fix: real-model robustness — purpose_agent/actor.py

purpose_agent/actor.py

@@ -285,30 +285,42 @@ class Actor:
         """Best-effort extraction of action JSON from free-form text."""
         import re
 
-        # Try to find JSON block
-        json_match = re.search(r'\{[^{}]*"thought"[^{}]*\}', raw, re.DOTALL)
-        if json_match:
-            try:
-                return json.loads(json_match.group())
-            except json.JSONDecodeError:
-                pass
+        # Strategy 1: Try json.loads on the entire response (works if LLM outputs pure JSON)
+        text = raw.strip()
+        try:
+            return json.loads(text)
+        except (json.JSONDecodeError, ValueError):
+            pass
 
-        # Try to find JSON in code blocks
-        code_match = re.search(r'```(?:json)?\s*(\{.*?\})\s*```', raw, re.DOTALL)
+        # Strategy 2: Extract JSON from markdown code blocks
+        code_match = re.search(r'```(?:json)?\s*(\{.*\})\s*```', text, re.DOTALL)
         if code_match:
             try:
                 return json.loads(code_match.group(1))
-            except json.JSONDecodeError:
+            except (json.JSONDecodeError, ValueError):
                 pass
 
-        # Last resort: extract key-value pairs
+        # Strategy 3: Find the outermost { ... } by brace matching
+        start = text.find('{')
+        if start >= 0:
+            depth = 0
+            for i in range(start, len(text)):
+                if text[i] == '{': depth += 1
+                elif text[i] == '}': depth -= 1
+                if depth == 0:
+                    try:
+                        return json.loads(text[start:i+1])
+                    except (json.JSONDecodeError, ValueError):
+                        break
+
+        # Strategy 4: Extract key fields by regex
         thought = ""
-        thought_match = re.search(r'"thought"\s*:\s*"([^"]*)"', raw)
+        thought_match = re.search(r'"thought"\s*:\s*"((?:[^"\\]|\\.)*)"', text)
         if thought_match:
             thought = thought_match.group(1)
 
         action_name = "UNKNOWN"
-        name_match = re.search(r'"name"\s*:\s*"([^"]*)"', raw)
+        name_match = re.search(r'"name"\s*:\s*"([^"]*)"', text)
         if name_match:
             action_name = name_match.group(1)