v1.0.166: classify prompt v2 (LLM-suggested) + safe JSON parse with auto-repair

deploy/services/classify.py

@@ -50,13 +50,7 @@ def _call_llm_classify(header_text):
         data = resp.json()
 
     raw = data.get("choices", [{}])[0].get("message", {}).get("content", "")
-    # LLM может обернуть JSON в markdown-блок ```json ... ```
-    json_text = raw
-    if "```json" in json_text:
-        json_text = json_text.split("```json")[1].split("```")[0]
-    elif "```" in json_text:
-        json_text = json_text.split("```")[1].split("```")[0]
-    return json.loads(json_text.strip())
+    return _safe_json_parse(raw)
 
 
 def classify_batch(batch_id):
@@ -103,7 +97,34 @@ def classify_batch(batch_id):
     return {"ok": True, "total": total, "done": done, "failed": failed}
 
 
-def _smart_extract(elements_json):
+def _safe_json_parse(raw):
+    """Parse LLM response, fixing common JSON errors (multiline, markdown, unescaped chars)."""
+    if not raw:
+        raise ValueError("empty LLM response")
+
+    # Strip markdown wrappers
+    text = raw.strip()
+    if "```json" in text:
+        text = text.split("```json")[1].split("```")[0].strip()
+    elif "```" in text:
+        text = text.split("```")[1].split("```")[0].strip()
+
+    # Try strict parse first
+    try:
+        return json.loads(text)
+    except json.JSONDecodeError:
+        pass
+
+    # Fix common issues: unescaped newlines inside strings, unterminated strings
+    # Replace literal newlines inside JSON values
+    import re as _re
+    # Collapse multiline to single line
+    text = _re.sub(r'\n\s*', ' ', text)
+    # Remove trailing commas before closing braces
+    text = _re.sub(r',\s*}', '}', text)
+    text = _re.sub(r',\s*]', ']', text)
+
+    return json.loads(text)
     """
     Умная выжимка текста для классификации (решение Q3 от Opus).