| """
|
| Defender Agent
|
|
|
| Multi-agent adversarial LLM evaluation framework - Defender module.
|
|
|
| This module provides:
|
| - DefenderRequest and DefenderResponse schemas
|
| - Hybrid rule-based + model-based detection
|
| - Injection detection
|
| - Role override detection
|
| - Toxicity classification
|
| - Semantic policy checking
|
| - Risk score computation
|
| """
|
|
|
| from agents.defender.base import (
|
| BaseDefenseStrategy,
|
| InjectionDefenseStrategy,
|
| RoleOverrideDefenseStrategy,
|
| ToxicityDefenseStrategy,
|
| SemanticPolicyDefenseStrategy,
|
| )
|
| from agents.defender.engine import (
|
| DefenderEngine,
|
| get_defender_engine,
|
| )
|
| from agents.defender.registry import (
|
| DEFENSE_REGISTRY,
|
| DefenseStrategy,
|
| DefaultDefenseStrategy,
|
| PromptSanitizationStrategy,
|
| OutputFilteringStrategy,
|
| get_defense_strategy,
|
| initialize_registry,
|
| list_defense_strategies,
|
| register_defense_strategy,
|
| )
|
| from agents.defender.schemas import (
|
| DefenderConfig,
|
| DefenderLog,
|
| DefenderRequest,
|
| DefenderResponse,
|
| )
|
|
|
| __version__ = "0.1.0"
|
|
|
| __all__ = [
|
|
|
| "BaseDefenseStrategy",
|
| "InjectionDefenseStrategy",
|
| "RoleOverrideDefenseStrategy",
|
| "ToxicityDefenseStrategy",
|
| "SemanticPolicyDefenseStrategy",
|
|
|
| "DefenderEngine",
|
| "get_defender_engine",
|
|
|
| "DefenderRequest",
|
| "DefenderResponse",
|
| "DefenderLog",
|
| "DefenderConfig",
|
|
|
| "DEFENSE_REGISTRY",
|
| "DefenseStrategy",
|
| "DefaultDefenseStrategy",
|
| "PromptSanitizationStrategy",
|
| "OutputFilteringStrategy",
|
| "get_defense_strategy",
|
| "list_defense_strategies",
|
| "register_defense_strategy",
|
| "initialize_registry",
|
| ]
|
|
|
