Telegram/STT: frame inbound voice-note transcripts as machine-generated,

untrusted text in agent context while preserving raw transcript mention

detection. Closes #33360. Thanks @smartchainark.

- `[embed ...]` for Control UI rich rendering

These directives are separate. `MEDIA:` and reply/voice tags remain delivery metadata; `[embed ...]` is the web-only rich render path.

Trusted tool-result media uses the same `MEDIA:` / `[[audio_as_voice]]` parser before delivery, so legacy tool outputs can still mark an audio attachment as a voice note.

When block streaming is enabled, `MEDIA:` remains single-delivery metadata for a

turn. If the same media URL is sent in a streamed block and repeated in the final

expect(ctx.state.pendingToolMediaUrls).toEqual(["/tmp/screenshot.png"]);

});

it("preserves legacy audio_as_voice when queuing trusted MEDIA tool output", async () => {

const onToolResult = vi.fn();

const ctx = createMockContext({

shouldEmitToolOutput: false,

onToolResult,

builtinToolNames: new Set(["tts"]),

});

await handleToolExecutionEnd(ctx, {

type: "tool_execution_end",

toolName: "tts",

toolCallId: "tc-1",

isError: false,

result: {

content: [

{

type: "text",

text: "Generated audio reply.\n[[audio_as_voice]]\nMEDIA:/tmp/reply.opus",

},

],

},

});

expect(onToolResult).not.toHaveBeenCalled();

expect(ctx.state.pendingToolMediaUrls).toEqual(["/tmp/reply.opus"]);

expect(ctx.state.pendingToolAudioAsVoice).toBe(true);

});

it("does NOT emit local media for untrusted tools", async () => {

const onToolResult = vi.fn();

const ctx = createMockContext({ shouldEmitToolOutput: false, onToolResult });

});

});

it("extracts audioAsVoice from legacy MEDIA text", () => {

expect(

extractToolResultMediaArtifact({

content: [

{ type: "text", text: "Generated audio\n[[audio_as_voice]]\nMEDIA:/tmp/reply.opus" },

],

}),

).toEqual({

mediaUrls: ["/tmp/reply.opus"],

audioAsVoice: true,

});

});

it("keeps legacy audioAsVoice when the tag and MEDIA path are in separate text blocks", () => {

expect(

extractToolResultMediaArtifact({

content: [

{ type: "text", text: "[[audio_as_voice]]" },

{ type: "text", text: "MEDIA:/tmp/reply.opus" },

],

}),

).toEqual({

mediaUrls: ["/tmp/reply.opus"],

audioAsVoice: true,

});

});

it("extracts structured media trust markers", () => {

expect(

extractToolResultMediaArtifact({

// parser so directive matching and validation stay in sync with outbound

// reply parsing.

const paths: string[] = [];

let audioAsVoice = false;

let hasImageContent = false;

for (const item of content) {

if (!item || typeof item !== "object") {

}

if (entry.type === "text" && typeof entry.text === "string") {

const parsed = splitMediaFromOutput(entry.text);

if (parsed.audioAsVoice) {

audioAsVoice = true;

}

if (parsed.mediaUrls?.length) {

paths.push(...parsed.mediaUrls);

}

}

}

if (paths.length > 0) {

return { mediaUrls: paths };

return {

mediaUrls: paths,

...(audioAsVoice ? { audioAsVoice: true } : {}),

};

}

// Fall back to legacy details.path when image content exists but no