Merge origin/main into fix/discord-allow-channel-threads

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_auto_compact.py

@@ -2,8 +2,8 @@
 
 import asyncio
 from datetime import datetime, timedelta
-from unittest.mock import AsyncMock, MagicMock
 from pathlib import Path
+from unittest.mock import AsyncMock, MagicMock
 
 import pytest
 
@@ -15,7 +15,10 @@ from nanobot.command import CommandContext
 from nanobot.providers.base import LLMResponse
 
 
-def _make_loop(tmp_path: Path, session_ttl_minutes: int = 15) -> AgentLoop:
+def _make_loop(
+    tmp_path: Path,
+    session_ttl_minutes: int = 15,
+) -> AgentLoop:
     """Create a minimal AgentLoop for testing."""
     bus = MessageBus()
     provider = MagicMock()
@@ -72,6 +75,12 @@ class TestSessionTTLConfig:
         assert data["idleCompactAfterMinutes"] == 30
         assert "sessionTtlMinutes" not in data
 
+    def test_session_history_and_file_cap_are_internal_constants(self):
+        """Session history/file cap should be internal constants, not config fields."""
+        from nanobot.session.manager import HISTORY_MAX_MESSAGES, FILE_MAX_MESSAGES
+        assert HISTORY_MAX_MESSAGES == 120
+        assert FILE_MAX_MESSAGES == 2000
+
 
 class TestAgentLoopTTLParam:
     """Test that AutoCompact receives and stores session_ttl_minutes."""
@@ -86,6 +95,75 @@ class TestAgentLoopTTLParam:
         loop = _make_loop(tmp_path, session_ttl_minutes=0)
         assert loop.auto_compact._ttl == 0
 
+    @pytest.mark.asyncio
+    async def test_process_message_reads_history_with_token_budget(self, tmp_path):
+        """_process_message should pass an auto-derived token budget to get_history."""
+        loop = _make_loop(tmp_path)
+        session = loop.sessions.get_or_create("cli:direct")
+        session.get_history = MagicMock(return_value=[])
+        loop.context.build_messages = MagicMock(return_value=[])
+        loop._run_agent_loop = AsyncMock(return_value=("ok", [], [], "stop", False))
+        loop._save_turn = MagicMock()
+
+        msg = InboundMessage(
+            channel="cli",
+            sender_id="u1",
+            chat_id="direct",
+            content="hello",
+        )
+        await loop._process_message(msg)
+        session.get_history.assert_called_once()
+        kwargs = session.get_history.call_args.kwargs
+        assert isinstance(kwargs.get("max_tokens"), int)
+        assert kwargs["max_tokens"] > 0
+        assert kwargs["include_timestamps"] is True
+
+    @pytest.mark.asyncio
+    async def test_session_file_cap_archives_and_trims_old_messages(self, tmp_path):
+        loop = _make_loop(tmp_path)
+        loop.context.memory.raw_archive = MagicMock()
+
+        for i in range(4):
+            msg = InboundMessage(
+                channel="cli",
+                sender_id="u1",
+                chat_id="direct",
+                content=f"hello {i}",
+            )
+            await loop._process_message(msg)
+
+        session = loop.sessions.get_or_create("cli:direct")
+        from nanobot.session.manager import FILE_MAX_MESSAGES
+        assert len(session.messages) <= FILE_MAX_MESSAGES
+
+    def test_session_enforce_file_cap_skips_archive_when_dropped_prefix_already_consolidated(self, tmp_path):
+        from nanobot.session.manager import Session
+        archive_fn = MagicMock()
+        session = Session(key="cli:direct")
+        for i in range(8):
+            session.add_message("user", f"u{i}")
+        session.last_consolidated = 6
+
+        session.enforce_file_cap(on_archive=archive_fn, limit=4)
+
+        assert len(session.messages) <= 4
+        archive_fn.assert_not_called()
+
+    def test_session_enforce_file_cap_archives_only_unconsolidated_dropped_prefix(self, tmp_path):
+        from nanobot.session.manager import Session
+        archive_fn = MagicMock()
+        session = Session(key="cli:direct")
+        for i in range(8):
+            session.add_message("user", f"u{i}")
+        session.last_consolidated = 2
+
+        session.enforce_file_cap(on_archive=archive_fn, limit=4)
+
+        assert len(session.messages) <= 4
+        archive_fn.assert_called_once()
+        archived = archive_fn.call_args.args[0]
+        assert [m["content"] for m in archived] == ["u2", "u3"]
+
 
 class TestAutoCompact:
     """Test the _archive method."""

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_consolidator.py

@@ -4,7 +4,12 @@ import pytest
 import asyncio
 from unittest.mock import AsyncMock, MagicMock, patch
 
-from nanobot.agent.memory import Consolidator, MemoryStore
+from nanobot.agent.memory import (
+    Consolidator,
+    MemoryStore,
+    _ARCHIVE_SUMMARY_MAX_CHARS,
+    _RAW_ARCHIVE_MAX_CHARS,
+)
 
 
 @pytest.fixture
@@ -117,8 +122,8 @@ class TestConsolidatorTokenBudget:
         await consolidator.maybe_consolidate_by_tokens(session)
         consolidator.archive.assert_not_called()
 
-    async def test_chunk_cap_preserves_user_turn_boundary(self, consolidator):
-        """Chunk cap should rewind to the last user boundary within the cap."""
+    async def test_large_chunk_archived_without_cap(self, consolidator):
+        """Without chunk cap, the full range from pick_consolidation_boundary is archived."""
         consolidator._SAFETY_BUFFER = 0
         session = MagicMock()
         session.last_consolidated = 0
@@ -133,19 +138,69 @@ class TestConsolidatorTokenBudget:
         consolidator.estimate_session_prompt_tokens = MagicMock(
             side_effect=[(1200, "tiktoken"), (400, "tiktoken")]
         )
-        consolidator.pick_consolidation_boundary = MagicMock(return_value=(61, 999))
+        # Use real pick_consolidation_boundary — it will find boundary at idx=50
+        # (user message at 50, token budget met)
         consolidator.archive = AsyncMock(return_value=True)
 
         await consolidator.maybe_consolidate_by_tokens(session)
 
         archived_chunk = consolidator.archive.await_args.args[0]
-        assert len(archived_chunk) == 50
+        # pick_consolidation_boundary returns (50, tokens) — user turn at idx 50
         assert archived_chunk[0]["content"] == "m0"
-        assert archived_chunk[-1]["content"] == "m49"
+        assert session.last_consolidated > 0
+
+    async def test_raw_archive_fallback_advances_last_consolidated(self, consolidator):
+        """When archive() falls back to raw-archive (LLM failed), the cursor
+        must still advance. Otherwise the same chunk gets raw-archived again
+        on every subsequent maybe_consolidate_by_tokens() call, spamming
+        duplicate [RAW] entries into history.jsonl."""
+        consolidator._SAFETY_BUFFER = 0
+        session = MagicMock()
+        session.last_consolidated = 0
+        session.key = "test:key"
+        session.messages = [
+            {"role": "user" if i in {0, 50} else "assistant", "content": f"m{i}"}
+            for i in range(70)
+        ]
+        session.metadata = {}
+        consolidator.estimate_session_prompt_tokens = MagicMock(
+            side_effect=[(1200, "tiktoken"), (400, "tiktoken")]
+        )
+        # LLM consolidation fails — archive() returns None (raw_archive fired).
+        consolidator.archive = AsyncMock(return_value=None)
+
+        await consolidator.maybe_consolidate_by_tokens(session)
+
+        consolidator.archive.assert_awaited_once()
+        # The chunk is considered "materialized" (as a raw-archive breadcrumb),
+        # so last_consolidated must have moved past it.
         assert session.last_consolidated == 50
 
-    async def test_chunk_cap_skips_when_no_user_boundary_within_cap(self, consolidator):
-        """If the cap would cut mid-turn, consolidation should skip that round."""
+    async def test_raw_archive_fallback_breaks_round_loop(self, consolidator):
+        """A degraded LLM should not trigger more archive() calls within the
+        same maybe_consolidate_by_tokens invocation — bail after one fallback."""
+        consolidator._SAFETY_BUFFER = 0
+        session = MagicMock()
+        session.last_consolidated = 0
+        session.key = "test:key"
+        session.messages = [
+            {"role": "user" if i in {0, 20, 40, 60} else "assistant", "content": f"m{i}"}
+            for i in range(70)
+        ]
+        session.metadata = {}
+        # Keep estimates high so the loop would otherwise run multiple rounds.
+        consolidator.estimate_session_prompt_tokens = MagicMock(
+            return_value=(1200, "tiktoken")
+        )
+        consolidator.archive = AsyncMock(return_value=None)
+
+        await consolidator.maybe_consolidate_by_tokens(session)
+
+        # Exactly one fallback per call — not _MAX_CONSOLIDATION_ROUNDS.
+        assert consolidator.archive.await_count == 1
+
+    async def test_boundary_respected_when_no_intermediate_user_turn(self, consolidator):
+        """When boundary points past a long tool chain, the full chunk is archived."""
         consolidator._SAFETY_BUFFER = 0
         session = MagicMock()
         session.last_consolidated = 0
@@ -157,11 +212,106 @@ class TestConsolidatorTokenBudget:
             }
             for i in range(70)
         ]
-        consolidator.estimate_session_prompt_tokens = MagicMock(return_value=(1200, "tiktoken"))
-        consolidator.pick_consolidation_boundary = MagicMock(return_value=(61, 999))
+        consolidator.estimate_session_prompt_tokens = MagicMock(
+            side_effect=[(1200, "tiktoken"), (400, "tiktoken")]
+        )
         consolidator.archive = AsyncMock(return_value=True)
 
         await consolidator.maybe_consolidate_by_tokens(session)
 
-        consolidator.archive.assert_not_awaited()
-        assert session.last_consolidated == 0
+        consolidator.archive.assert_awaited_once()
+        # pick_consolidation_boundary finds the only boundary at idx=61
+        assert session.last_consolidated == 61
+
+
+class TestRawArchiveTruncation:
+    """raw_archive() must cap entry size to avoid bloating history.jsonl."""
+
+    def test_raw_archive_truncates_large_content(self, store):
+        """Large messages should be truncated to _RAW_ARCHIVE_MAX_CHARS."""
+        big = "x" * 50_000
+        messages = [{"role": "user", "content": big}]
+        store.raw_archive(messages)
+        entries = store.read_unprocessed_history(since_cursor=0)
+        assert len(entries) == 1
+        assert len(entries[0]["content"]) < 50_000
+        assert "[RAW]" in entries[0]["content"]
+
+    def test_raw_archive_preserves_small_content(self, store):
+        """Small messages should not be truncated."""
+        messages = [{"role": "user", "content": "hello"}]
+        store.raw_archive(messages)
+        entries = store.read_unprocessed_history(since_cursor=0)
+        assert len(entries) == 1
+        assert "hello" in entries[0]["content"]
+
+    def test_raw_archive_custom_max_chars(self, store):
+        """max_chars parameter should override default limit."""
+        messages = [{"role": "user", "content": "a" * 200}]
+        store.raw_archive(messages, max_chars=100)
+        entries = store.read_unprocessed_history(since_cursor=0)
+        assert len(entries[0]["content"]) < 200
+
+
+class TestArchiveTruncation:
+    """archive() must truncate formatted text before sending to consolidation LLM."""
+
+    async def test_archive_truncates_large_formatted_text(self, consolidator, mock_provider, store):
+        """Large formatted text should be truncated to token budget before LLM call."""
+        # context_window_tokens=1000, max_completion_tokens=100, _SAFETY_BUFFER=1024
+        # budget = 1000 - 100 - 1024 = -124 → fallback via truncate_text(budget*4)
+        big_messages = [{"role": "user", "content": "x" * 100_000}]
+        mock_provider.chat_with_retry.return_value = MagicMock(
+            content="Summary of large input.", finish_reason="stop"
+        )
+        await consolidator.archive(big_messages)
+
+        call_args = mock_provider.chat_with_retry.call_args
+        user_content = call_args.kwargs["messages"][1]["content"]
+        # Should be significantly shorter than 100K
+        assert len(user_content) < 50_000
+
+    async def test_archive_truncates_with_small_token_budget(self, consolidator, mock_provider, store):
+        """Small context window: truncation uses actual tokenizer count."""
+        consolidator.context_window_tokens = 500
+        big_messages = [{"role": "user", "content": "word " * 50_000}]
+        mock_provider.chat_with_retry.return_value = MagicMock(
+            content="Summary.", finish_reason="stop"
+        )
+        await consolidator.archive(big_messages)
+
+        sent_messages = mock_provider.chat_with_retry.call_args.kwargs["messages"]
+        user_content = sent_messages[1]["content"]
+        # budget = 500 - 100 - 1024 = negative, fallback char-based
+        # Should be truncated
+        assert len(user_content) < 250_000
+
+    async def test_oversized_summary_is_capped_before_append(self, consolidator, mock_provider, store):
+        """A pathologically large LLM summary must not land full-length in
+        history.jsonl — that would re-open the #3412 bloat vector from the
+        *success* path instead of the fallback path."""
+        mock_provider.chat_with_retry.return_value = MagicMock(
+            content="S" * (_ARCHIVE_SUMMARY_MAX_CHARS * 10),
+            finish_reason="stop",
+        )
+        await consolidator.archive([{"role": "user", "content": "hi"}])
+
+        entry = store.read_unprocessed_history(since_cursor=0)[0]
+        assert len(entry["content"]) <= _ARCHIVE_SUMMARY_MAX_CHARS + 50
+
+    async def test_archive_truncates_via_tiktoken_with_positive_budget(self, consolidator, mock_provider, store):
+        """Positive token budget should use tiktoken for precise truncation."""
+        consolidator.context_window_tokens = 10_000
+        consolidator._SAFETY_BUFFER = 0
+        # budget = 10000 - 100 - 0 = 9900 tokens
+        big_messages = [{"role": "user", "content": "word " * 50_000}]
+        mock_provider.chat_with_retry.return_value = MagicMock(
+            content="Summary.", finish_reason="stop"
+        )
+        await consolidator.archive(big_messages)
+
+        import tiktoken
+        enc = tiktoken.get_encoding("cl100k_base")
+        sent_content = mock_provider.chat_with_retry.call_args.kwargs["messages"][1]["content"]
+        token_count = len(enc.encode(sent_content))
+        assert token_count <= 9_900 + 10  # small margin for truncation suffix

tests/agent/test_context_prompt_cache.py

@@ -116,6 +116,20 @@ def test_recent_history_capped_at_max(tmp_path) -> None:
     assert f"entry-{builder._MAX_RECENT_HISTORY + 19}" in prompt
 
 
+def test_recent_history_truncated_at_max_chars(tmp_path) -> None:
+    """Recent History section must be truncated at _MAX_HISTORY_CHARS."""
+    workspace = _make_workspace(tmp_path)
+    builder = ContextBuilder(workspace)
+
+    big_entry = "x" * (builder._MAX_HISTORY_CHARS + 5_000)
+    builder.memory.append_history(big_entry)
+
+    prompt = builder.build_system_prompt()
+    history_section = prompt.split("# Recent History\n\n", 1)
+    assert len(history_section) == 2
+    assert len(history_section[1]) < builder._MAX_HISTORY_CHARS + 200
+
+
 def test_no_recent_history_when_dream_has_processed_all(tmp_path) -> None:
     """If Dream has consumed everything, no Recent History section should appear."""
     workspace = _make_workspace(tmp_path)
@@ -174,6 +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_dream.py

@@ -1,5 +1,7 @@
 """Tests for the Dream class — two-phase memory consolidation via AgentRunner."""
 
+import json
+
 import pytest
 
 from unittest.mock import AsyncMock, MagicMock, patch
@@ -256,3 +258,52 @@ class TestDreamRun:
         # The template renders with stale_threshold_days=14 → LLM must see "N>14"
         assert "N>14" in system_msg
 
+
+class TestDreamPromptCaps:
+    """Dream's Phase 1/2 prompt must not be poisoned by a legacy oversized
+    history entry or a runaway MEMORY.md. Without caps, a single pre-#3412
+    raw_archive dump in history.jsonl would make every subsequent Dream run
+    exceed the context window and silently advance the cursor past real work.
+    """
+
+    async def test_phase1_caps_huge_memory_file(
+        self, dream, mock_provider, mock_runner, store,
+    ):
+        """A MEMORY.md much larger than _MEMORY_FILE_MAX_CHARS must be truncated
+        in the prompt preview (full content is still reachable via read_file)."""
+        store.write_memory("M" * (dream._MEMORY_FILE_MAX_CHARS * 5))
+        store.append_history("some event")
+        mock_provider.chat_with_retry.return_value = MagicMock(content="[SKIP]")
+        mock_runner.run = AsyncMock(return_value=_make_run_result())
+
+        await dream.run()
+
+        user_msg = mock_provider.chat_with_retry.call_args.kwargs["messages"][1]["content"]
+        memory_section = user_msg.split("## Current MEMORY.md")[1].split("## Current SOUL.md")[0]
+        assert len(memory_section) < dream._MEMORY_FILE_MAX_CHARS + 500
+
+    async def test_phase1_caps_huge_history_entry(
+        self, dream, mock_provider, mock_runner, store,
+    ):
+        """A legacy oversized history entry (e.g. pre-#3412 raw_archive dump)
+        must not explode the Phase 1 prompt — each entry is capped in the
+        preview, even though the JSONL record itself stays full-size."""
+        # Bypass the append_history cap by writing directly, simulating a
+        # record that was written by an older nanobot build before any caps.
+        store.history_file.write_text(
+            json.dumps({
+                "cursor": 1,
+                "timestamp": "2026-04-01 10:00",
+                "content": "H" * (dream._HISTORY_ENTRY_PREVIEW_MAX_CHARS * 8),
+            }) + "\n",
+            encoding="utf-8",
+        )
+        mock_provider.chat_with_retry.return_value = MagicMock(content="[SKIP]")
+        mock_runner.run = AsyncMock(return_value=_make_run_result())
+
+        await dream.run()
+
+        user_msg = mock_provider.chat_with_retry.call_args.kwargs["messages"][1]["content"]
+        history_section = user_msg.split("## Conversation History\n")[1].split("\n\n## Current Date")[0]
+        assert len(history_section) < dream._HISTORY_ENTRY_PREVIEW_MAX_CHARS + 500
+

tests/agent/test_loop_progress.py

@@ -0,0 +1,130 @@
+"""Tests for structured tool-event progress metadata emitted by AgentLoop."""
+
+from pathlib import Path
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+
+from nanobot.agent.loop import AgentLoop
+from nanobot.bus.events import InboundMessage
+from nanobot.bus.queue import MessageBus
+from nanobot.providers.base import LLMResponse, ToolCallRequest
+
+
+def _make_loop(tmp_path: Path) -> AgentLoop:
+    bus = MessageBus()
+    provider = MagicMock()
+    provider.get_default_model.return_value = "test-model"
+    return AgentLoop(bus=bus, provider=provider, workspace=tmp_path, model="test-model")
+
+
+class TestToolEventProgress:
+    """_run_agent_loop emits structured tool_events via on_progress."""
+
+    @pytest.mark.asyncio
+    async def test_start_and_finish_events_emitted(self, tmp_path: Path) -> None:
+        loop = _make_loop(tmp_path)
+        tool_call = ToolCallRequest(id="call1", name="custom_tool", arguments={"path": "foo.txt"})
+        calls = iter([
+            LLMResponse(content="Visible", tool_calls=[tool_call]),
+            LLMResponse(content="Done", tool_calls=[]),
+        ])
+        loop.provider.chat_with_retry = AsyncMock(side_effect=lambda *a, **kw: next(calls))
+        loop.tools.get_definitions = MagicMock(return_value=[])
+        loop.tools.prepare_call = MagicMock(return_value=(None, {"path": "foo.txt"}, None))
+        loop.tools.execute = AsyncMock(return_value="ok")
+
+        progress: list[tuple[str, bool, list[dict] | None]] = []
+
+        async def on_progress(
+            content: str,
+            *,
+            tool_hint: bool = False,
+            tool_events: list[dict] | None = None,
+        ) -> None:
+            progress.append((content, tool_hint, tool_events))
+
+        final_content, _, _, _, _ = await loop._run_agent_loop([], on_progress=on_progress)
+
+        assert final_content == "Done"
+        assert progress == [
+            ("Visible", False, None),
+            (
+                'custom_tool("foo.txt")',
+                True,
+                [{
+                    "version": 1,
+                    "phase": "start",
+                    "call_id": "call1",
+                    "name": "custom_tool",
+                    "arguments": {"path": "foo.txt"},
+                    "result": None,
+                    "error": None,
+                    "files": [],
+                    "embeds": [],
+                }],
+            ),
+            (
+                "",
+                False,
+                [{
+                    "version": 1,
+                    "phase": "end",
+                    "call_id": "call1",
+                    "name": "custom_tool",
+                    "arguments": {"path": "foo.txt"},
+                    "result": "ok",
+                    "error": None,
+                    "files": [],
+                    "embeds": [],
+                }],
+            ),
+        ]
+
+    @pytest.mark.asyncio
+    async def test_bus_progress_forwards_tool_events_to_outbound_metadata(self, tmp_path: Path) -> None:
+        """When run() handles a bus message, _tool_events lands in OutboundMessage metadata."""
+        bus = MessageBus()
+        provider = MagicMock()
+        provider.get_default_model.return_value = "test-model"
+        loop = AgentLoop(bus=bus, provider=provider, workspace=tmp_path, model="test-model")
+
+        tool_call = ToolCallRequest(id="tc1", name="exec", arguments={"command": "ls"})
+        calls = iter([
+            LLMResponse(content="", tool_calls=[tool_call]),
+            LLMResponse(content="Done", tool_calls=[]),
+        ])
+        loop.provider.chat_with_retry = AsyncMock(side_effect=lambda *a, **kw: next(calls))
+        loop.tools.get_definitions = MagicMock(return_value=[])
+        loop.tools.prepare_call = MagicMock(return_value=(None, {"command": "ls"}, None))
+        loop.tools.execute = AsyncMock(return_value="file.txt")
+
+        msg = InboundMessage(
+            channel="telegram",
+            sender_id="u1",
+            chat_id="chat1",
+            content="run ls",
+        )
+        await loop._dispatch(msg)
+
+        # Drain all outbound messages and find the one carrying _tool_events
+        outbound = []
+        while bus.outbound_size > 0:
+            outbound.append(await bus.consume_outbound())
+
+        tool_event_msgs = [m for m in outbound if m.metadata and m.metadata.get("_tool_events")]
+        assert tool_event_msgs, "expected at least one outbound message with _tool_events"
+
+        start_msgs = [m for m in tool_event_msgs if m.metadata["_tool_events"][0]["phase"] == "start"]
+        finish_msgs = [m for m in tool_event_msgs if m.metadata["_tool_events"][0]["phase"] in ("end", "error")]
+        assert start_msgs, "expected a start-phase tool event"
+        assert finish_msgs, "expected a finish-phase tool event"
+
+        start = start_msgs[0].metadata["_tool_events"][0]
+        assert start["name"] == "exec"
+        assert start["call_id"] == "tc1"
+        assert start["result"] is None
+
+        finish = finish_msgs[0].metadata["_tool_events"][0]
+        assert finish["phase"] == "end"
+        assert finish["result"] == "file.txt"

tests/agent/test_loop_save_turn.py

@@ -395,7 +395,7 @@ def test_set_tool_context_uses_effective_key_for_spawn_tool(tmp_path: Path) -> N
     loop._set_tool_context(
         "discord",
         "thread-777",
-        effective_key="discord:parent-456:thread:thread-777",
+        session_key="discord:parent-456:thread:thread-777",
     )
 
     assert spawn_tool._origin_channel.get() == "discord"  # type: ignore[attr-defined]
@@ -590,7 +590,14 @@ 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"]
 
@@ -712,3 +719,63 @@ 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_memory_store.py

@@ -5,7 +5,7 @@ from datetime import datetime
 
 import pytest
 
-from nanobot.agent.memory import MemoryStore
+from nanobot.agent.memory import MemoryStore, _HISTORY_ENTRY_HARD_CAP
 
 
 @pytest.fixture
@@ -142,6 +142,49 @@ class TestHistoryWithCursor:
         assert entries[0]["cursor"] in {4, 5}
 
 
+class TestAppendHistoryHardCap:
+    """append_history has a defensive cap that catches new callers who forgot
+    to set their own tighter cap. The default is intentionally larger than
+    any current caller's per-call cap, so normal operation never trips it."""
+
+    def test_oversized_entry_is_truncated(self, store):
+        """An entry above _HISTORY_ENTRY_HARD_CAP is truncated before being persisted."""
+        huge = "x" * (_HISTORY_ENTRY_HARD_CAP + 10_000)
+        store.append_history(huge)
+        entry = store.read_unprocessed_history(since_cursor=0)[0]
+        assert len(entry["content"]) <= _HISTORY_ENTRY_HARD_CAP + 50
+
+    def test_oversize_warning_is_emitted_once(self, store, caplog):
+        """Repeated oversized writes should warn only on the first occurrence."""
+        from loguru import logger as loguru_logger
+
+        records: list[str] = []
+        handler_id = loguru_logger.add(lambda m: records.append(m), level="WARNING")
+        try:
+            huge = "x" * (_HISTORY_ENTRY_HARD_CAP + 1)
+            store.append_history(huge)
+            store.append_history(huge)
+            store.append_history(huge)
+        finally:
+            loguru_logger.remove(handler_id)
+
+        oversize_warnings = [r for r in records if "exceeds" in r and "chars" in r]
+        assert len(oversize_warnings) == 1
+
+    def test_custom_max_chars_overrides_default(self, store):
+        """Callers that pass max_chars should get their tighter cap applied."""
+        store.append_history("a" * 500, max_chars=100)
+        entry = store.read_unprocessed_history(since_cursor=0)[0]
+        assert len(entry["content"]) <= 150  # 100 + "\n... (truncated)"
+
+    def test_normal_sized_entries_unaffected(self, store):
+        """The hard cap must not alter entries that fit within it."""
+        msg = "normal short entry"
+        store.append_history(msg)
+        entry = store.read_unprocessed_history(since_cursor=0)[0]
+        assert entry["content"] == msg
+
+
 class TestDreamCursor:
     def test_initial_cursor_is_zero(self, store):
         assert store.get_last_dream_cursor() == 0

tests/agent/test_runner.py

@@ -252,6 +252,35 @@ async def test_runner_returns_max_iterations_fallback():
     assert result.messages[-1]["role"] == "assistant"
     assert result.messages[-1]["content"] == result.final_content
 
+
+@pytest.mark.asyncio
+async def test_runner_times_out_hung_llm_request():
+    from nanobot.agent.runner import AgentRunSpec, AgentRunner
+
+    provider = MagicMock()
+
+    async def chat_with_retry(**kwargs):
+        await asyncio.sleep(3600)
+
+    provider.chat_with_retry = chat_with_retry
+    tools = MagicMock()
+    tools.get_definitions.return_value = []
+
+    runner = AgentRunner(provider)
+    started = time.monotonic()
+    result = await runner.run(AgentRunSpec(
+        initial_messages=[{"role": "user", "content": "hello"}],
+        tools=tools,
+        model="test-model",
+        max_iterations=1,
+        max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
+        llm_timeout_s=0.05,
+    ))
+
+    assert (time.monotonic() - started) < 1.0
+    assert result.stop_reason == "error"
+    assert "timed out" in (result.final_content or "").lower()
+
 @pytest.mark.asyncio
 async def test_runner_returns_structured_tool_error():
     from nanobot.agent.runner import AgentRunSpec, AgentRunner
@@ -1031,11 +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():
@@ -269,3 +350,66 @@ def test_get_history_ignores_media_kwarg_on_non_user_rows():
     # List content is passed through verbatim — the synthesizer only
     # rewrites plain-string content.
     assert history[0]["content"] == [{"type": "text", "text": "structured"}]
+
+
+def test_get_history_respects_max_tokens(monkeypatch):
+    session = Session(key="test:token-cap")
+    session.messages.extend(
+        [
+            {"role": "user", "content": "u1"},
+            {"role": "assistant", "content": "a1"},
+            {"role": "user", "content": "u2"},
+            {"role": "assistant", "content": "a2"},
+            {"role": "user", "content": "u3"},
+            {"role": "assistant", "content": "a3"},
+        ]
+    )
+
+    token_map = {"u1": 50, "a1": 50, "u2": 50, "a2": 50, "u3": 50, "a3": 50}
+    monkeypatch.setattr(
+        "nanobot.session.manager.estimate_message_tokens",
+        lambda message: token_map.get(message.get("content"), 0),
+    )
+
+    history = session.get_history(max_messages=500, max_tokens=120)
+    assert [m["content"] for m in history] == ["u3", "a3"]
+
+
+def test_get_history_recovers_user_when_token_slice_would_be_assistant_only(monkeypatch):
+    session = Session(key="test:assistant-only-slice")
+    session.messages.extend(
+        [
+            {"role": "user", "content": "u1"},
+            {"role": "assistant", "content": "a1"},
+            {"role": "user", "content": "u2"},
+            {"role": "assistant", "content": "a2"},
+        ]
+    )
+    token_map = {"u1": 100, "a1": 100, "u2": 100, "a2": 100}
+    monkeypatch.setattr(
+        "nanobot.session.manager.estimate_message_tokens",
+        lambda message: token_map.get(message.get("content"), 0),
+    )
+
+    history = session.get_history(max_messages=500, max_tokens=100)
+    assert [m["content"] for m in history] == ["u2", "a2"]
+
+
+def test_retain_recent_legal_suffix_hard_cap_with_long_non_user_chain():
+    session = Session(key="test:hard-cap-chain")
+    session.messages.append({"role": "user", "content": "u0"})
+    session.messages.append(
+        {
+            "role": "assistant",
+            "content": None,
+            "tool_calls": [
+                {"id": "c1", "type": "function", "function": {"name": "x", "arguments": "{}"}}
+            ],
+        }
+    )
+    for i in range(12):
+        session.messages.append({"role": "assistant", "content": f"a{i}"})
+
+    session.retain_recent_legal_suffix(6)
+
+    assert len(session.messages) <= 6