@braze/sanity-plugin-ai-translation 0.2.55 → 0.2.57

package/dist/index.js

@@ -1527,8 +1527,35 @@ const globalTransformRegistry = new TransformRegistry(), callGPT5 = async (apiKe
   }, modelPricing = pricing[model] || pricing["gpt-5"], cachedInputRate = model === "gpt-5.1" ? 0.125 / 1e6 : modelPricing.input;
   return Math.max(0, tokens.prompt - tokens.cached) * modelPricing.input + tokens.cached * cachedInputRate + tokens.completion * modelPricing.output + tokens.reasoning * modelPricing.reasoning;
 }, deserializeDocument = (originalDoc, reviewedText) => {
-  const parsed = parseReviewFormat(reviewedText, originalDoc), result = cloneDeep(originalDoc);
+  devLog.debug("Raw GPT-5 response received", {
+    length: reviewedText.length,
+    preview: reviewedText.substring(0, 1e3),
+    hasReasoning: reviewedText.includes("<reasoning>"),
+    hasCodeBlocks: reviewedText.includes("```")
+  });
+  const cleanedText = preprocessResponse(reviewedText);
+  cleanedText.length !== reviewedText.length && devLog.debug("Response after preprocessing", {
+    originalLength: reviewedText.length,
+    cleanedLength: cleanedText.length,
+    preview: cleanedText.substring(0, 500)
+  });
+  const parsed = parseReviewFormat(cleanedText, originalDoc), result = cloneDeep(originalDoc);
   return applyTextChanges(result, parsed), result;
+}, preprocessResponse = (text) => {
+  const originalLength = text.length;
+  let cleaned = text;
+  const beforeReasoning = cleaned.length;
+  cleaned = cleaned.replace(/<reasoning>[\s\S]*?<\/reasoning>/gi, "");
+  const reasoningRemoved = beforeReasoning !== cleaned.length, beforeCodeBlocks = cleaned.length;
+  cleaned = cleaned.replace(/^```[a-z]*\s*$/gim, ""), cleaned = cleaned.replace(/^```\s*$/gim, "");
+  const codeBlocksRemoved = beforeCodeBlocks !== cleaned.length, finalLength = cleaned.trim().length;
+  return (reasoningRemoved || codeBlocksRemoved) && devLog.debug("Preprocessed GPT-5 response", {
+    originalLength,
+    finalLength,
+    reasoningRemoved,
+    codeBlocksRemoved,
+    preview: cleaned.trim().substring(0, 200)
+  }), cleaned.trim();
 }, parseReviewFormat = (text, originalDoc) => {
   const root = [], stack = [], lines = text.split(`
 `);
@@ -3902,10 +3929,14 @@ function serializeVTTCaptionText(captions) {
     firstCaption: captions.length > 0 ? captions[0] : null
   }), !Array.isArray(captions) || captions.length === 0)
     return "";
-  const result = captions.map((caption, index) => (devLog.debug(`Processing caption ${index}:`, {
-    text: caption.text,
-    hasText: typeof caption.text == "string"
-  }), `[${index}] ${caption.text || ""}`)).join(`
+  const result = captions.map((caption, index) => {
+    devLog.debug(`Processing caption ${index}:`, {
+      text: caption.text,
+      hasText: typeof caption.text == "string"
+    });
+    const escapedText = (caption.text || "").replace(/\n/g, "\\n");
+    return `[${index}] ${escapedText}`;
+  }).join(`
 `);
   return devLog.debug("serializeVTTCaptionText - output:", {
     resultLength: result.length,
@@ -3922,29 +3953,46 @@ function deserializeVTTCaptionText(translationText, originalMetadata) {
     return [];
   const result = [], lines = translationText.split(`
 `), lineRegex = /^\[(\d+)\]\s*(.*)$/;
+  let currentIndex = null, currentTextParts = [];
   for (const line of lines) {
     const match = line.match(lineRegex);
-    if (!match) {
-      devLog.warn("Failed to parse caption line", { line });
-      continue;
-    }
-    const index = Number.parseInt(match[1], 10), text = match[2];
-    devLog.debug("Found caption:", {
-      index,
-      text
-    });
-    const metadata = originalMetadata.find((m) => m._originalIndex === index);
+    if (match) {
+      if (currentIndex !== null) {
+        const text = currentTextParts.join(`
+`).replace(/\\n/g, `
+`), metadata = originalMetadata.find((m) => m._originalIndex === currentIndex);
+        if (metadata) {
+          const { _originalIndex, ...metadataWithoutIndex } = metadata;
+          result.push({
+            ...metadataWithoutIndex,
+            text
+          });
+        } else
+          devLog.warn("Missing metadata for caption", { index: currentIndex }), result.push({
+            _type: "vtt.cue",
+            _key: `caption-${currentIndex}`,
+            text,
+            startTime: 0,
+            endTime: 0
+          });
+      }
+      currentIndex = Number.parseInt(match[1], 10), currentTextParts = [match[2]];
+    } else currentIndex !== null ? currentTextParts.push(line) : devLog.warn("Failed to parse caption line", { line });
+  }
+  if (currentIndex !== null) {
+    const text = currentTextParts.join(`
+`).replace(/\\n/g, `
+`), metadata = originalMetadata.find((m) => m._originalIndex === currentIndex);
     if (metadata) {
-      devLog.debug(`Found matching metadata for index ${index}`);
       const { _originalIndex, ...metadataWithoutIndex } = metadata;
       result.push({
         ...metadataWithoutIndex,
         text
       });
     } else
-      devLog.warn("Missing metadata for caption", { index }), result.push({
+      devLog.warn("Missing metadata for caption", { index: currentIndex }), result.push({
         _type: "vtt.cue",
-        _key: `caption-${index}`,
+        _key: `caption-${currentIndex}`,
         text,
         startTime: 0,
         endTime: 0
@@ -4181,10 +4229,14 @@ function serializeWistiaCaptionText(cues) {
     firstCue: cues.length > 0 ? cues[0] : null
   }), !Array.isArray(cues) || cues.length === 0)
     return "";
-  const result = cues.map((cue) => (devLog.debug(`Processing cue ${cue.index}:`, {
-    text: cue.text,
-    hasText: typeof cue.text == "string"
-  }), `[${cue.index}] ${cue.text}`)).join(`
+  const result = cues.map((cue) => {
+    devLog.debug(`Processing cue ${cue.index}:`, {
+      text: cue.text,
+      hasText: typeof cue.text == "string"
+    });
+    const escapedText = cue.text.replace(/\n/g, "\\n");
+    return `[${cue.index}] ${escapedText}`;
+  }).join(`
 `);
   return devLog.debug("serializeWistiaCaptionText - output:", {
     resultLength: result.length,
@@ -4203,19 +4255,24 @@ function deserializeWistiaCaptionText(translationText, srtCuesMetadata) {
 `), lineRegex = /^\[(\d+)\]\s*(.*)$/, metadataByIndex = /* @__PURE__ */ new Map();
   for (const metadata of srtCuesMetadata)
     metadataByIndex.set(metadata.index, { startTime: metadata.startTime, endTime: metadata.endTime });
+  let currentIndex = null, currentTextParts = [];
   for (const line of lines) {
     const match = line.match(lineRegex);
-    if (!match) {
-      devLog.warn("Failed to parse caption line", { line });
-      continue;
-    }
-    const index = Number.parseInt(match[1], 10), text = match[2];
-    devLog.debug("Found caption:", {
-      index,
-      text
-    });
-    const metadata = metadataByIndex.get(index);
-    metadata ? (devLog.debug(`Found matching metadata for index ${index}`), result.push(`${index}`), result.push(`${metadata.startTime} --> ${metadata.endTime}`), result.push(text), result.push("")) : devLog.warn("Missing metadata for caption", { index });
+    if (match) {
+      if (currentIndex !== null) {
+        const text = currentTextParts.join(`
+`).replace(/\\n/g, `
+`), metadata = metadataByIndex.get(currentIndex);
+        metadata ? (result.push(`${currentIndex}`), result.push(`${metadata.startTime} --> ${metadata.endTime}`), result.push(text), result.push("")) : devLog.warn("Missing metadata for caption", { index: currentIndex });
+      }
+      currentIndex = Number.parseInt(match[1], 10), currentTextParts = [match[2]];
+    } else currentIndex !== null ? currentTextParts.push(line) : devLog.warn("Failed to parse caption line", { line });
+  }
+  if (currentIndex !== null) {
+    const text = currentTextParts.join(`
+`).replace(/\\n/g, `
+`), metadata = metadataByIndex.get(currentIndex);
+    metadata ? (result.push(`${currentIndex}`), result.push(`${metadata.startTime} --> ${metadata.endTime}`), result.push(text), result.push("")) : devLog.warn("Missing metadata for caption", { index: currentIndex });
   }
   result.length > 0 && result[result.length - 1] === "" && result.pop();
   const reconstructed = result.join(`