Skip to main content

hackagent.risks.base

Base vulnerability class for all hackagent risk assessments.

Architecture (mirrors the attack layer): HackAgent.hack() → AttackOrchestrator → BaseAttack (technique) ↕ BaseVulnerability ← vulnerability.assess()

Each concrete vulnerability:

  1. Defines an Enum of sub-types in its types.py
  2. Provides prompt templates in its templates.py
  3. Extends this class in its main module (e.g. bias.py)

BaseVulnerability Objects​

class BaseVulnerability(abc.ABC)

Abstract base class for all vulnerabilities.

Each vulnerability carries an Enum of sub-types that can be individually selected.

Subclasses must set the class-level attributes: - name – human-readable name - description – one-liner for reports - ALLOWED_TYPES – list of valid sub-type values (strings) - _type_enum – the Enum class used for validation

Parameters​

types : list[Enum] Sub-types to evaluate (defaults to all allowed types).

get_types​

def get_types() -> List[Enum]

Return the list of selected sub-type enums.

get_values​

def get_values() -> List[str]

Return selected sub-type values as plain strings.

assess​

def assess(model_callback: Any = None,
           purpose: Optional[str] = None) -> Dict[str, Any]

Evaluate the target model for this vulnerability.

Returns a dict mapping each sub-type value to its test-case results.

a_assess​

async def a_assess(model_callback: Any = None,
                   purpose: Optional[str] = None) -> Dict[str, Any]

Async variant of :pymeth:assess.

simulate_attacks​

def simulate_attacks(purpose: Optional[str] = None) -> List[str]

Generate baseline attack prompts for each selected sub-type.

Returns a flat list of attack strings.

a_simulate_attacks​

async def a_simulate_attacks(purpose: Optional[str] = None) -> List[str]

Async variant of :pymeth:simulate_attacks.