[pull] main from OpenHands:main

openhands-sdk/openhands/sdk/conversation/impl/local_conversation.py

@@ -55,6 +55,7 @@
 from openhands.sdk.io import FileStore, LocalFileStore
 from openhands.sdk.llm import LLM, Message, TextContent, content_to_str
 from openhands.sdk.llm.auth.openai import create_subscription_llm_from_config
+from openhands.sdk.llm.exceptions import LLMAuthenticationError
 from openhands.sdk.llm.llm import LLMCallContext
 from openhands.sdk.llm.llm_profile_store import LLMProfileStore
 from openhands.sdk.llm.llm_registry import LLMRegistry
@@ -2040,6 +2041,21 @@ def run(self) -> None:
                             )
                         )
                         break
+        except LLMAuthenticationError as e:
+            with self._state:
+                self._state.execution_status = ConversationExecutionStatus.ERROR
+                self._on_event(
+                    ConversationErrorEvent(
+                        source="environment",
+                        code="LLMAuthenticationError",
+                        detail=(
+                            "Your LLM API key appears to be invalid or has expired."
+                        ),
+                    )
+                )
+            raise ConversationRunError(
+                self._state.id, e, persistence_dir=self._state.persistence_dir
+            ) from e
         except Exception as e:
             with self._state:
                 self._state.execution_status = ConversationExecutionStatus.ERROR
@@ -2540,6 +2556,21 @@ async def arun(self) -> None:
 
                 self._state.execution_status = ConversationExecutionStatus.PAUSED
                 self._on_event(InterruptEvent())
+        except LLMAuthenticationError as e:
+            with self._state:
+                self._state.execution_status = ConversationExecutionStatus.ERROR
+                self._on_event(
+                    ConversationErrorEvent(
+                        source="environment",
+                        code="LLMAuthenticationError",
+                        detail=(
+                            "Your LLM API key appears to be invalid or has expired."
+                        ),
+                    )
+                )
+            raise ConversationRunError(
+                self._state.id, e, persistence_dir=self._state.persistence_dir
+            ) from e
         except Exception as e:
             with self._state:
                 updated_agent_state = dict(self._state.agent_state)

openhands-sdk/openhands/sdk/hooks/executor.py

@@ -566,9 +566,10 @@ def execute(
                 try:
                     output_data = json.loads(result.stdout)
                     if isinstance(output_data, dict):
-                        # Parse decision
-                        if "decision" in output_data:
-                            decision_str = output_data["decision"].lower()
+                        # Non-string values represent no decision.
+                        decision_value = output_data.get("decision")
+                        if isinstance(decision_value, str):
+                            decision_str = decision_value.lower()
                             if decision_str == "allow":
                                 hook_result.decision = HookDecision.ALLOW
                             elif decision_str == "deny":

openhands-sdk/openhands/sdk/llm/utils/AGENTS.md

@@ -6,7 +6,7 @@ See the [SDK AGENTS.md](../../AGENTS.md) for package-wide policies.
 
 These lists are a curated set of models that work well, not a catalog of everything a provider offers. Keep them short.
 
-- For each model line, keep only the **two latest versions** (for example `gpt-6` and `gpt-5.6`; `claude-opus-5` and `claude-opus-4-8`).
+- For each model line, keep only the **two latest versions** (for example `gpt-6` and `gpt-5.6`; `claude-opus-5-5` and `claude-opus-5`).
 - Variants of a kept version (`-pro`, `-mini`, `-codex`, `-flash`, dated aliases) stay with that version. Unversioned "current" aliases (`deepseek-chat`, `kimi-for-coding`) stay.
 - When you add a new version, remove the oldest version in the same line, in every list where it appears (the provider list and `VERIFIED_OPENHANDS_MODELS`).
 - Every entry in `VERIFIED_OPENHANDS_MODELS` must also appear in a provider list, unless it is OpenHands-only; `tests/sdk/llm/test_model_list.py` checks this.

openhands-sdk/openhands/sdk/llm/utils/model_features.py

@@ -109,7 +109,6 @@ def _normalized_supported_openai_params(model: str | None) -> frozenset[str]:
 
 
 REASONING_EFFORT_MODEL_OVERRIDES = {
-    "gpt-5.2-codex": "gpt-5.2-codex",
     "kimi-k3": "moonshot/kimi-k3",
 }
 

openhands-sdk/openhands/sdk/llm/utils/verified_models.py

@@ -11,6 +11,8 @@
 
 # GPT: gpt-6 and gpt-5.6. Codex: gpt-5.3-codex and gpt-5.2-codex.
 VERIFIED_OPENAI_MODELS = [
+    "gpt-6-sol",
+    "gpt-6-luna",
     "gpt-6-astra",
     "gpt-5.6",
     "gpt-5.6-sol",
@@ -20,10 +22,10 @@
     "gpt-5.2-codex",
 ]
 
-# Opus: 5 and 4.8. Sonnet: 5 and 4.6. Haiku: 4.5. Fable: 5.1 and 5.
+# Opus: 5.5 and 5. Sonnet: 5 and 4.6. Haiku: 4.5. Fable: 5.1 and 5.
 VERIFIED_ANTHROPIC_MODELS = [
+    "claude-opus-5-5",
     "claude-opus-5",
-    "claude-opus-4-8",
     "claude-sonnet-5",
     "claude-sonnet-4-6",
     "claude-haiku-4-5-20251001",
@@ -104,12 +106,14 @@
 # What the ``openhands/`` provider serves. Same rule; every entry must also be in
 # a provider list above, except OpenHands-only models.
 VERIFIED_OPENHANDS_MODELS = [
+    "claude-opus-5-5",
     "claude-opus-5",
-    "claude-opus-4-8",
     "claude-sonnet-5",
     "claude-sonnet-4-6",
     "claude-fable-5-1",
     "claude-fable-5",
+    "gpt-6-sol",
+    "gpt-6-luna",
     "gpt-6-astra",
     "gpt-5.6",
     "gpt-5.3-codex",

tests/sdk/conversation/local/test_auth_error_friendly_message.py

@@ -0,0 +1,118 @@
+"""Tests that LLMAuthenticationError surfaces a user-friendly message.
+
+Regression tests for:
+  https://github.com/OpenHands/software-agent-sdk/issues/3411
+
+When a user has an invalid/expired API key the raw litellm error (e.g. the
+full AnthropicException JSON) must NOT appear in the ConversationErrorEvent
+detail that is sent to the UI.  Instead, a clear, actionable message should
+be emitted, while the ConversationRunError is still raised so server logs
+remain unaffected.
+"""
+
+import asyncio
+import tempfile
+
+import pytest
+
+from openhands.sdk.agent import Agent
+from openhands.sdk.conversation import Conversation, LocalConversation
+from openhands.sdk.conversation.exceptions import ConversationRunError
+from openhands.sdk.event.conversation_error import ConversationErrorEvent
+from openhands.sdk.llm import Message, TextContent
+from openhands.sdk.llm.exceptions import LLMAuthenticationError
+from openhands.sdk.testing import TestLLM
+
+
+_RAW_LITELLM_ERROR = (
+    "litellm.AuthenticationError: AnthropicException - "
+    '{"type":"error","error":{"type":"authentication_error",'
+    '"message":"invalid x-api-key"},'
+    '"request_id":"req_011CbTfF4jtKVAB95FSH6ESb"}'
+)
+_FRIENDLY_SUBSTRING = "invalid or has expired"
+
+
+def _make_auth_failing_conversation(tmpdir: str) -> LocalConversation:
+    llm = TestLLM.from_messages([LLMAuthenticationError(_RAW_LITELLM_ERROR)])
+    agent = Agent(llm=llm, tools=[])
+    conv = Conversation(agent=agent, persistence_dir=tmpdir, workspace=tmpdir)
+    assert isinstance(conv, LocalConversation)
+    conv.send_message(Message(role="user", content=[TextContent(text="hello")]))
+    return conv
+
+
+# ---------------------------------------------------------------------------
+# Sync path (run)
+# ---------------------------------------------------------------------------
+
+
+def test_auth_error_run_raises_conversation_run_error():
+    """ConversationRunError is still raised so server logs are unaffected."""
+    with tempfile.TemporaryDirectory() as tmpdir:
+        conv = _make_auth_failing_conversation(tmpdir)
+        with pytest.raises(ConversationRunError) as exc_info:
+            conv.run()
+        assert isinstance(exc_info.value.__cause__, LLMAuthenticationError)
+
+
+def test_auth_error_run_emits_friendly_detail():
+    """ConversationErrorEvent.detail is user-readable, not the raw litellm string."""
+    with tempfile.TemporaryDirectory() as tmpdir:
+        conv = _make_auth_failing_conversation(tmpdir)
+        with pytest.raises(ConversationRunError):
+            conv.run()
+
+        error_events = [
+            e for e in conv.state.events if isinstance(e, ConversationErrorEvent)
+        ]
+        assert error_events, "Expected at least one ConversationErrorEvent"
+
+        auth_error_event = next(
+            (e for e in error_events if e.code == "LLMAuthenticationError"), None
+        )
+        assert auth_error_event is not None, (
+            "Expected a ConversationErrorEvent with code='LLMAuthenticationError'"
+        )
+        assert _FRIENDLY_SUBSTRING in auth_error_event.detail, (
+            f"Expected friendly message in detail, got: {auth_error_event.detail!r}"
+        )
+        assert _RAW_LITELLM_ERROR not in auth_error_event.detail, (
+            "Raw litellm error string must not appear in the UI-facing detail"
+        )
+
+
+# ---------------------------------------------------------------------------
+# Async path (arun)
+# ---------------------------------------------------------------------------
+
+
+def test_auth_error_arun_raises_conversation_run_error():
+    """Async path: ConversationRunError is still raised."""
+    with tempfile.TemporaryDirectory() as tmpdir:
+        conv = _make_auth_failing_conversation(tmpdir)
+        with pytest.raises(ConversationRunError) as exc_info:
+            asyncio.run(conv.arun())
+        assert isinstance(exc_info.value.__cause__, LLMAuthenticationError)
+
+
+def test_auth_error_arun_emits_friendly_detail():
+    """Async path: ConversationErrorEvent.detail is user-readable."""
+    with tempfile.TemporaryDirectory() as tmpdir:
+        conv = _make_auth_failing_conversation(tmpdir)
+        with pytest.raises(ConversationRunError):
+            asyncio.run(conv.arun())
+
+        error_events = [
+            e for e in conv.state.events if isinstance(e, ConversationErrorEvent)
+        ]
+        assert error_events, "Expected at least one ConversationErrorEvent"
+
+        auth_error_event = next(
+            (e for e in error_events if e.code == "LLMAuthenticationError"), None
+        )
+        assert auth_error_event is not None, (
+            "Expected a ConversationErrorEvent with code='LLMAuthenticationError'"
+        )
+        assert _FRIENDLY_SUBSTRING in auth_error_event.detail
+        assert _RAW_LITELLM_ERROR not in auth_error_event.detail

tests/sdk/hooks/test_executor.py

@@ -56,6 +56,36 @@ def test_execute_receives_json_stdin(self, executor, sample_event, tmp_path):
         assert output_data["event_type"] == "PreToolUse"
         assert output_data["tool_name"] == "BashTool"
 
+    def test_null_decision_leaves_a_successful_hook_successful(
+        self, executor, sample_event
+    ):
+        hook = HookDefinition(command="""echo '{"decision": null}'""")
+
+        result = executor.execute(hook, sample_event)
+
+        assert result.success
+        assert result.exit_code == 0
+        assert result.decision is None
+        assert not result.error
+
+    def test_non_string_decision_is_ignored_rather_than_fatal(
+        self, executor, sample_event
+    ):
+        """Any non-string decision means no decision, not a failed hook."""
+        hook = HookDefinition(command="""echo '{"decision": 1}'""")
+
+        result = executor.execute(hook, sample_event)
+
+        assert result.success
+        assert result.decision is None
+
+    def test_string_decisions_still_parse(self, executor, sample_event):
+        hook = HookDefinition(command="""echo '{"decision": "allow"}'""")
+
+        result = executor.execute(hook, sample_event)
+
+        assert result.decision == HookDecision.ALLOW
+
     def test_execute_blocking_exit_code(self, executor, sample_event):
         """Test that exit code 2 blocks the operation."""
         hook = HookDefinition(command=python_command("import sys; sys.exit(2)"))

tests/sdk/llm/test_llm_profile_store.py

@@ -154,15 +154,15 @@ def test_load_migrates_legacy_openhands_proxy_profile(
         json.dumps(
             {
                 "schema_version": 1,
-                "model": "litellm_proxy/claude-opus-4-8",
+                "model": "litellm_proxy/claude-opus-5",
                 "base_url": "https://llm-proxy.app.all-hands.dev/",
             }
         )
     )
 
     loaded = profile_store.load("legacy")
 
-    assert loaded.model == "openhands/claude-opus-4-8"
+    assert loaded.model == "openhands/claude-opus-5"
     assert loaded.base_url is None
 
 
@@ -174,7 +174,7 @@ def test_list_summaries_migrates_legacy_openhands_proxy_profile(
         json.dumps(
             {
                 "schema_version": 1,
-                "model": "litellm_proxy/claude-opus-4-8",
+                "model": "litellm_proxy/claude-opus-5",
                 "base_url": "https://llm-proxy.app.all-hands.dev/",
             }
         )
@@ -185,7 +185,7 @@ def test_list_summaries_migrates_legacy_openhands_proxy_profile(
     assert summaries == [
         {
             "name": "legacy",
-            "model": "openhands/claude-opus-4-8",
+            "model": "openhands/claude-opus-5",
             "base_url": None,
             "provider_connection_id": None,
             "provider_connection_broken": False,

tests/sdk/llm/test_model_list.py

@@ -157,7 +157,10 @@ def test_verified_lists_keep_two_latest_versions_per_line():
     """
     expectations = {
         "openai": ({"gpt-6-astra", "gpt-5.6"}, {"gpt-5.5", "gpt-5.4", "gpt-4o", "o3"}),
-        "anthropic": ({"claude-opus-5", "claude-opus-4-8"}, {"claude-opus-4-7"}),
+        "anthropic": (
+            {"claude-opus-5-5", "claude-opus-5"},
+            {"claude-opus-4-8", "claude-opus-4-7"},
+        ),
         "mistral": (
             {"devstral-2512", "devstral-medium-2512"},
             {"devstral-medium-2507"},
@@ -175,8 +178,10 @@ def test_verified_lists_keep_two_latest_versions_per_line():
         models = set(VERIFIED_MODELS[provider])
         assert present <= models, f"{provider}: missing {present - models}"
         assert not (absent & models), f"{provider}: stale {absent & models}"
-    assert {"gpt-6-astra", "gpt-5.6", "claude-opus-5"} <= set(VERIFIED_OPENHANDS_MODELS)
-    assert not {"gpt-5.5", "claude-opus-4-7", "minimax-m2.5"} & set(
+    assert {"gpt-6-astra", "gpt-5.6", "claude-opus-5-5", "claude-opus-5"} <= set(
+        VERIFIED_OPENHANDS_MODELS
+    )
+    assert not {"gpt-5.5", "claude-opus-4-8", "claude-opus-4-7", "minimax-m2.5"} & set(
         VERIFIED_OPENHANDS_MODELS
     )
 

tests/sdk/test_settings.py

@@ -523,15 +523,15 @@ def test_validate_agent_settings_migrates_legacy_openhands_proxy_llm() -> None:
             "schema_version": 3,
             "agent_kind": "openhands",
             "llm": {
-                "model": "litellm_proxy/claude-opus-4-8",
+                "model": "litellm_proxy/claude-opus-5",
                 "base_url": "https://llm-proxy.app.all-hands.dev/",
             },
         }
     )
 
     assert isinstance(settings, OpenHandsAgentSettings)
     assert settings.schema_version == AGENT_SETTINGS_SCHEMA_VERSION
-    assert settings.llm.model == "openhands/claude-opus-4-8"
+    assert settings.llm.model == "openhands/claude-opus-5"
     assert settings.llm.base_url is None