From d6648e77bc6b4ddd731c811500d3dfa227a1bb67 Mon Sep 17 00:00:00 2001
From: Eulices Lopez <eulices@users.noreply.github.com>
Date: Wed, 18 Mar 2026 01:52:58 -0400
Subject: [PATCH] fix: preserve synthetic transcript attachment order

---
 src/media-understanding/apply.ts       |  1 +
 src/media-understanding/format.test.ts | 25 +++++++++++++++++++++++++
 2 files changed, 26 insertions(+)

diff --git a/src/media-understanding/apply.ts b/src/media-understanding/apply.ts
index 1f97eaba5d7..d18bdbb282b 100644
--- a/src/media-understanding/apply.ts
+++ b/src/media-understanding/apply.ts
@@ -532,6 +532,7 @@ export async function applyMediaUnderstanding(params: {
       (output) => !audioOutputAttachmentIndexes.has(output.attachmentIndex),
     );
     outputs.push(...syntheticSkippedAudioOutputs);
+    outputs.sort((left, right) => left.attachmentIndex - right.attachmentIndex);
 
     if (decisions.length > 0) {
       ctx.MediaUnderstandingDecisions = [...(ctx.MediaUnderstandingDecisions ?? []), ...decisions];
diff --git a/src/media-understanding/format.test.ts b/src/media-understanding/format.test.ts
index 172ecadf985..57ea5ebf83b 100644
--- a/src/media-understanding/format.test.ts
+++ b/src/media-understanding/format.test.ts
@@ -88,4 +88,29 @@ describe("formatMediaUnderstandingBody", () => {
     });
     expect(body).toBe("[Image]\nDescription:\na cat");
   });
+
+  it("labels audio transcripts by their attachment order", () => {
+    const body = formatMediaUnderstandingBody({
+      outputs: [
+        {
+          kind: "audio.transcription",
+          attachmentIndex: 0,
+          text: "first clip was silent",
+          provider: "openclaw",
+        },
+        {
+          kind: "audio.transcription",
+          attachmentIndex: 1,
+          text: "second clip has speech",
+          provider: "groq",
+        },
+      ],
+    });
+    expect(body).toBe(
+      [
+        "[Audio 1/2]\nTranscript:\nfirst clip was silent",
+        "[Audio 2/2]\nTranscript:\nsecond clip has speech",
+      ].join("\n\n"),
+    );
+  });
 });