Skip to main content

hackagent.attacks.techniques.baseline.attack

Baseline attack implementation.

Uses predefined prompt templates to attempt jailbreaks by combining templates with harmful goals.

BaselineAttack Objects​

class BaselineAttack(BaseAttack)

Baseline attack using predefined prompt templates.

Combines a library of prompt templates across several jailbreak categories with each goal string to produce attack prompts, sends them to the target model, and evaluates responses using a configurable evaluator (pattern-matching, keyword, or LLM judge).

Pipeline stages​

  1. Generation (:func:~hackagent.attacks.techniques.baseline.generation.execute) — selects up to templates_per_category templates from each category in template_categories, injects each goal, and collects target-model responses.
  2. Evaluation (:func:~hackagent.attacks.techniques.baseline.evaluation.execute) — scores responses for jailbreak success using the configured evaluator_type ("pattern", "keyword", or "llm_judge").

This attack is useful as a sanity-check baseline: it requires no additional LLM (unlike PAIR/TAP/AdvPrefix) and surfaces naive template weaknesses in the target model.

Attributes:

  • ``4 - Merged baseline configuration dictionary.
  • ``5 - Authenticated HackAgent API client.
  • ``6 - Router for the victim model.
  • 7 - Hierarchical logger at hackagent.attacks.baseline``.

__init__​

def __init__(config: Optional[Dict[str, Any]] = None,
             client: Optional[AuthenticatedClient] = None,
             agent_router: Optional[AgentRouter] = None)

Initialize baseline attack.

Arguments:

  • config - Configuration override dictionary merged into :data:~hackagent.attacks.techniques.baseline.config.DEFAULT_TEMPLATE_CONFIG.
  • client - Authenticated HackAgent API client.
  • agent_router - Router for the victim model.

Raises:

  • ValueError - If client or agent_router is None.

run​

@with_tui_logging(logger_name="hackagent.attacks", level=logging.INFO)
def run(goals: List[str]) -> Dict[str, Any]

Execute baseline attack.

Uses TrackingCoordinator for unified pipeline and goal tracking.

Arguments:

  • goals - List of harmful goals to test

Returns:

Dictionary with 'evaluated' and 'summary' DataFrames