Merge 463c0788bc into 09be5d45d5

src/media-understanding/apply.test.ts

@@ -547,6 +547,73 @@ describe("applyMediaUnderstanding", () => {
     expect(ctx.Body).toContain("a\tb\tc");
   });
 
+  it("does not treat OGG audio as text even with ASCII-heavy headers (issue #1989)", async () => {
+    const { applyMediaUnderstanding } = await loadApply();
+    const dir = await fs.mkdtemp(path.join(os.tmpdir(), "moltbot-media-"));
+    const oggPath = path.join(dir, "voice.ogg");
+    // Simulate OGG file with "OggS" magic bytes followed by ASCII-heavy content
+    // that would pass looksLikeUtf8Text() if not for magic byte detection
+    // Real OGG Opus files (Telegram voice) have similar structure
+    const oggMagic = Buffer.from([0x4f, 0x67, 0x67, 0x53]); // "OggS"
+    const fakeMetadata = Buffer.from("ENCODER=test\tVERSION=1\nTITLE=hello\t");
+    const oggBuffer = Buffer.concat([oggMagic, fakeMetadata]);
+    await fs.writeFile(oggPath, oggBuffer);
+
+    const ctx: MsgContext = {
+      Body: "<media:audio>",
+      MediaPath: oggPath,
+      MediaType: "audio/ogg",
+    };
+    const cfg: MoltbotConfig = {
+      tools: {
+        media: {
+          audio: { enabled: false },
+          image: { enabled: false },
+          video: { enabled: false },
+        },
+      },
+    };
+
+    const result = await applyMediaUnderstanding({ ctx, cfg });
+
+    // OGG should NOT be treated as a text file
+    expect(result.appliedFile).toBe(false);
+    expect(ctx.Body).not.toContain("<file");
+  });
+
+  it("does not treat MP3 with ID3 tag as text even with ASCII-heavy metadata", async () => {
+    const { applyMediaUnderstanding } = await loadApply();
+    const dir = await fs.mkdtemp(path.join(os.tmpdir(), "moltbot-media-"));
+    const mp3Path = path.join(dir, "song.mp3");
+    // Simulate MP3 file with ID3v2 tag followed by ASCII-heavy metadata
+    // ID3 tags can contain lyrics, comments, and other text that passes looksLikeUtf8Text()
+    const id3Magic = Buffer.from([0x49, 0x44, 0x33]); // "ID3"
+    const fakeMetadata = Buffer.from("TIT2=Song Title\tTPE1=Artist Name\nTALB=Album\t");
+    const mp3Buffer = Buffer.concat([id3Magic, fakeMetadata]);
+    await fs.writeFile(mp3Path, mp3Buffer);
+
+    const ctx: MsgContext = {
+      Body: "<media:audio>",
+      MediaPath: mp3Path,
+      MediaType: "audio/mpeg",
+    };
+    const cfg: MoltbotConfig = {
+      tools: {
+        media: {
+          audio: { enabled: false },
+          image: { enabled: false },
+          video: { enabled: false },
+        },
+      },
+    };
+
+    const result = await applyMediaUnderstanding({ ctx, cfg });
+
+    // MP3 with ID3 should NOT be treated as a text file
+    expect(result.appliedFile).toBe(false);
+    expect(ctx.Body).not.toContain("<file");
+  });
+
   it("escapes XML special characters in filenames to prevent injection", async () => {
     const { applyMediaUnderstanding } = await loadApply();
     const dir = await fs.mkdtemp(path.join(os.tmpdir(), "openclaw-media-"));

src/media-understanding/apply.ts

@@ -162,6 +162,48 @@ function looksLikeUtf8Text(buffer?: Buffer): boolean {
   return printable / total > 0.85;
 }
 
+/**
+ * Detects binary audio/video formats by magic bytes (file signatures).
+ *
+ * OGG files (used by Telegram voice messages as OGG Opus) have ASCII-heavy
+ * headers that can pass looksLikeUtf8Text() due to Vorbis/Opus comment metadata
+ * containing printable strings. This causes them to be misidentified as text.
+ *
+ * Magic bytes provide authoritative format detection independent of MIME type
+ * or file extension, following established file format specifications.
+ *
+ * References:
+ * - OGG container: RFC 3533 (https://datatracker.ietf.org/doc/html/rfc3533)
+ *   Section 6: "OggS" capture pattern at byte offset 0
+ * - MP3 ID3v2: id3.org spec (https://id3.org/id3v2.4.0-structure)
+ *   Section 3.1: "ID3" identifier at file start
+ *
+ * @see https://github.com/moltbot/moltbot/issues/1989
+ */
+function hasBinaryAudioMagic(buffer?: Buffer): boolean {
+  if (!buffer || buffer.length < 4) return false;
+  // OGG container format: "OggS" signature (RFC 3533 Section 6)
+  // Covers OGG Vorbis, OGG Opus (Telegram voice), OGG Theora, etc.
+  if (
+    buffer[0] === 0x4f && // 'O'
+    buffer[1] === 0x67 && // 'g'
+    buffer[2] === 0x67 && // 'g'
+    buffer[3] === 0x53 // 'S'
+  ) {
+    return true;
+  }
+  // MP3 with ID3v2 tag: "ID3" signature (id3.org spec Section 3.1)
+  // ID3v2 tags can contain large amounts of ASCII text (lyrics, comments)
+  if (
+    buffer[0] === 0x49 && // 'I'
+    buffer[1] === 0x44 && // 'D'
+    buffer[2] === 0x33 // '3'
+  ) {
+    return true;
+  }
+  return false;
+}
+
 function decodeTextSample(buffer?: Buffer): string {
   if (!buffer || buffer.length === 0) return "";
   const sample = buffer.subarray(0, Math.min(buffer.length, 8192));
@@ -242,7 +284,11 @@ async function extractFileBlocks(params: {
     const forcedTextMimeResolved = forcedTextMime ?? resolveTextMimeFromName(nameHint ?? "");
     const utf16Charset = resolveUtf16Charset(bufferResult?.buffer);
     const textSample = decodeTextSample(bufferResult?.buffer);
-    const textLike = Boolean(utf16Charset) || looksLikeUtf8Text(bufferResult?.buffer);
+    // Check if content looks like text, but exclude files with known binary audio magic bytes
+    // OGG files can pass looksLikeUtf8Text() due to ASCII-heavy headers (>85% printable)
+    const textLike =
+      (Boolean(utf16Charset) || looksLikeUtf8Text(bufferResult?.buffer)) &&
+      !hasBinaryAudioMagic(bufferResult?.buffer);
     if (!forcedTextMimeResolved && kind === "audio" && !textLike) {
       continue;
     }