Update main.py

main.py

@@ -103,24 +103,48 @@ def _normalize_lines(text: str) -> List[str]:
         return [raw.strip()] if raw.strip() else []
 
 def _ensure_numbering(text: str) -> str:
+    """
+    Cleanly number steps: 1) 2) 3) ...
+    - Merge dangling numeric lines like '2' that are followed by text.
+    - Strip existing numbers/bullets and re-number.
+    """
     lines = _normalize_lines(text)
     if not lines:
         return text or ""
-    normalized: List[str] = []
-    for ln in lines:
-        ln2 = re.sub(
+
+    # Merge 'dangling number' lines (e.g., "2" on its own line)
+    merged: List[str] = []
+    i = 0
+    while i < len(lines):
+        ln = lines[i].strip()
+        if re.fullmatch(r"\d+[.)]?", ln) and (i + 1) < len(lines):
+            # merge number with the next line's text
+            merged.append(lines[i + 1].strip())
+            i += 2
+        else:
+            merged.append(ln)
+            i += 1
+
+    # Remove any bullet/number prefix
+    cleaned = [
+        re.sub(
             r"^\s*(?:\d+[.\)\]]\s+|(?:step\s*\d+:?)\s+|[\-\*\u2022]\s+)",
             "",
             ln,
             flags=re.IGNORECASE,
         ).strip()
-        normalized.append(ln2)
-    if len(normalized) == 1:
-        para = normalized[0]
+        for ln in merged
+    ]
+
+    # If single paragraph, split on separators
+    if len(cleaned) == 1:
+        para = cleaned[0]
         parts = [p.strip() for p in re.split(r"(?:;|\.\s+(?=[A-Z0-9]))", para) if p.strip()]
         if len(parts) > 1:
-            normalized = parts
-    return "\n".join([f"{i+1}) {ln}" for i, ln in enumerate(normalized)])
+            cleaned = parts
+
+    # Final numbering
+    return "\n".join([f"{idx+1}) {ln}" for idx, ln in enumerate(cleaned) if ln])
 
 def _friendly_permission_reply(raw: str) -> str:
     line = (raw or "").strip()
@@ -235,7 +259,8 @@ def _normalize_for_match(text: str) -> str:
     return t
 
 def _split_sentences(ctx: str) -> List[str]:
-    raw_sents = re.split(r"(?<=([.!?]))\s+|\n+|\-\s*|\*\s*", ctx or "")
+    # Non-capturing split to avoid stray tokens that break numbering
+    raw_sents = re.split(r"(?<=[.!?])\s+|\n+|\-\s*|\*\s*", ctx or "")
     return [s.strip() for s in raw_sents if s and len(s.strip()) > 2]
 
 def _filter_context_for_query(context: str, query: str) -> Tuple[str, Dict[str, Any]]:
@@ -316,7 +341,7 @@ async def chat_with_ai(input_data: ChatInput):
     try:
         msg_norm = (input_data.user_message or "").lower().strip()
 
-        # Yes/No
+        # Quick yes/no
         if msg_norm in ("yes", "y", "sure", "ok", "okay"):
             return {
                 "bot_response": "Great! Tell me what you'd like to do next — check another ticket, create an incident, or describe your issue.",