fix: reorder fake tool call messages at tail to correct timing

astrbot/core/provider/__init__.py

@@ -1,4 +1,9 @@
 from .entities import ProviderMetaData
-from .provider import Provider, STTProvider
+from .provider import Provider, STTProvider, reorder_tailing_tool_call_user
 
-__all__ = ["Provider", "ProviderMetaData", "STTProvider"]
+__all__ = [
+    "Provider",
+    "ProviderMetaData",
+    "STTProvider",
+    "reorder_tailing_tool_call_user",
+]

astrbot/core/provider/provider.py

@@ -2,7 +2,7 @@
 import asyncio
 import os
 from collections.abc import AsyncGenerator
-from typing import Literal, TypeAlias, Union
+from typing import Any, Literal, TypeAlias, Union
 
 from astrbot.core.agent.message import ContentPart, Message, is_checkpoint_message
 from astrbot.core.agent.tool import ToolSet
@@ -211,6 +211,52 @@ async def test(self, timeout: float = 45.0) -> None:
         )
 
 
+def _is_valid_tool_pair(asst_msg: dict[str, Any], tool_msg: dict[str, Any]) -> bool:
+    """判断 assistant(tool_calls) 与 tool 是否为 tool_call_id 匹配的一对。"""
+    if asst_msg.get("role") != "assistant" or tool_msg.get("role") != "tool":
+        return False
+    tool_calls = asst_msg.get("tool_calls")
+    if not tool_calls:
+        return False
+    tc_ids = {tc.get("id") for tc in tool_calls if isinstance(tc, dict)}
+    return tool_msg.get("tool_call_id") in tc_ids
+
+
+def reorder_tailing_tool_call_user(messages: list[dict[str, Any]]) -> None:
+    """重排因伪造工具调用导致尾部 assistant(tc) → tool → user 乱序的消息。
+
+    此为轻量妥协修复。未来若实现专用的上下文操作钩子或伪造工具调用钩子，
+    可考虑移除此函数。
+    """
+    if not isinstance(messages, list) or len(messages) < 2:
+        return
+
+    last = messages[-1]
+    if last.get("role") != "user":
+        return
+
+    # 从最后一个非 user 元素向前扫描 assistant(tool_calls) + tool 成对消息
+    pairs: list[tuple[dict[str, Any], dict[str, Any]]] = []
+    i = len(messages) - 2
+    while i >= 1:
+        if not _is_valid_tool_pair(messages[i - 1], messages[i]):
+            break
+        pairs.append((messages[i - 1], messages[i]))
+        i -= 2
+
+    if not pairs:
+        return
+
+    # 重排：user → assistant_1, tool_1 → ... → assistant_N, tool_N
+    # pairs 从内到外收集（N, N-1, ..., 1），反转后按 1..N 顺序回插
+    new_tail: list[dict[str, Any]] = [last]
+    for asst_msg, tool_msg in reversed(pairs):
+        new_tail.append(asst_msg)
+        new_tail.append(tool_msg)
+
+    messages[:] = messages[: i + 1] + new_tail
+
+
 class STTProvider(AbstractProvider):
     def __init__(self, provider_config: dict, provider_settings: dict) -> None:
         super().__init__(provider_config)

astrbot/core/provider/sources/anthropic_source.py

@@ -14,6 +14,7 @@
 from astrbot.api.provider import Provider
 from astrbot.core.agent.message import AudioURLPart, ContentPart, ImageURLPart, TextPart
 from astrbot.core.exceptions import EmptyModelOutputError
+from astrbot.core.provider import reorder_tailing_tool_call_user
 from astrbot.core.provider.entities import LLMResponse, TokenUsage
 from astrbot.core.provider.func_tool_manager import ToolSet
 from astrbot.core.utils.media_utils import (
@@ -789,6 +790,8 @@ async def text_chat(
                 for tool_call_result in tool_calls_result:
                     context_query.extend(tool_call_result.to_openai_messages())
 
+        # 伪造工具调用对需前置到用户消息之后，与真实工具调用时序对齐
+        reorder_tailing_tool_call_user(context_query)
         system_prompt, new_messages = self._prepare_payload(context_query)
 
         model = model or self.get_model()
@@ -861,6 +864,8 @@ async def text_chat_stream(
                 for tool_call_result in tool_calls_result:
                     context_query.extend(tool_call_result.to_openai_messages())
 
+        # 伪造工具调用对需前置到用户消息之后，与真实工具调用时序对齐
+        reorder_tailing_tool_call_user(context_query)
         system_prompt, new_messages = self._prepare_payload(context_query)
 
         model = model or self.get_model()

astrbot/core/provider/sources/gemini_source.py

@@ -17,6 +17,7 @@
 from astrbot.core.agent.message import AudioURLPart, ContentPart, ImageURLPart, TextPart
 from astrbot.core.exceptions import EmptyModelOutputError
 from astrbot.core.message.message_event_result import MessageChain
+from astrbot.core.provider import reorder_tailing_tool_call_user
 from astrbot.core.provider.entities import LLMResponse, TokenUsage
 from astrbot.core.provider.func_tool_manager import ToolSet
 from astrbot.core.utils.media_utils import (
@@ -857,6 +858,10 @@ async def text_chat(
                 for tcr in tool_calls_result:
                     context_query.extend(tcr.to_openai_messages())
 
+        # 伪造工具调用对需前置到用户消息之后，与真实工具调用时序对齐。
+        # Gemini API 要求 functionCall 回合必须紧跟 user 回合，否则返回 400。
+        reorder_tailing_tool_call_user(context_query)
+
         model = model or self.get_model()
 
         payloads = {"messages": context_query, "model": model}
@@ -924,6 +929,10 @@ async def text_chat_stream(
                 for tcr in tool_calls_result:
                     context_query.extend(tcr.to_openai_messages())
 
+        # 伪造工具调用对需前置到用户消息之后，与真实工具调用时序对齐。
+        # Gemini API 要求 functionCall 回合必须紧跟 user 回合，否则返回 400。
+        reorder_tailing_tool_call_user(context_query)
+
         model = model or self.get_model()
 
         payloads = {"messages": context_query, "model": model}

astrbot/core/provider/sources/openai_source.py

@@ -28,6 +28,7 @@
 from astrbot.core.agent.tool import ToolSet
 from astrbot.core.exceptions import EmptyModelOutputError
 from astrbot.core.message.message_event_result import MessageChain
+from astrbot.core.provider import reorder_tailing_tool_call_user
 from astrbot.core.provider.entities import LLMResponse, TokenUsage, ToolCallsResult
 from astrbot.core.utils.media_utils import (
     describe_media_ref,
@@ -528,6 +529,8 @@ def _is_empty(content: Any) -> bool:
             )
         payloads["messages"] = final
 
+        reorder_tailing_tool_call_user(payloads["messages"])
+
     async def _query(
         self,
         payloads: dict,

tests/fixtures/fake_tool_call.py

@@ -0,0 +1,23 @@
+"""伪造工具调用（fake tool call）共享测试数据。
+
+各 provider 测试（OpenAI / Anthropic / Gemini 格式）共用同一组伪造
+assistant(tool_calls) + tool 消息对，避免场景漂移。
+"""
+
+FAKE_TOOL_CALL_CONTEXTS = [
+    {
+        "role": "assistant",
+        "content": None,
+        "tool_calls": [
+            {
+                "id": "fake_recall_abc",
+                "type": "function",
+                "function": {
+                    "name": "recall_long_term_memory",
+                    "arguments": '{"query": "我的名字是？", "k": 5}',
+                },
+            }
+        ],
+    },
+    {"role": "tool", "tool_call_id": "fake_recall_abc", "content": "memory json"},
+]

tests/test_anthropic_kimi_code_provider.py

@@ -9,6 +9,7 @@
 import astrbot.core.provider.sources.request_retry as request_retry
 from astrbot.core.exceptions import EmptyModelOutputError
 from astrbot.core.provider.entities import LLMResponse
+from tests.fixtures.fake_tool_call import FAKE_TOOL_CALL_CONTEXTS
 
 
 class _FakeAsyncAnthropic:
@@ -864,3 +865,75 @@ async def test_tool_choice_empty_tool_list_skips_tool_choice(monkeypatch):
     kwargs = _capture_payloads_create.last_kwargs
     assert "tools" not in kwargs
     assert "tool_choice" not in kwargs
+
+
+# ── fake tool call 重排 ───────────────────────────────────────────────────────
+
+_EXPECTED_REORDERED_MESSAGES = [
+    {"role": "user", "content": "我的名字是？"},
+    {
+        "role": "assistant",
+        "content": [
+            {
+                "type": "tool_use",
+                "name": "recall_long_term_memory",
+                "input": {"query": "我的名字是？", "k": 5},
+                "id": "fake_recall_abc",
+            }
+        ],
+    },
+    {
+        "role": "user",
+        "content": [
+            {
+                "type": "tool_result",
+                "tool_use_id": "fake_recall_abc",
+                "content": "memory json",
+            }
+        ],
+    },
+]
+
+
+@pytest.mark.asyncio
+async def test_text_chat_reorders_fake_tool_call_pair(monkeypatch):
+    """伪造工具调用对应重排到用户消息之后，与真实工具调用时序对齐。"""
+    provider = _setup_provider_with_mock_client(monkeypatch)
+
+    await provider.text_chat(prompt="我的名字是？", contexts=FAKE_TOOL_CALL_CONTEXTS)
+
+    assert _capture_payloads_create.last_kwargs["messages"] == (
+        _EXPECTED_REORDERED_MESSAGES
+    )
+
+
+@pytest.mark.asyncio
+async def test_text_chat_stream_reorders_fake_tool_call_pair(monkeypatch):
+    """流式路径同样应将伪造工具调用对重排到用户消息之后。"""
+    monkeypatch.setattr(anthropic_source, "AsyncAnthropic", _FakeAsyncAnthropic)
+
+    provider = anthropic_source.ProviderAnthropic(
+        provider_config={
+            "id": "anthropic-test",
+            "type": "anthropic_chat_completion",
+            "model": "claude-test",
+            "key": ["test-key"],
+        },
+        provider_settings={},
+    )
+
+    captured: dict[str, object] = {}
+
+    async def fake_query_stream(payloads, tools, *, request_max_retries=None):
+        captured["messages"] = payloads["messages"]
+        return
+        yield  # pragma: no cover  # 保持 async generator 形态
+
+    monkeypatch.setattr(provider, "_query_stream", fake_query_stream)
+
+    async for _ in provider.text_chat_stream(
+        prompt="我的名字是？", contexts=FAKE_TOOL_CALL_CONTEXTS
+    ):
+        pass
+
+    assert captured["messages"] == _EXPECTED_REORDERED_MESSAGES

tests/test_gemini_source.py

@@ -3,10 +3,11 @@
 import httpx
 import pytest
 
-from astrbot.core.exceptions import EmptyModelOutputError
 import astrbot.core.provider.sources.request_retry as request_retry
+from astrbot.core.exceptions import EmptyModelOutputError
 from astrbot.core.provider.entities import LLMResponse
 from astrbot.core.provider.sources.gemini_source import ProviderGoogleGenAI
+from tests.fixtures.fake_tool_call import FAKE_TOOL_CALL_CONTEXTS
 
 
 def test_gemini_empty_output_raises_empty_model_output_error():
@@ -33,6 +34,71 @@ def test_gemini_reasoning_only_output_is_allowed():
     )
 
 
+def _make_provider() -> ProviderGoogleGenAI:
+    provider = ProviderGoogleGenAI.__new__(ProviderGoogleGenAI)
+    provider.api_keys = [""]
+    provider.model_name = ""
+    return provider
+
+
+def _assert_reordered_tail(messages: list[dict]) -> None:
+    assert messages[-3]["role"] == "user"
+    assert messages[-3]["content"] == "我的名字是？"
+    assert messages[-2]["role"] == "assistant"
+    assert messages[-2]["tool_calls"][0]["id"] == "fake_recall_abc"
+    assert (
+        messages[-2]["tool_calls"][0]["function"]["name"] == "recall_long_term_memory"
+    )
+    assert messages[-1]["role"] == "tool"
+    assert messages[-1]["tool_call_id"] == "fake_recall_abc"
+
+
+@pytest.mark.asyncio
+async def test_text_chat_reorders_fake_tool_call_pair(monkeypatch):
+    captured = {}
+
+    async def fake_query(payloads, tools, *, request_max_retries=None):
+        captured["payloads"] = payloads
+        return LLMResponse(role="assistant")
+
+    provider = _make_provider()
+    monkeypatch.setattr(provider, "_query", fake_query)
+
+    await provider.text_chat(prompt="我的名字是？", contexts=FAKE_TOOL_CALL_CONTEXTS)
+
+    messages = captured["payloads"]["messages"]
+    _assert_reordered_tail(messages)
+
+    # 转换后的 contents 应为 user → model(functionCall) → user(functionResponse)
+    contents = provider._prepare_conversation(captured["payloads"])
+    assert [content.role for content in contents] == ["user", "model", "user"]
+    assert contents[0].parts[0].text == "我的名字是？"
+    assert contents[1].parts[0].function_call.name == "recall_long_term_memory"
+    assert contents[2].parts[0].function_response.name == "fake_recall_abc"
+
+
+@pytest.mark.asyncio
+async def test_text_chat_stream_reorders_fake_tool_call_pair(monkeypatch):
+    captured = {}
+
+    async def fake_query_stream(payloads, tools, *, request_max_retries=None):
+        captured["payloads"] = payloads
+        return
+        yield  # pragma: no cover
+
+    provider = _make_provider()
+    monkeypatch.setattr(provider, "_query_stream", fake_query_stream)
+
+    async for _ in provider.text_chat_stream(
+        prompt="我的名字是？",
+        contexts=FAKE_TOOL_CALL_CONTEXTS,
+    ):
+        pass
+
+    messages = captured["payloads"]["messages"]
+    _assert_reordered_tail(messages)
+
+
 @pytest.mark.asyncio
 async def test_gemini_get_models_retries_transient_request_error(monkeypatch):
     monkeypatch.setattr(request_retry, "REQUEST_RETRY_WAIT_MIN_S", 0)