[agent] Add LLM risk analyzer (#9349)

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: openhands <openhands@all-hands.dev>
Co-authored-by: Graham Neubig <neubig@gmail.com>
Co-authored-by: llamantino <213239228+llamantino@users.noreply.github.com>
Co-authored-by: mamoodi <mamoodiha@gmail.com>
Co-authored-by: Tim O'Farrell <tofarr@gmail.com>
Co-authored-by: Hiep Le <69354317+hieptl@users.noreply.github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Ryan H. Tran <descience.thh10@gmail.com>
Co-authored-by: Neeraj Panwar <49247372+npneeraj@users.noreply.github.com>
Co-authored-by: sp.wack <83104063+amanape@users.noreply.github.com>
Co-authored-by: Insop <1240382+insop@users.noreply.github.com>
Co-authored-by: test <test@test.com>
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
Co-authored-by: Zhonghao Jiang <zhonghao.J@outlook.com>
Co-authored-by: Ray Myers <ray.myers@gmail.com>

openhands/server/session/agent_session.py

@@ -29,7 +29,6 @@ from openhands.runtime import get_runtime_cls
 from openhands.runtime.base import Runtime
 from openhands.runtime.impl.remote.remote_runtime import RemoteRuntime
 from openhands.runtime.runtime_status import RuntimeStatus
-from openhands.security import SecurityAnalyzer, options
 from openhands.server.services.conversation_stats import ConversationStats
 from openhands.storage.data_models.user_secrets import UserSecrets
 from openhands.storage.files import FileStore
@@ -54,7 +53,7 @@ class AgentSession:
     file_store: FileStore
     controller: AgentController | None = None
     runtime: Runtime | None = None
-    security_analyzer: SecurityAnalyzer | None = None
+
     memory: Memory | None = None
     _starting: bool = False
     _started_at: float = 0
@@ -133,7 +132,6 @@ class AgentSession:
             custom_secrets=custom_secrets if custom_secrets else {}  # type: ignore[arg-type]
         )
         try:
-            self._create_security_analyzer(config.security.security_analyzer)
             runtime_connected = await self._create_runtime(
                 runtime_name=runtime_name,
                 config=config,
@@ -245,8 +243,6 @@ class AgentSession:
             await self.controller.close()
         if self.runtime is not None:
             EXECUTOR.submit(self.runtime.close)
-        if self.security_analyzer is not None:
-            await self.security_analyzer.close()
 
     def _run_replay(
         self,
@@ -278,18 +274,6 @@ class AgentSession:
         assert isinstance(replay_events[0], MessageAction)
         return replay_events[0]
 
-    def _create_security_analyzer(self, security_analyzer: str | None) -> None:
-        """Creates a SecurityAnalyzer instance that will be used to analyze the agent actions
-
-        Parameters:
-        - security_analyzer: The name of the security analyzer to use
-        """
-        if security_analyzer:
-            self.logger.debug(f'Using security analyzer: {security_analyzer}')
-            self.security_analyzer = options.SecurityAnalyzers.get(
-                security_analyzer, SecurityAnalyzer
-            )(self.event_stream)
-
     def override_provider_tokens_with_custom_secret(
         self,
         git_provider_tokens: PROVIDER_TOKEN_TYPE | None,
@@ -461,6 +445,7 @@ class AgentSession:
             status_callback=self._status_callback,
             initial_state=initial_state,
             replay_events=replay_events,
+            security_analyzer=self.runtime.security_analyzer if self.runtime else None,
         )
 
         return (controller, initial_state is not None)

openhands/server/session/conversation.py

@@ -5,7 +5,6 @@ from openhands.events.stream import EventStream
 from openhands.llm.llm_registry import LLMRegistry
 from openhands.runtime import get_runtime_cls
 from openhands.runtime.base import Runtime
-from openhands.security import SecurityAnalyzer, options
 from openhands.storage.files import FileStore
 from openhands.utils.async_utils import call_sync_from_async
 
@@ -36,11 +35,6 @@ class ServerConversation:
             event_stream = EventStream(sid, file_store, user_id)
         self.event_stream = event_stream
 
-        if config.security.security_analyzer:
-            self.security_analyzer = options.SecurityAnalyzers.get(
-                config.security.security_analyzer, SecurityAnalyzer
-            )(self.event_stream)
-
         if runtime:
             self._attach_to_existing = True
         else:
@@ -55,6 +49,11 @@ class ServerConversation:
             )
         self.runtime = runtime
 
+    @property
+    def security_analyzer(self):
+        """Access security analyzer through runtime."""
+        return self.runtime.security_analyzer
+
     async def connect(self) -> None:
         if not self._attach_to_existing:
             await self.runtime.connect()

openhands/server/session/session.py

@@ -118,7 +118,9 @@ class Session:
             else settings.confirmation_mode
         )
         self.config.security.security_analyzer = (
-            settings.security_analyzer or self.config.security.security_analyzer
+            self.config.security.security_analyzer
+            if settings.security_analyzer is None
+            else settings.security_analyzer
         )
         self.config.sandbox.base_container_image = (
             settings.sandbox_base_container_image