diff --git a/backend/app/core/ai_tree_generator_service.py b/backend/app/core/ai_tree_generator_service.py
index 805c84f7..34af11d2 100644
--- a/backend/app/core/ai_tree_generator_service.py
+++ b/backend/app/core/ai_tree_generator_service.py
@@ -9,6 +9,7 @@ System prompts are static constants to enable Anthropic prompt caching.
 """
 import json
 import logging
+import re
 import uuid
 from typing import Any
 
@@ -94,6 +95,15 @@ Return a corrected full JSON object only. No markdown, no prose, no code fences.
 Fix ALL listed errors while maintaining the same troubleshooting/procedural logic."""
 
 
+def _strip_markdown_fences(text: str) -> str:
+    """Strip markdown code fences if the model wrapped its JSON response."""
+    text = text.strip()
+    match = re.match(r"^```(?:json)?\s*([\s\S]*?)```$", text)
+    if match:
+        return match.group(1).strip()
+    return text
+
+
 def _get_client() -> anthropic.AsyncAnthropic:
     """Get configured async Anthropic client."""
     if not settings.ANTHROPIC_API_KEY:
@@ -141,7 +151,7 @@ async def scaffold_branches(
         messages=[{"role": "user", "content": user_message}],
     )
 
-    raw_text = response.content[0].text
+    raw_text = _strip_markdown_fences(response.content[0].text)
     input_tokens = response.usage.input_tokens
     output_tokens = response.usage.output_tokens
     cost = _estimate_cost(input_tokens, output_tokens)
@@ -197,7 +207,7 @@ async def generate_branch_detail(
             messages=messages,
         )
 
-        raw_text = response.content[0].text
+        raw_text = _strip_markdown_fences(response.content[0].text)
         total_input += response.usage.input_tokens
         total_output += response.usage.output_tokens