Merge branch 'main' of github.com:NousResearch/hermes-agent into bb/gui

tests/run_agent/test_860_dedup.py

@@ -170,33 +170,7 @@ class TestFlushDeduplication:
 # ---------------------------------------------------------------------------
 
 class TestAppendToTranscriptSkipDb:
-    """Verify skip_db=True writes JSONL but not SQLite."""
-
-    @pytest.fixture()
-    def store(self, tmp_path):
-        from gateway.config import GatewayConfig
-        from gateway.session import SessionStore
-        config = GatewayConfig()
-        with patch("gateway.session.SessionStore._ensure_loaded"):
-            s = SessionStore(sessions_dir=tmp_path, config=config)
-        s._db = None  # no SQLite for these JSONL-focused tests
-        s._loaded = True
-        return s
-
-    def test_skip_db_writes_jsonl_only(self, store, tmp_path):
-        """With skip_db=True, message appears in JSONL but not SQLite."""
-        session_id = "test-skip-db"
-        msg = {"role": "assistant", "content": "hello world"}
-        store.append_to_transcript(session_id, msg, skip_db=True)
-
-        # JSONL should have the message
-        jsonl_path = store.get_transcript_path(session_id)
-        assert jsonl_path.exists()
-        with open(jsonl_path) as f:
-            lines = f.readlines()
-        assert len(lines) == 1
-        parsed = json.loads(lines[0])
-        assert parsed["content"] == "hello world"
+    """Verify skip_db=True skips the SQLite write."""
 
     def test_skip_db_prevents_sqlite_write(self, tmp_path):
         """With skip_db=True and a real DB, message does NOT appear in SQLite."""
@@ -223,14 +197,8 @@ class TestAppendToTranscriptSkipDb:
         rows = db.get_messages(session_id)
         assert len(rows) == 0, f"Expected 0 DB rows with skip_db=True, got {len(rows)}"
 
-        # But JSONL should have it
-        jsonl_path = store.get_transcript_path(session_id)
-        with open(jsonl_path) as f:
-            lines = f.readlines()
-        assert len(lines) == 1
-
-    def test_default_writes_both(self, tmp_path):
-        """Without skip_db, message appears in both JSONL and SQLite."""
+    def test_default_writes_to_sqlite(self, tmp_path):
+        """Without skip_db, message appears in SQLite."""
         from gateway.config import GatewayConfig
         from gateway.session import SessionStore
         from hermes_state import SessionDB
@@ -250,13 +218,7 @@ class TestAppendToTranscriptSkipDb:
         msg = {"role": "user", "content": "test message"}
         store.append_to_transcript(session_id, msg)
 
-        # JSONL should have the message
-        jsonl_path = store.get_transcript_path(session_id)
-        with open(jsonl_path) as f:
-            lines = f.readlines()
-        assert len(lines) == 1
-
-        # SQLite should also have the message
+        # SQLite should have the message
         rows = db.get_messages(session_id)
         assert len(rows) == 1
 

tests/run_agent/test_background_review_cache_parity.py

@@ -38,6 +38,9 @@ def _make_agent_stub(agent_cls):
     agent._MEMORY_REVIEW_PROMPT = "review memory"
     agent._SKILL_REVIEW_PROMPT = "review skills"
     agent._COMBINED_REVIEW_PROMPT = "review both"
+    # Non-None so the test catches a missing-kwarg regression.
+    agent.enabled_toolsets = ["memory", "skills", "terminal"]
+    agent.disabled_toolsets = ["spotify", "feishu_doc"]
     return agent
 
 
@@ -183,3 +186,54 @@ def test_review_fork_pins_session_start_and_session_id():
         "Review fork did not inherit parent's session_id — "
         "system-prompt rebuild paths would diverge."
     )
+
+
+def test_review_fork_inherits_parent_toolset_config():
+    """``tools[]`` byte-stability: fork must inherit parent's toolset config."""
+    import run_agent
+
+    agent = _make_agent_stub(run_agent.AIAgent)
+
+    captured = {}
+
+    class _Recorder:
+        def __init__(self, *args, **kwargs):
+            captured["enabled_toolsets"] = kwargs.get("enabled_toolsets")
+            captured["disabled_toolsets"] = kwargs.get("disabled_toolsets")
+            self._cached_system_prompt = None
+            self._memory_write_origin = None
+            self._memory_write_context = None
+            self._memory_store = None
+            self._memory_enabled = None
+            self._user_profile_enabled = None
+            self._memory_nudge_interval = None
+            self._skill_nudge_interval = None
+            self.suppress_status_output = None
+            self.session_start = None
+            self.session_id = None
+
+        def run_conversation(self, *args, **kwargs):
+            raise RuntimeError("stop after recording — don't actually call the API")
+
+        def shutdown_memory_provider(self):
+            pass
+
+        def close(self):
+            pass
+
+    with patch.object(run_agent, "AIAgent", _Recorder), \
+         patch("threading.Thread", _SyncThread):
+        agent._spawn_background_review(
+            messages_snapshot=[],
+            review_memory=True,
+            review_skills=False,
+        )
+
+    assert captured.get("enabled_toolsets") == agent.enabled_toolsets, (
+        f"enabled_toolsets mismatch: {captured.get('enabled_toolsets')!r} "
+        f"vs expected {agent.enabled_toolsets!r}"
+    )
+    assert captured.get("disabled_toolsets") == agent.disabled_toolsets, (
+        f"disabled_toolsets mismatch: {captured.get('disabled_toolsets')!r} "
+        f"vs expected {agent.disabled_toolsets!r}"
+    )

tests/run_agent/test_background_review_toolset_restriction.py

@@ -38,6 +38,9 @@ def _make_agent_stub(agent_cls):
     agent._MEMORY_REVIEW_PROMPT = "review memory"
     agent._SKILL_REVIEW_PROMPT = "review skills"
     agent._COMBINED_REVIEW_PROMPT = "review both"
+    # Non-None so the test catches a missing-kwarg regression.
+    agent.enabled_toolsets = ["memory", "skills", "terminal"]
+    agent.disabled_toolsets = ["spotify", "feishu_doc"]
     return agent
 
 
@@ -52,13 +55,8 @@ class _SyncThread:
             self._target()
 
 
-def test_background_review_does_not_narrow_toolset_schema():
-    """The review fork must NOT pass enabled_toolsets to AIAgent.
-
-    Narrowing the schema diverges the ``tools`` cache key from the parent's,
-    which sits above ``system`` in Anthropic's cache hierarchy and forces a
-    full prefix-cache miss on every review (see #25322, PR #17276).
-    """
+def test_background_review_matches_parent_toolset_config():
+    """Fork must receive parent's toolset config so ``tools[]`` cache key matches."""
     import run_agent
 
     agent = _make_agent_stub(run_agent.AIAgent)
@@ -66,6 +64,7 @@ def test_background_review_does_not_narrow_toolset_schema():
 
     def _capture_init(self, *args, **kwargs):
         captured["enabled_toolsets"] = kwargs.get("enabled_toolsets", "UNSET")
+        captured["disabled_toolsets"] = kwargs.get("disabled_toolsets", "UNSET")
         raise RuntimeError("stop after capturing init args")
 
     with patch.object(run_agent.AIAgent, "__init__", _capture_init), \
@@ -77,11 +76,13 @@ def test_background_review_does_not_narrow_toolset_schema():
         )
 
     assert "enabled_toolsets" in captured, "AIAgent.__init__ was not called"
-    # The kwarg must be absent — letting AIAgent inherit the default full
-    # toolset so the schema bytes match the parent's.
-    assert captured["enabled_toolsets"] == "UNSET", (
-        f"Review fork narrowed the toolset schema (got {captured['enabled_toolsets']!r}), "
-        "which breaks prefix-cache parity with the parent."
+    assert captured["enabled_toolsets"] == agent.enabled_toolsets, (
+        f"enabled_toolsets mismatch: {captured['enabled_toolsets']!r} "
+        f"vs expected {agent.enabled_toolsets!r}"
+    )
+    assert captured["disabled_toolsets"] == agent.disabled_toolsets, (
+        f"disabled_toolsets mismatch: {captured['disabled_toolsets']!r} "
+        f"vs expected {agent.disabled_toolsets!r}"
     )
 
 

tests/run_agent/test_codex_xai_oauth_recovery.py

@@ -19,11 +19,15 @@ Three distinct failure modes the user community hit during rollout:
    one-line hint pointing the user at https://grok.com and ``/model``.
 
 3. Multi-turn replay of ``codex_reasoning_items`` (with
-   ``encrypted_content``) is now suppressed for ``is_xai_responses=True``
-   in ``_chat_messages_to_responses_input``.  xAI's OAuth/SuperGrok
-   surface rejects replayed encrypted reasoning items; Grok still
-   reasons natively each turn, so coherence rides on visible message
-   text.
+   ``encrypted_content``) was briefly suppressed for ``is_xai_responses``
+   in PR #26644 on the theory that xAI's OAuth/SuperGrok surface
+   rejected replayed encrypted reasoning items.  That suppression was
+   reverted shortly after: xAI confirmed they explicitly want Hermes to
+   thread encrypted reasoning back across turns, and the original
+   multi-turn failure mode was actually the prelude-SSE issue closed by
+   Fix A above.  The remaining tests here lock in that xAI receives
+   replayed reasoning AND that we ask xAI to echo it back in the
+   ``include`` array.
 """
 
 from types import SimpleNamespace
@@ -353,8 +357,15 @@ def test_codex_reasoning_replay_default_includes_encrypted_content():
     assert reasoning[0]["encrypted_content"] == "enc_blob"
 
 
-def test_codex_reasoning_replay_stripped_for_xai_oauth():
-    """xAI OAuth surface must NOT receive replayed encrypted reasoning."""
+def test_codex_reasoning_replay_includes_encrypted_content_for_xai():
+    """xAI must receive replayed encrypted reasoning items (May 2026 reversal).
+
+    Earlier we stripped these on the theory that the OAuth/SuperGrok
+    surface rejected them.  xAI subsequently confirmed they explicitly
+    want Hermes to thread encrypted reasoning back across turns for
+    cross-turn coherence — that's the whole point of the partnership
+    integration.
+    """
     from agent.codex_responses_adapter import _chat_messages_to_responses_input
 
     msgs = [
@@ -365,10 +376,13 @@ def test_codex_reasoning_replay_stripped_for_xai_oauth():
 
     items = _chat_messages_to_responses_input(msgs, is_xai_responses=True)
     reasoning = [it for it in items if it.get("type") == "reasoning"]
-    assert reasoning == []
+    assert len(reasoning) == 1, (
+        "xAI must receive replayed reasoning items — see docstring for the "
+        "May 2026 reversal of the earlier suppression gate."
+    )
+    assert reasoning[0]["encrypted_content"] == "enc_blob"
 
-    # The assistant's visible text must still survive — coherence across
-    # turns rides on the message text alone.
+    # And the assistant's visible text must still be present alongside it.
     assistant_items = [
         it for it in items
         if it.get("role") == "assistant" or it.get("type") == "message"
@@ -376,8 +390,12 @@ def test_codex_reasoning_replay_stripped_for_xai_oauth():
     assert assistant_items, "assistant message must still be present"
 
 
-def test_codex_transport_xai_request_omits_encrypted_content_include():
-    """Verify the xAI ``include`` array no longer requests encrypted reasoning."""
+def test_codex_transport_xai_request_includes_encrypted_content():
+    """xAI ``include`` array must request ``reasoning.encrypted_content``.
+
+    This is the request-side half of the May 2026 reversal: we ask xAI
+    to echo back encrypted reasoning so the next turn can replay it.
+    """
     from agent.transports.codex import ResponsesApiTransport
 
     transport = ResponsesApiTransport()
@@ -392,14 +410,11 @@ def test_codex_transport_xai_request_omits_encrypted_content_include():
         reasoning_config={"enabled": True, "effort": "medium"},
         is_xai_responses=True,
     )
-    # Without this gate, xAI would echo back encrypted_content blobs we'd
-    # then store in codex_reasoning_items and replay next turn — which is
-    # exactly the multi-turn failure mode we're closing.
-    assert kwargs["include"] == []
+    assert kwargs["include"] == ["reasoning.encrypted_content"]
 
 
-def test_codex_transport_xai_strips_replayed_reasoning_in_input():
-    """End-to-end: build_kwargs on xai-oauth must strip prior reasoning."""
+def test_codex_transport_xai_replays_reasoning_in_input():
+    """End-to-end: build_kwargs on xAI must replay prior encrypted reasoning."""
     from agent.transports.codex import ResponsesApiTransport
 
     transport = ResponsesApiTransport()
@@ -418,7 +433,8 @@ def test_codex_transport_xai_strips_replayed_reasoning_in_input():
     )
     input_items = kwargs["input"]
     reasoning_items = [it for it in input_items if it.get("type") == "reasoning"]
-    assert reasoning_items == []
+    assert len(reasoning_items) == 1
+    assert reasoning_items[0]["encrypted_content"] == "enc_blob"
 
 
 def test_codex_transport_native_codex_still_replays_reasoning_in_input():

tests/run_agent/test_create_openai_client_reuse.py

@@ -16,6 +16,7 @@ with ``APIConnectionError('Connection error.')`` whose cause was
 That is the exact scenario this test reproduces at object level without a
 network, so it runs in CI on every PR.
 """
+from types import SimpleNamespace
 from unittest.mock import MagicMock, patch
 
 from run_agent import AIAgent
@@ -186,3 +187,32 @@ def test_replace_primary_openai_client_survives_repeated_rebuilds():
         "Some _create_openai_client calls returned the same object across "
         "a teardown — rebuild is not producing fresh clients"
     )
+
+
+def test_force_close_tcp_sockets_descends_httpcore_1_connection_wrapper():
+    """httpcore 1.x stores the real stream below conn._connection."""
+    from agent.agent_runtime_helpers import force_close_tcp_sockets
+
+    class FakeSocket:
+        def __init__(self):
+            self.shutdown_calls = 0
+            self.close_calls = 0
+
+        def shutdown(self, _how):
+            self.shutdown_calls += 1
+
+        def close(self):
+            self.close_calls += 1
+
+    sock = FakeSocket()
+    stream = SimpleNamespace(_sock=sock)
+    http11 = SimpleNamespace(_network_stream=stream)
+    pool_entry = SimpleNamespace(_connection=http11)
+    pool = SimpleNamespace(_connections=[pool_entry])
+    transport = SimpleNamespace(_pool=pool)
+    http_client = SimpleNamespace(_transport=transport)
+    openai_client = SimpleNamespace(_client=http_client)
+
+    assert force_close_tcp_sockets(openai_client) == 1
+    assert sock.shutdown_calls == 1
+    assert sock.close_calls == 1

tests/run_agent/test_openai_client_lifecycle.py

@@ -1,5 +1,6 @@
 import sys
 import threading
+import time
 import types
 from types import SimpleNamespace
 
@@ -64,6 +65,7 @@ def _build_agent(shared_client=None):
     agent.stream_delta_callback = None
     agent._stream_callback = None
     agent.reasoning_callback = None
+    agent.status_callback = None
     return agent
 
 
@@ -93,6 +95,24 @@ def test_retry_after_api_connection_error_recreates_request_client(monkeypatch):
     assert second_request.close_calls >= 1
 
 
+def test_stale_non_stream_close_is_single_owner(monkeypatch):
+    def slow_responder(**kwargs):
+        time.sleep(0.1)
+        raise _connection_error()
+
+    request_client = FakeRequestClient(slow_responder)
+    factory = OpenAIFactory([request_client])
+    monkeypatch.setattr(run_agent, "OpenAI", factory)
+
+    agent = _build_agent()
+    agent._compute_non_stream_stale_timeout = lambda _messages: 0.01
+
+    with pytest.raises(APIConnectionError):
+        agent._interruptible_api_call({"model": agent.model, "messages": []})
+
+    assert request_client.close_calls == 1
+
+
 def test_closed_shared_client_is_recreated_before_request(monkeypatch):
     stale_shared = FakeSharedClient(lambda **kwargs: (_ for _ in ()).throw(AssertionError("stale shared client used")))
     stale_shared._client.is_closed = True

tests/run_agent/test_vision_aware_preprocessing.py

@@ -168,3 +168,43 @@ class TestModelSupportsVision:
         agent = _make_agent()
         with patch("agent.models_dev.get_model_capabilities", side_effect=RuntimeError("boom")):
             assert agent._model_supports_vision() is False
+
+    def test_top_level_model_override_wins(self):
+        agent = _make_agent()
+        agent.provider = "custom"
+        agent.model = "my-llava"
+        with patch("hermes_cli.config.load_config", return_value={"model": {"supports_vision": True}}), \
+             patch("agent.models_dev.get_model_capabilities", return_value=None):
+            assert agent._model_supports_vision() is True
+
+    def test_per_provider_per_model_override_wins(self):
+        agent = _make_agent()
+        agent.provider = "custom"
+        agent.model = "my-llava"
+        cfg = {"providers": {"custom": {"models": {"my-llava": {"supports_vision": True}}}}}
+        with patch("hermes_cli.config.load_config", return_value=cfg), \
+             patch("agent.models_dev.get_model_capabilities", return_value=None):
+            assert agent._model_supports_vision() is True
+
+    def test_named_custom_provider_resolved_via_config_provider(self):
+        # Named custom providers get runtime self.provider rewritten to
+        # "custom" while the config keeps the original name under
+        # model.provider. The override must still resolve.
+        agent = _make_agent()
+        agent.provider = "custom"
+        agent.model = "my-llava"
+        cfg = {
+            "model": {"provider": "my-vllm", "default": "my-llava"},
+            "providers": {"my-vllm": {"models": {"my-llava": {"supports_vision": True}}}},
+        }
+        with patch("hermes_cli.config.load_config", return_value=cfg), \
+             patch("agent.models_dev.get_model_capabilities", return_value=None):
+            assert agent._model_supports_vision() is True
+
+    def test_override_false_disables_vision_for_models_dev_models(self):
+        agent = _make_agent()
+        fake_caps = MagicMock()
+        fake_caps.supports_vision = True
+        with patch("hermes_cli.config.load_config", return_value={"model": {"supports_vision": False}}), \
+             patch("agent.models_dev.get_model_capabilities", return_value=fake_caps):
+            assert agent._model_supports_vision() is False