fix: handle truncated AI responses and relax progressive tree validation

- Strip unclosed [TREE_UPDATE] and [METADATA] blocks from display when
  response is truncated by max_tokens
- Increase send_message max_tokens from 2000 to 8000 to prevent
  truncation of large tree JSON
- Use lightweight validation for progressive tree updates (valid root
  node only) instead of strict 5-node minimum — strict validation
  still applies at final /generate step

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

backend/app/core/ai_chat_service.py

@@ -187,6 +187,13 @@ def _parse_ai_response(raw_response: str) -> dict[str, Any]:
         except (json.JSONDecodeError, ValueError) as e:
             logger.warning("Failed to parse tree update JSON: %s", e)
         result["content"] = raw_response[: tree_match.start()] + raw_response[tree_match.end() :]
+    else:
+        # Handle truncated response — opening tag exists but no closing tag
+        # (happens when max_tokens cuts off the JSON block)
+        truncated_match = re.search(r"\[TREE_UPDATE\][\s\S]*$", raw_response)
+        if truncated_match:
+            logger.warning("Truncated [TREE_UPDATE] block detected (no closing tag) — stripping from display")
+            result["content"] = raw_response[: truncated_match.start()]
 
     # Extract [PHASE:name]
     phase_match = re.search(r"\[PHASE:(\w+)\]", result["content"])
@@ -205,6 +212,11 @@ def _parse_ai_response(raw_response: str) -> dict[str, Any]:
         except (json.JSONDecodeError, ValueError) as e:
             logger.warning("Failed to parse metadata JSON: %s", e)
         result["content"] = result["content"][: meta_match.start()] + result["content"][meta_match.end() :]
+    else:
+        truncated_meta = re.search(r"\[METADATA\][\s\S]*$", result["content"])
+        if truncated_meta:
+            logger.warning("Truncated [METADATA] block detected — stripping from display")
+            result["content"] = result["content"][: truncated_meta.start()]
 
     # Clean up extra whitespace from marker removal
     result["content"] = re.sub(r"\n{3,}", "\n\n", result["content"]).strip()
@@ -294,18 +306,21 @@ async def send_message(
     response_text, input_tokens, output_tokens = await provider.generate_text(
         system_prompt=system_prompt,
         messages=provider_messages,
-        max_tokens=2000,
+        max_tokens=8000,
     )
 
     parsed = _parse_ai_response(response_text)
 
-    # Validate tree update if present
+    # Validate tree update if present (lightweight check for progressive builds —
+    # only require valid root structure, not min node counts)
     tree_update = parsed["tree_update"]
     if tree_update:
-        errors = validate_generated_tree(tree_update)
+        if not isinstance(tree_update, dict) or tree_update.get("type") != "decision":
-        if errors:
+            logger.warning("AI tree update rejected: root must be a decision node")
-            logger.warning("AI tree update failed validation: %s", errors)
+            tree_update = None
-            tree_update = None  # Silently discard invalid updates
+        elif not tree_update.get("id"):
+            logger.warning("AI tree update rejected: root node missing id")
+            tree_update = None
 
     # Update session state
     history.append({"role": "assistant", "content": parsed["content"], "timestamp": now_iso})