hackagent.orchestrator.planning.planner

Attack planner.

An LLM chooses one registered technique, goals and parameters. Parameters are validated against the technique's pydantic JSON schema (the registry), not the TUI form specs.

PlannerError Objects

class PlannerError(Exception)

Raised when the planner cannot produce a usable plan.

AttackPlan Objects

@dataclass
class AttackPlan()

An LLM-chosen attack strategy for a target.

to_attack_config

def to_attack_config() -> Dict[str, Any]

Build the attack the planner chose, as a campaign's attacks entry.

summary

def summary() -> str

Human-readable one-screen summary of the plan.

plan_attack

def plan_attack(target: Dict[str, Any],
                *,
                model: str = DEFAULT_PLANNER_MODEL,
                goals: Optional[List[str]] = None,
                api_key: Optional[str] = None,
                temperature: float = 0.2,
                max_tokens: int = 1500) -> AttackPlan

Ask an LLM to choose an attack strategy and parameters for target.

AutoPlanResult Objects

@dataclass
class AutoPlanResult()

Combined output of :func:auto_plan.

auto_plan

def auto_plan(url: str,
              *,
              model: str = DEFAULT_PLANNER_MODEL,
              goals: Optional[List[str]] = None,
              target_kwargs: Optional[Dict[str, Any]] = None,
              **plan_kwargs: Any) -> AutoPlanResult

Build a web target for url and plan an attack against it.