Agent Self-Attribution Errors

class AttributionCorrector:
    """Correct self-attribution errors in responses"""
    
    FORBIDDEN_SELF_ACTIONS = [
        r"i('ll| will) send",
        r"i('ll| will) share",
        r"i('ll| will) schedule",
        r"i('ll| will) process",
        r"i('ll| will) make sure",
        r"i('ll| will) follow up",
        r"i('ll| will) add you",
        r"i('ll| will) book",
        r"i('ll| will) arrange",
        r"i('ll| will) personally",
    ]
    
    CORRECTIONS = {
        r"i('ll| will) send (you |it |the )?(on whatsapp|the playbook)?":
            "the team can share",
        r"i('ll| will) make sure (you |someone )?":
            "",  # Remove entirely
        r"i('ll| will) follow up":
            "someone from the team may reach out",
        r"i('ll| will) schedule (a |your )?callback":
            "the team will call back",
        r"i('ll| will) process":
            "that'll be reviewed",
        r"i('ll| will) add you to":
            "you may be added to",
    }
    
    def check_attribution(self, response: str) -> list:
        """Find self-attribution errors"""
        errors = []
        response_lower = response.lower()
        
        for pattern in self.FORBIDDEN_SELF_ACTIONS:
            if re.search(pattern, response_lower):
                errors.append({
                    "pattern": pattern,
                    "severity": "high"
                })
        
        return errors
    
    def correct(self, response: str) -> str:
        """Correct self-attribution to team attribution"""
        result = response
        
        for pattern, replacement in self.CORRECTIONS.items():
            result = re.sub(pattern, replacement, result, 
                           flags=re.IGNORECASE)
        
        # Clean up extra spaces
        result = re.sub(r'\s+', ' ', result).strip()
        
        return result
    
    def reframe_as_permission(self, response: str) -> str:
        """Reframe sending as permission capture"""
        # "I'll send the playbook" → "Can the team share the playbook?"
        send_patterns = [
            (r"i('ll| will) send (you )?the playbook",
             "can the team share the playbook on WhatsApp"),
            (r"i('ll| will) share (the )?details",
             "want me to note down that the team can share details"),
        ]
        
        result = response
        for pattern, replacement in send_patterns:
            result = re.sub(pattern, replacement, result,
                           flags=re.IGNORECASE)
        
        return result


class CapabilityBoundaryEnforcer:
    """Enforce agent capability boundaries"""
    
    AGENT_CAN = [
        "capture information",
        "ask questions",
        "explain program",
        "note preferences",
        "get permission"
    ]
    
    AGENT_CANNOT = [
        "send messages",
        "schedule callbacks",
        "process applications",
        "make decisions",
        "guarantee timelines",
        "follow up personally"
    ]
    
    def get_appropriate_framing(self, action: str) -> str:
        """Get correct framing for action"""
        framings = {
            "send_playbook": 
                "Can the team share the playbook on WhatsApp?",
            "schedule_callback":
                "Evening or weekend—when's better for a callback?",
            "process_application":
                "The playbook has next steps.",
            "follow_up":
                "The next steps are in the playbook.",
            "add_to_group":
                "The format is explained in the playbook."
        }
        return framings.get(action, "")

instructions: |
  ## ATTRIBUTION RULES (CRITICAL)
  
  You capture information. You do NOT execute follow-up.
  
  NEVER SAY:
  - "I'll send you..."
  - "I'll share..."  
  - "I'll make sure..."
  - "I'll follow up..."
  - "I'll schedule..."
  - "I'll process..."
  - "I'll personally..."
  
  INSTEAD SAY:
  - "Can the TEAM share the playbook on WhatsApp?"
  - "The TEAM will handle follow-up."
  - "SOMEONE from the team may reach out."
  - "That's noted for the TEAM."
  - "The playbook has next steps."
  
  WHY: You cannot send messages, schedule calls, or process
  applications. Saying "I'll" creates false expectations.
  
  PERMISSION FRAMING:
  - You ask: "Can the team share X?"
  - You capture: "yes" or "no"
  - You do NOT: promise it will be sent
  
  TIMELINE:
  - Never promise "by tomorrow" or "within X days"
  - You don't control when the team acts