hackagent 0.5.0__tar.gz → 0.7.0__tar.gz

{hackagent-0.5.0 → hackagent-0.7.0}/.gitignore

@@ -4,6 +4,7 @@ logs/
 .ruff_cache/
 reports/
 .adk/
+slurm/
 
 # Editors
 .vscode/
@@ -129,4 +130,7 @@ venv.bak/
 # mypy
 .mypy_cache/
 .dmypy.json
-dmypy.json
+dmypy.json
+
+
+.copilotignore

{hackagent-0.5.0 → hackagent-0.7.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hackagent
-Version: 0.5.0
+Version: 0.7.0
 Summary: HackAgent is an open-source security toolkit to detect vulnerabilities of your AI Agents.
 Author-email: AI Security Lab <ais@ai4i.it>
 License: Apache-2.0
@@ -15,18 +15,19 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Requires-Python: >=3.10
-Requires-Dist: attrs>=21.0.0
 Requires-Dist: click>=8.1.0
+Requires-Dist: datasets>=2.14.0
+Requires-Dist: faiss-cpu>=1.13.2
+Requires-Dist: httpx>=0.27.0
 Requires-Dist: litellm>=1.69.2
+Requires-Dist: nicegui>=2.0
 Requires-Dist: openai>=1.0.0
-Requires-Dist: pydantic>=2.0
+Requires-Dist: pydantic[email]>=2.0
 Requires-Dist: python-dateutil>=2.8.0
 Requires-Dist: pyyaml>=6.0.0
 Requires-Dist: requests>=2.31.0
 Requires-Dist: rich>=14.0.0
 Requires-Dist: textual>=1.0.0
-Provides-Extra: datasets
-Requires-Dist: datasets>=2.14.0; extra == 'datasets'
 Description-Content-Type: text/markdown
 
 <div align="center">

hackagent-0.7.0/hackagent/__init__.py

@@ -0,0 +1,25 @@
+# Copyright 2026 - AI4I. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""A client library for accessing HackAgent API"""
+
+from .agent import HackAgent
+from .server.client import AuthenticatedClient, Client
+from .logger import setup_package_logging
+from .router.types import AgentTypeEnum
+from .server.storage.base import StorageBackend
+from .server.storage.local import LocalBackend
+from .server.storage.remote import RemoteBackend
+
+# Configure RichHandler for all hackagent.* loggers on first import.
+setup_package_logging()
+
+__all__ = (
+    "AgentTypeEnum",
+    "AuthenticatedClient",
+    "Client",
+    "HackAgent",
+    "LocalBackend",
+    "RemoteBackend",
+    "StorageBackend",
+)

{hackagent-0.5.0 → hackagent-0.7.0}/hackagent/agent.py

@@ -1,20 +1,36 @@
 # Copyright 2026 - AI4I. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
-import logging
+from hackagent.logger import get_logger
 from typing import TYPE_CHECKING, Any, Dict, Optional, Union
 
 from hackagent import utils
-from hackagent.client import AuthenticatedClient
 from hackagent.errors import HackAgentError
 from hackagent.router import AgentRouter
 from hackagent.router.types import AgentTypeEnum
+from hackagent.server.storage.base import StorageBackend
 
 # Lazy import for attack orchestrators to avoid ~0.5s startup delay
 if TYPE_CHECKING:
     pass
 
-logger = logging.getLogger(__name__)
+logger = get_logger(__name__)
+
+
+def _resolve_target_config(target_config: Optional[Dict[str, Any]]) -> Dict[str, Any]:
+    """Return normalized victim request defaults for the configured router."""
+    from hackagent.attacks.techniques.config import default_target
+
+    resolved = default_target()
+    if not target_config:
+        return resolved
+
+    merged = {key: value for key, value in target_config.items() if value is not None}
+    if "request_timeout" in merged and "timeout" not in merged:
+        merged["timeout"] = merged.pop("request_timeout")
+
+    resolved.update(merged)
+    return resolved
 
 
 class HackAgent:
@@ -50,6 +66,7 @@ class HackAgent:
         raise_on_unexpected_status: bool = False,
         timeout: Optional[float] = None,
         metadata: Optional[Dict[str, Any]] = None,
+        target_config: Optional[Dict[str, Any]] = None,
         adapter_operational_config: Optional[Dict[str, Any]] = None,
     ):
         """
@@ -84,32 +101,70 @@ class HackAgent:
                 authenticated client. Defaults to `None` (which might mean a
                 default timeout from the underlying HTTP library is used).
             metadata: Optional dictionary containing agent-specific metadata.
+            target_config: Optional default request settings for the configured
+                victim model. This is the preferred place to define target-side
+                generation defaults such as `max_tokens`, `temperature`,
+                and `timeout`.
             adapter_operational_config: Optional configuration for the agent adapter.
         """
 
         resolved_auth_token = utils.resolve_api_token(direct_api_key_param=api_key)
 
-        # Use default base_url if not provided
-        if base_url is None:
-            base_url = "https://api.hackagent.dev"
+        if resolved_auth_token:
+            from hackagent.server.client import AuthenticatedClient
+            from hackagent.server.storage.remote import RemoteBackend
 
-        self.client = AuthenticatedClient(
-            base_url=base_url,
-            token=resolved_auth_token,
-            prefix="Bearer",
-            raise_on_unexpected_status=raise_on_unexpected_status,
-            timeout=timeout,
-        )
+            _base_url = base_url or "https://api.hackagent.dev"
+            _client = AuthenticatedClient(
+                base_url=_base_url,
+                token=resolved_auth_token,
+                prefix="Bearer",
+                raise_on_unexpected_status=raise_on_unexpected_status,
+                timeout=timeout,
+            )
+            self.backend: StorageBackend = RemoteBackend(_client)
+            logger.info("HackAgent using remote backend → %s", _base_url)
+        else:
+            from hackagent.server.storage.local import LocalBackend
+
+            self.backend = LocalBackend()
+            logger.info(
+                "HackAgent using local backend → ~/.local/share/hackagent/hackagent.db"
+            )
+
+        # Keep self.client as the raw HTTP client for backward compat
+        # (adapters that need it can access it via backend.get_api_key())
+        self.client = getattr(self.backend, "_client", None)
 
         processed_agent_type = utils.resolve_agent_type(agent_type)
+        self.target_config = _resolve_target_config(target_config)
+        explicit_target_config = (
+            {
+                key: value
+                for key, value in (target_config or {}).items()
+                if value is not None
+            }
+            if target_config
+            else {}
+        )
+
+        router_metadata = {
+            key: value
+            for key, value in {**(metadata or {}), **explicit_target_config}.items()
+            if value is not None
+        }
+        router_operational_config = {
+            **self.target_config,
+            **(adapter_operational_config or {}),
+        }
 
         self.router = AgentRouter(
-            client=self.client,
-            name=name,
+            backend=self.backend,
+            name=name or endpoint,  # fall back to endpoint if no name provided
             agent_type=processed_agent_type,
             endpoint=endpoint,
-            metadata=metadata,
-            adapter_operational_config=adapter_operational_config,
+            metadata=router_metadata,
+            adapter_operational_config=router_operational_config,
         )
 
         # Attack strategies are lazy-loaded to improve startup time
@@ -122,16 +177,28 @@ class HackAgent:
             # Import here to avoid circular imports and improve startup time
             from hackagent.attacks.registry import (
                 AdvPrefixOrchestrator,
+                AutoDANTurboOrchestrator,
                 BaselineOrchestrator,
+                BoNOrchestrator,
+                CipherChatOrchestrator,
+                H4rm3lOrchestrator,
+                PAPOrchestrator,
                 PAIROrchestrator,
                 FlipAttackOrchestrator,
+                TAPOrchestrator,
             )
 
             self._attack_strategies = {
                 "advprefix": AdvPrefixOrchestrator(hack_agent=self),
+                "autodan_turbo": AutoDANTurboOrchestrator(hack_agent=self),
                 "baseline": BaselineOrchestrator(hack_agent=self),
+                "bon": BoNOrchestrator(hack_agent=self),
+                "cipherchat": CipherChatOrchestrator(hack_agent=self),
                 "pair": PAIROrchestrator(hack_agent=self),
                 "flipattack": FlipAttackOrchestrator(hack_agent=self),
+                "tap": TAPOrchestrator(hack_agent=self),
+                "h4rm3l": H4rm3lOrchestrator(hack_agent=self),
+                "pap": PAPOrchestrator(hack_agent=self),
             }
         return self._attack_strategies
 

{hackagent-0.5.0 → hackagent-0.7.0}/hackagent/attacks/__init__.py

@@ -33,13 +33,21 @@ The module integrates with the HackAgent backend for result tracking and reporti
 from .registry import (
     ATTACK_REGISTRY,
     AdvPrefixOrchestrator,
+    AutoDANTurboOrchestrator,
     BaselineOrchestrator,
+    CipherChatOrchestrator,
     PAIROrchestrator,
+    FlipAttackOrchestrator,
+    TAPOrchestrator,
 )
 
 __all__ = [
     "ATTACK_REGISTRY",
     "AdvPrefixOrchestrator",
+    "AutoDANTurboOrchestrator",
     "BaselineOrchestrator",
+    "CipherChatOrchestrator",
     "PAIROrchestrator",
+    "FlipAttackOrchestrator",
+    "TAPOrchestrator",
 ]

{hackagent-0.5.0 → hackagent-0.7.0}/hackagent/attacks/evaluator/__init__.py

@@ -23,6 +23,7 @@ Usage:
         NuancedEvaluator,
         JailbreakBenchEvaluator,
         HarmBenchEvaluator,
+        HarmBenchVariantEvaluator,
         EVALUATOR_MAP,
         AssertionResult,
         # Pattern-based evaluators
@@ -33,6 +34,10 @@ Usage:
         calculate_success_rate,
         calculate_per_goal_metrics,
         generate_summary_report,
+        calculate_confidence_score,
+        calculate_majority_vote_asr,
+        calculate_fleiss_kappa,
+        calculate_per_judge_strictness,
         # Server sync
         sync_evaluation_to_server,
         update_single_result,
@@ -44,8 +49,10 @@ from hackagent.attacks.evaluator.evaluation_step import BaseEvaluationStep
 from hackagent.attacks.evaluator.judge_evaluators import (
     EVALUATOR_MAP,
     HarmBenchEvaluator,
+    HarmBenchVariantEvaluator,
     JailbreakBenchEvaluator,
     NuancedEvaluator,
+    OnTopicEvaluator,
 )
 from hackagent.attacks.evaluator.metrics import (
     calculate_confidence_score,
@@ -53,6 +60,9 @@ from hackagent.attacks.evaluator.metrics import (
     calculate_success_rate,
     generate_summary_report,
     group_by_goal,
+    calculate_majority_vote_asr,
+    calculate_fleiss_kappa,
+    calculate_per_judge_strictness,
 )
 from hackagent.attacks.evaluator.pattern_evaluators import (
     KeywordEvaluator,
@@ -73,6 +83,8 @@ __all__ = [
     "NuancedEvaluator",
     "JailbreakBenchEvaluator",
     "HarmBenchEvaluator",
+    "HarmBenchVariantEvaluator",
+    "OnTopicEvaluator",
     "EVALUATOR_MAP",
     # Pattern evaluators
     "PatternEvaluator",
@@ -84,6 +96,9 @@ __all__ = [
     "calculate_per_goal_metrics",
     "generate_summary_report",
     "group_by_goal",
+    "calculate_majority_vote_asr",
+    "calculate_fleiss_kappa",
+    "calculate_per_judge_strictness",
     # Sync
     "sync_evaluation_to_server",
     "update_single_result",