Implement Basic Guardrails for Agent Output

Agent Guardrails

Guardrails prevent harmful, off-topic, or policy-violating outputs from being returned to users.

Task

Build OutputGuardrail that:

1. Blocks outputs containing prohibited patterns.
2. Enforces maximum output length.
3. Returns a structured validation result with safety status and reason.
4. Sanitizes output by truncating or masking violations.

Constraints

• Pattern matching is case-insensitive.
• Sanitized output replaces blocked patterns with [BLOCKED].
• If length exceeded, truncate and append '...[truncated]'.

from typing import List, Dict

class OutputGuardrail:
    def __init__(self):
        self.blocked_patterns: List[str] = []
        self.max_length: int = 2000

    def add_blocked_pattern(self, pattern: str) -> None:
        pass

    def set_max_length(self, length: int) -> None:
        pass

    def validate(self, output: str) -> Dict:
        # TODO: Return {'safe': bool, 'reason': str, 'sanitized': str}
        pass