From e286da75c45e4b62793c25365a7621c59cff03d8 Mon Sep 17 00:00:00 2001
From: Yufeng He <40085740+he-yufeng@users.noreply.github.com>
Date: Fri, 20 Mar 2026 10:06:20 +0800
Subject: [PATCH 1/6] =?UTF-8?q?fix:=20=E6=88=AA=E6=96=AD=E5=99=A8=E4=B8=A2?=
 =?UTF-8?q?=E5=A4=B1=E5=94=AF=E4=B8=80=20user=20=E6=B6=88=E6=81=AF?=
 =?UTF-8?q?=E5=AF=BC=E8=87=B4=E6=99=BA=E8=B0=B1=E7=AD=89=20provider=20?=
 =?UTF-8?q?=E8=BF=94=E5=9B=9E=20400=20(#6581)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

* fix: 截断器丢失唯一 user 消息导致 API 400

修复 #6196

当对话只有一条 user 消息（长 tool chain 场景：system → user → assistant
→ tool → assistant → tool → ...），三个截断方法都会把这条 user 消息丢掉，
导致智谱、Gemini 等要求 user 消息的 provider 返回 400。

改动：
- 提取 `_split_system_rest()` 去掉三个方法里重复的 system/non-system 拆分
- 新增 `_ensure_user_message()`：截断后如果没有 user 了，从原始消息里补回
  第一条 user，避免违反 API 格式要求
- 删掉 `truncate_by_dropping_oldest_turns` 里把没有 user 就清空全部消息的逻辑
- 5 个新测试覆盖单 user + 长 tool chain 场景，3 个旧测试更新断言

* style: format code

---------

Co-authored-by: Yufeng He <40085740+universeplayer@users.noreply.github.com>
Co-authored-by: RC-CHN <1051989940@qq.com>
---
 astrbot/core/agent/context/truncator.py | 79 ++++++++++++++-----------
 tests/agent/test_truncator.py           | 72 ++++++++++++++++++++--
 2 files changed, 112 insertions(+), 39 deletions(-)

diff --git a/astrbot/core/agent/context/truncator.py b/astrbot/core/agent/context/truncator.py
index afd89f2be..b552b5ef2 100644
--- a/astrbot/core/agent/context/truncator.py
+++ b/astrbot/core/agent/context/truncator.py
@@ -12,6 +12,37 @@ class ContextTruncator:
             and len(message.tool_calls) > 0
         )
 
+    @staticmethod
+    def _split_system_rest(
+        messages: list[Message],
+    ) -> tuple[list[Message], list[Message]]:
+        """把 system 消息和后面的对话消息分开。"""
+        first_non_system = 0
+        for i, msg in enumerate(messages):
+            if msg.role != "system":
+                first_non_system = i
+                break
+        return messages[:first_non_system], messages[first_non_system:]
+
+    @staticmethod
+    def _ensure_user_message(
+        system_messages: list[Message],
+        truncated: list[Message],
+        original_messages: list[Message],
+    ) -> list[Message]:
+        """截断后如果没有 user 消息了，从原始列表里把第一条 user 补回来。
+        很多 provider (智谱、Gemini 等) 要求 system 之后必须紧跟 user，否则直接 400。
+        """
+        if any(m.role == "user" for m in truncated):
+            return system_messages + truncated
+
+        # 从原始消息里找第一条 user
+        first_user = next((m for m in original_messages if m.role == "user"), None)
+        if first_user is None:
+            return system_messages + truncated
+
+        return system_messages + [first_user] + truncated
+
     def fix_messages(self, messages: list[Message]) -> list[Message]:
         """修复消息列表，确保 tool call 和 tool response 的配对关系有效。
 
@@ -81,14 +112,7 @@ class ContextTruncator:
         if keep_most_recent_turns == -1:
             return messages
 
-        first_non_system = 0
-        for i, msg in enumerate(messages):
-            if msg.role != "system":
-                first_non_system = i
-                break
-
-        system_messages = messages[:first_non_system]
-        non_system_messages = messages[first_non_system:]
+        system_messages, non_system_messages = self._split_system_rest(messages)
 
         if len(non_system_messages) // 2 <= keep_most_recent_turns:
             return messages
@@ -99,7 +123,7 @@ class ContextTruncator:
         else:
             truncated_contexts = non_system_messages[-num_to_keep * 2 :]
 
-        # 找到第一个 role 为 user 的索引，确保上下文格式正确
+        # 对齐到第一条 user 消息
         index = next(
             (i for i, item in enumerate(truncated_contexts) if item.role == "user"),
             None,
@@ -107,8 +131,9 @@ class ContextTruncator:
         if index is not None and index > 0:
             truncated_contexts = truncated_contexts[index:]
 
-        result = system_messages + truncated_contexts
-
+        result = self._ensure_user_message(
+            system_messages, truncated_contexts, messages
+        )
         return self.fix_messages(result)
 
     def truncate_by_dropping_oldest_turns(
@@ -120,31 +145,24 @@ class ContextTruncator:
         if drop_turns <= 0:
             return messages
 
-        first_non_system = 0
-        for i, msg in enumerate(messages):
-            if msg.role != "system":
-                first_non_system = i
-                break
-
-        system_messages = messages[:first_non_system]
-        non_system_messages = messages[first_non_system:]
+        system_messages, non_system_messages = self._split_system_rest(messages)
 
         if len(non_system_messages) // 2 <= drop_turns:
             truncated_non_system = []
         else:
             truncated_non_system = non_system_messages[drop_turns * 2 :]
 
+        # 对齐到第一条 user
         index = next(
             (i for i, item in enumerate(truncated_non_system) if item.role == "user"),
             None,
         )
         if index is not None:
             truncated_non_system = truncated_non_system[index:]
-        elif truncated_non_system:
-            truncated_non_system = []
-
-        result = system_messages + truncated_non_system
 
+        result = self._ensure_user_message(
+            system_messages, truncated_non_system, messages
+        )
         return self.fix_messages(result)
 
     def truncate_by_halving(
@@ -155,14 +173,7 @@ class ContextTruncator:
         if len(messages) <= 2:
             return messages
 
-        first_non_system = 0
-        for i, msg in enumerate(messages):
-            if msg.role != "system":
-                first_non_system = i
-                break
-
-        system_messages = messages[:first_non_system]
-        non_system_messages = messages[first_non_system:]
+        system_messages, non_system_messages = self._split_system_rest(messages)
 
         messages_to_delete = len(non_system_messages) // 2
         if messages_to_delete == 0:
@@ -170,6 +181,7 @@ class ContextTruncator:
 
         truncated_non_system = non_system_messages[messages_to_delete:]
 
+        # 对齐到第一条 user
         index = next(
             (i for i, item in enumerate(truncated_non_system) if item.role == "user"),
             None,
@@ -177,6 +189,7 @@ class ContextTruncator:
         if index is not None:
             truncated_non_system = truncated_non_system[index:]
 
-        result = system_messages + truncated_non_system
-
+        result = self._ensure_user_message(
+            system_messages, truncated_non_system, messages
+        )
         return self.fix_messages(result)
diff --git a/tests/agent/test_truncator.py b/tests/agent/test_truncator.py
index c85fb7d55..7dac80f9c 100644
--- a/tests/agent/test_truncator.py
+++ b/tests/agent/test_truncator.py
@@ -104,8 +104,9 @@ class TestContextTruncator:
             messages, keep_most_recent_turns=0, drop_turns=1
         )
 
-        # Should result in empty or minimal list
-        assert len(result) == 0
+        # 截断后至少保留一条 user 消息 (#6196)
+        assert len(result) >= 1
+        assert result[0].role == "user"
 
     def test_truncate_by_turns_below_threshold(self):
         """Test truncate_by_turns when messages are below threshold."""
@@ -201,8 +202,9 @@ class TestContextTruncator:
         messages = self.create_messages(4)
         result = truncator.truncate_by_dropping_oldest_turns(messages, drop_turns=2)
 
-        # Should drop all turns
-        assert len(result) == 0
+        # 即使 drop 掉所有 turn，也会把 user 消息补回来 (#6196)
+        assert len(result) >= 1
+        assert result[0].role == "user"
 
     def test_truncate_by_dropping_oldest_turns_drop_more_than_available(self):
         """Test truncate_by_dropping_oldest_turns with drop_turns > available turns."""
@@ -211,8 +213,9 @@ class TestContextTruncator:
         messages = self.create_messages(4)
         result = truncator.truncate_by_dropping_oldest_turns(messages, drop_turns=5)
 
-        # Should result in empty list
-        assert len(result) == 0
+        # 同理，user 消息会被保留 (#6196)
+        assert len(result) >= 1
+        assert result[0].role == "user"
 
     def test_truncate_by_dropping_oldest_turns_ensures_user_first(self):
         """Test that result starts with user message after dropping."""
@@ -372,3 +375,60 @@ class TestContextTruncator:
         assert len(result) >= 0  # May keep system messages or clear all
         if len(result) > 0:
             assert all(msg.role == "system" for msg in result)
+
+    # ==================== #6196: 长 tool chain 只有一条 user 消息 ====================
+
+    def _build_tool_chain(self, tool_rounds: int = 20) -> list[Message]:
+        """构造 system -> user -> (assistant -> tool) * N 的长链，只有一条 user。"""
+        msgs = [
+            self.create_message("system", "You are a helpful assistant."),
+            self.create_message("user", "帮我查一下天气"),
+        ]
+        for i in range(tool_rounds):
+            msgs.append(self.create_message("assistant", f"调用工具 {i}"))
+            msgs.append(self.create_message("tool", f"工具结果 {i}"))
+        return msgs
+
+    def test_drop_oldest_preserves_sole_user(self):
+        """#6196: drop 1 turn 不应丢掉唯一的 user 消息。"""
+        truncator = ContextTruncator()
+        msgs = self._build_tool_chain(20)  # 1 system + 1 user + 40 asst/tool = 42
+        result = truncator.truncate_by_dropping_oldest_turns(msgs, drop_turns=1)
+        roles = [m.role for m in result]
+        assert "user" in roles, "唯一的 user 消息被丢掉了"
+        assert roles[0] == "system"
+
+    def test_halving_preserves_sole_user(self):
+        """#6196: 对半砍不应丢掉唯一的 user 消息。"""
+        truncator = ContextTruncator()
+        msgs = self._build_tool_chain(20)
+        result = truncator.truncate_by_halving(msgs)
+        roles = [m.role for m in result]
+        assert "user" in roles, "唯一的 user 消息被丢掉了"
+
+    def test_truncate_by_turns_preserves_sole_user(self):
+        """#6196: keep_most_recent_turns 也不应丢掉唯一的 user 消息。"""
+        truncator = ContextTruncator()
+        msgs = self._build_tool_chain(20)
+        result = truncator.truncate_by_turns(
+            msgs, keep_most_recent_turns=3, drop_turns=1
+        )
+        roles = [m.role for m in result]
+        assert "user" in roles, "唯一的 user 消息被丢掉了"
+
+    def test_drop_oldest_heavy_drops_still_has_user(self):
+        """#6196: 大量 drop 也不会丢 user。"""
+        truncator = ContextTruncator()
+        msgs = self._build_tool_chain(30)
+        result = truncator.truncate_by_dropping_oldest_turns(msgs, drop_turns=10)
+        roles = [m.role for m in result]
+        assert "user" in roles
+
+    def test_normal_multi_user_not_affected(self):
+        """正常多 user 对话不受影响。"""
+        truncator = ContextTruncator()
+        msgs = self.create_messages(20, include_system=True)
+        result_before = truncator.truncate_by_dropping_oldest_turns(msgs, drop_turns=2)
+        # 多 user 场景下截断后仍有 user
+        roles = [m.role for m in result_before]
+        assert "user" in roles

From 0b690344913cda8acc1e2bca07114b10717bd76c Mon Sep 17 00:00:00 2001
From: leonforcode <leonbeyourside01@gmail.com>
Date: Fri, 20 Mar 2026 12:05:54 +0800
Subject: [PATCH 2/6] fix: prevent truncation logic from removing the only user
 message in long tool-calling conversations (#6198)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

* fix: 压缩算法删除 user 消息 Bug 修复

* perf: improve truncate algo

---------

Co-authored-by: Soulter <905617992@qq.com>
---
 astrbot/core/agent/context/truncator.py | 63 ++++++++++++++-----------
 1 file changed, 35 insertions(+), 28 deletions(-)

diff --git a/astrbot/core/agent/context/truncator.py b/astrbot/core/agent/context/truncator.py
index b552b5ef2..9abf57433 100644
--- a/astrbot/core/agent/context/truncator.py
+++ b/astrbot/core/agent/context/truncator.py
@@ -16,7 +16,11 @@ class ContextTruncator:
     def _split_system_rest(
         messages: list[Message],
     ) -> tuple[list[Message], list[Message]]:
-        """把 system 消息和后面的对话消息分开。"""
+        """Split messages into system messages and the rest.
+
+        Returns:
+            tuple: (system_messages, non_system_messages)
+        """
         first_non_system = 0
         for i, msg in enumerate(messages):
             if msg.role != "system":
@@ -30,13 +34,14 @@ class ContextTruncator:
         truncated: list[Message],
         original_messages: list[Message],
     ) -> list[Message]:
-        """截断后如果没有 user 消息了，从原始列表里把第一条 user 补回来。
-        很多 provider (智谱、Gemini 等) 要求 system 之后必须紧跟 user，否则直接 400。
+        """Ensure the result always contains the first user message right after
+        system messages. This is required by many LLM APIs (e.g. Zhipu) that
+        mandate a ``user`` message immediately following the ``system`` message.
         """
-        if any(m.role == "user" for m in truncated):
+        if truncated and truncated[0].role == "user":
             return system_messages + truncated
 
-        # 从原始消息里找第一条 user
+        # Locate the first user message from the *original* list.
         first_user = next((m for m in original_messages if m.role == "user"), None)
         if first_user is None:
             return system_messages + truncated
@@ -44,13 +49,13 @@ class ContextTruncator:
         return system_messages + [first_user] + truncated
 
     def fix_messages(self, messages: list[Message]) -> list[Message]:
-        """修复消息列表，确保 tool call 和 tool response 的配对关系有效。
+        """Fix the message list to ensure the validity of tool call and tool response pairing.
 
-        此方法确保：
-        1. 每个 `tool` 消息前面都有一个包含 tool_calls 的 `assistant` 消息
-        2. 每个包含 tool_calls 的 `assistant` 消息后面都有对应的 `tool` 响应
+        This method ensures that:
+        1. Each `tool` message is preceded by an `assistant` message containing `tool_calls`.
+        2. Each `assistant` message containing `tool_calls` is followed by corresponding `
 
-        这是 OpenAI Chat Completions API 规范的要求（Gemini 对此执行严格检查）。
+        This is a requirement of the OpenAI Chat Completions API specification (Gemini enforces this strictly).
         """
         if not messages:
             return messages
@@ -69,24 +74,25 @@ class ContextTruncator:
 
         for msg in messages:
             if msg.role == "tool":
-                # 只有在有挂起的 assistant(tool_calls) 时才记录 tool 响应
+                # Only record tool responses when there is a pending assistant(tool_calls)
                 if pending_assistant is not None:
                     pending_tools.append(msg)
-                # else: 孤立的 tool 消息，直接忽略
+                # Isolated tool messages without a preceding assistant(tool_calls) are ignored
                 continue
 
             if self._has_tool_calls(msg):
-                # 遇到新的 assistant(tool_calls) 前，先处理旧的 pending 链
+                # When encountering a new assistant(tool_calls), first process the old pending chain
                 flush_pending_if_valid()
                 pending_assistant = msg
                 continue
 
-            # 非 tool，且不含 tool_calls 的消息
-            # 先结束任何 pending 链，再正常追加
+            # Non-tool messages that do not contain tool_calls will break the pending chain.
+            # Flush any pending chain first, then append the current message normally.
             flush_pending_if_valid()
             fixed_messages.append(msg)
 
-        # 结束时处理最后一个 pending 链
+        # Flush the last pending chain at the end,
+        # ensuring that any remaining valid assistant(tool_calls) and its tools are included in the final list.
         flush_pending_if_valid()
 
         return fixed_messages
@@ -97,17 +103,18 @@ class ContextTruncator:
         keep_most_recent_turns: int,
         drop_turns: int = 1,
     ) -> list[Message]:
-        """截断上下文列表，确保不超过最大长度。
-        一个 turn 包含一个 user 消息和一个 assistant 消息。
-        这个方法会保证截断后的上下文列表符合 OpenAI 的上下文格式。
+        """
+        Turn-based truncation strategy, which drops the oldest turns while keeping the most recent N turns.
+        A turn consists of a user message and an assistant message.
+        This method ensures that the truncated context list conforms to OpenAI's context format.
 
         Args:
-            messages: 上下文列表
-            keep_most_recent_turns: 保留最近的对话轮数
-            drop_turns: 一次性丢弃的对话轮数
+            messages: The original list of messages in the context.
+            keep_most_recent_turns: The number of most recent turns to keep. If set to -1, it means keeping all turns (no truncation).
+            drop_turns: The number of turns to drop from the beginning.
 
         Returns:
-            截断后的上下文列表
+            The truncated list of messages.
         """
         if keep_most_recent_turns == -1:
             return messages
@@ -123,7 +130,7 @@ class ContextTruncator:
         else:
             truncated_contexts = non_system_messages[-num_to_keep * 2 :]
 
-        # 对齐到第一条 user 消息
+        # Find the first user message
         index = next(
             (i for i, item in enumerate(truncated_contexts) if item.role == "user"),
             None,
@@ -141,7 +148,7 @@ class ContextTruncator:
         messages: list[Message],
         drop_turns: int = 1,
     ) -> list[Message]:
-        """丢弃最旧的 N 个对话轮次。"""
+        """Drop the oldest N turns, regardless of the number of turns to keep."""
         if drop_turns <= 0:
             return messages
 
@@ -152,7 +159,7 @@ class ContextTruncator:
         else:
             truncated_non_system = non_system_messages[drop_turns * 2 :]
 
-        # 对齐到第一条 user
+        # Find the first user message
         index = next(
             (i for i, item in enumerate(truncated_non_system) if item.role == "user"),
             None,
@@ -169,7 +176,7 @@ class ContextTruncator:
         self,
         messages: list[Message],
     ) -> list[Message]:
-        """对半砍策略，删除 50% 的消息"""
+        """Halve the number of messages, keeping the most recent ones."""
         if len(messages) <= 2:
             return messages
 
@@ -181,7 +188,7 @@ class ContextTruncator:
 
         truncated_non_system = non_system_messages[messages_to_delete:]
 
-        # 对齐到第一条 user
+        # Find the first user message
         index = next(
             (i for i, item in enumerate(truncated_non_system) if item.role == "user"),
             None,

From b07dbb3d26d41a5397f5b7ca9ed4335264db5e9b Mon Sep 17 00:00:00 2001
From: daniel5u <danielsuuuuuu@gmail.com>
Date: Fri, 20 Mar 2026 13:32:06 +0800
Subject: [PATCH 3/6] feat: add Kimi Coding Plan provider with Anthropic API
 compatibility (#6559)

* Add Kimi Code provider

* Add icon mapping for Kimi Code provider

* Clarify Kimi CodingPlan provider labeling

* Refine Kimi Code header handling

* modified docker compose

* fix: correct Kimi Coding Plan label and update API base URL

---------

Co-authored-by: Soulter <905617992@qq.com>
---
 astrbot/core/config/default.py                | 14 ++++
 astrbot/core/provider/manager.py              |  4 +
 .../core/provider/sources/anthropic_source.py | 33 +++++++-
 .../core/provider/sources/kimi_code_source.py | 27 +++++++
 compose-with-shipyard.yml                     |  5 +-
 .../i18n/locales/en-US/features/provider.json |  1 +
 .../i18n/locales/ru-RU/features/provider.json |  3 +-
 .../i18n/locales/zh-CN/features/provider.json |  1 +
 dashboard/src/utils/providerUtils.js          |  8 +-
 tests/test_anthropic_kimi_code_provider.py    | 81 +++++++++++++++++++
 10 files changed, 171 insertions(+), 6 deletions(-)
 create mode 100644 astrbot/core/provider/sources/kimi_code_source.py
 create mode 100644 tests/test_anthropic_kimi_code_provider.py

diff --git a/astrbot/core/config/default.py b/astrbot/core/config/default.py
index 58c172681..3842b1177 100644
--- a/astrbot/core/config/default.py
+++ b/astrbot/core/config/default.py
@@ -1112,6 +1112,20 @@ CONFIG_METADATA_2 = {
                         "api_base": "https://api.anthropic.com/v1",
                         "timeout": 120,
                         "proxy": "",
+                        "custom_headers": {},
+                        "anth_thinking_config": {"type": "", "budget": 0, "effort": ""},
+                    },
+                    "Kimi Coding Plan": {
+                        "id": "kimi-code",
+                        "provider": "kimi-code",
+                        "type": "kimi_code_chat_completion",
+                        "provider_type": "chat_completion",
+                        "enable": True,
+                        "key": [],
+                        "api_base": "https://api.kimi.com/coding/",
+                        "timeout": 120,
+                        "proxy": "",
+                        "custom_headers": {"User-Agent": "claude-code/0.1.0"},
                         "anth_thinking_config": {"type": "", "budget": 0, "effort": ""},
                     },
                     "Moonshot": {
diff --git a/astrbot/core/provider/manager.py b/astrbot/core/provider/manager.py
index 0df9f791a..761b84c9d 100644
--- a/astrbot/core/provider/manager.py
+++ b/astrbot/core/provider/manager.py
@@ -375,6 +375,10 @@ class ProviderManager:
                 from .sources.anthropic_source import (
                     ProviderAnthropic as ProviderAnthropic,
                 )
+            case "kimi_code_chat_completion":
+                from .sources.kimi_code_source import (
+                    ProviderKimiCode as ProviderKimiCode,
+                )
             case "googlegenai_chat_completion":
                 from .sources.gemini_source import (
                     ProviderGoogleGenAI as ProviderGoogleGenAI,
diff --git a/astrbot/core/provider/sources/anthropic_source.py b/astrbot/core/provider/sources/anthropic_source.py
index be70fdc74..203d0610f 100644
--- a/astrbot/core/provider/sources/anthropic_source.py
+++ b/astrbot/core/provider/sources/anthropic_source.py
@@ -16,7 +16,6 @@ from astrbot.core.provider.entities import LLMResponse, TokenUsage
 from astrbot.core.provider.func_tool_manager import ToolSet
 from astrbot.core.utils.io import download_image_by_url
 from astrbot.core.utils.network_utils import (
-    create_proxy_client,
     is_connection_error,
     log_connection_failure,
 )
@@ -29,6 +28,30 @@ from ..register import register_provider_adapter
     "Anthropic Claude API 提供商适配器",
 )
 class ProviderAnthropic(Provider):
+    @staticmethod
+    def _normalize_custom_headers(provider_config: dict) -> dict[str, str] | None:
+        custom_headers = provider_config.get("custom_headers", {})
+        if not isinstance(custom_headers, dict) or not custom_headers:
+            return None
+        normalized_headers: dict[str, str] = {}
+        for key, value in custom_headers.items():
+            normalized_headers[str(key)] = str(value)
+        return normalized_headers or None
+
+    @classmethod
+    def _resolve_custom_headers(
+        cls,
+        provider_config: dict,
+        *,
+        required_headers: dict[str, str] | None = None,
+    ) -> dict[str, str] | None:
+        merged_headers = cls._normalize_custom_headers(provider_config) or {}
+        if required_headers:
+            for header_name, header_value in required_headers.items():
+                if not merged_headers.get(header_name, "").strip():
+                    merged_headers[header_name] = header_value
+        return merged_headers or None
+
     def __init__(
         self,
         provider_config,
@@ -46,6 +69,7 @@ class ProviderAnthropic(Provider):
         if isinstance(self.timeout, str):
             self.timeout = int(self.timeout)
         self.thinking_config = provider_config.get("anth_thinking_config", {})
+        self.custom_headers = self._resolve_custom_headers(provider_config)
 
         if use_api_key:
             self._init_api_key(provider_config)
@@ -66,7 +90,12 @@ class ProviderAnthropic(Provider):
     def _create_http_client(self, provider_config: dict) -> httpx.AsyncClient | None:
         """创建带代理的 HTTP 客户端"""
         proxy = provider_config.get("proxy", "")
-        return create_proxy_client("Anthropic", proxy)
+        if proxy:
+            logger.info(f"[Anthropic] 使用代理: {proxy}")
+            return httpx.AsyncClient(proxy=proxy, headers=self.custom_headers)
+        if self.custom_headers:
+            return httpx.AsyncClient(headers=self.custom_headers)
+        return None
 
     def _apply_thinking_config(self, payloads: dict) -> None:
         thinking_type = self.thinking_config.get("type", "")
diff --git a/astrbot/core/provider/sources/kimi_code_source.py b/astrbot/core/provider/sources/kimi_code_source.py
new file mode 100644
index 000000000..02c200271
--- /dev/null
+++ b/astrbot/core/provider/sources/kimi_code_source.py
@@ -0,0 +1,27 @@
+from ..register import register_provider_adapter
+from .anthropic_source import ProviderAnthropic
+
+KIMI_CODE_API_BASE = "https://api.kimi.com/coding"
+KIMI_CODE_DEFAULT_MODEL = "kimi-for-coding"
+KIMI_CODE_USER_AGENT = "claude-code/0.1.0"
+
+
+@register_provider_adapter(
+    "kimi_code_chat_completion",
+    "Kimi Code Provider Adapter",
+)
+class ProviderKimiCode(ProviderAnthropic):
+    def __init__(
+        self,
+        provider_config: dict,
+        provider_settings: dict,
+    ) -> None:
+        merged_provider_config = dict(provider_config)
+        merged_provider_config.setdefault("api_base", KIMI_CODE_API_BASE)
+        merged_provider_config.setdefault("model", KIMI_CODE_DEFAULT_MODEL)
+        merged_provider_config["custom_headers"] = self._resolve_custom_headers(
+            merged_provider_config,
+            required_headers={"User-Agent": KIMI_CODE_USER_AGENT},
+        )
+
+        super().__init__(merged_provider_config, provider_settings)
diff --git a/compose-with-shipyard.yml b/compose-with-shipyard.yml
index 24ced5a95..7703293fa 100644
--- a/compose-with-shipyard.yml
+++ b/compose-with-shipyard.yml
@@ -4,7 +4,10 @@ version: '3.8'
 
 services:
   astrbot:
-    image: soulter/astrbot:latest
+    build:
+      context: .
+      dockerfile: Dockerfile
+    image: astrbot:kimi-code
     container_name: astrbot
     restart: always
     ports: # mappings description: https://github.com/AstrBotDevs/AstrBot/issues/497
diff --git a/dashboard/src/i18n/locales/en-US/features/provider.json b/dashboard/src/i18n/locales/en-US/features/provider.json
index 87e26e225..1dca1a4a5 100644
--- a/dashboard/src/i18n/locales/en-US/features/provider.json
+++ b/dashboard/src/i18n/locales/en-US/features/provider.json
@@ -21,6 +21,7 @@
     },
     "description": {
       "openai": "Supports all OpenAI API compatible providers.",
+      "kimi_code": "Dedicated Kimi CodingPlan / Kimi Code integration using a custom Anthropic API compatibility layer.",
       "default": ""
     }
   },
diff --git a/dashboard/src/i18n/locales/ru-RU/features/provider.json b/dashboard/src/i18n/locales/ru-RU/features/provider.json
index 1287bfbd4..9bc9f31ed 100644
--- a/dashboard/src/i18n/locales/ru-RU/features/provider.json
+++ b/dashboard/src/i18n/locales/ru-RU/features/provider.json
@@ -21,6 +21,7 @@
         },
         "description": {
             "openai": "Поддерживаются все провайдеры, совместимые с OpenAI API.",
+            "kimi_code": "Специальная интеграция Kimi CodingPlan / Kimi Code с отдельной совместимой адаптацией Anthropic API.",
             "vllm_rerank": "Также поддерживает Jina AI, Cohere, PPIO и другие.",
             "default": "Преобразование речи в текст"
         }
@@ -148,4 +149,4 @@
             "modelId": "ID модели"
         }
     }
-}
\ No newline at end of file
+}
diff --git a/dashboard/src/i18n/locales/zh-CN/features/provider.json b/dashboard/src/i18n/locales/zh-CN/features/provider.json
index d3f5d95b1..7a8a1a4d8 100644
--- a/dashboard/src/i18n/locales/zh-CN/features/provider.json
+++ b/dashboard/src/i18n/locales/zh-CN/features/provider.json
@@ -21,6 +21,7 @@
     },
     "description": {
       "openai": "也支持所有兼容 OpenAI API 的模型提供商。",
+      "kimi_code": "Kimi 的 CodingPlan / Kimi Code 专用接入，使用特殊的 Anthropic API 兼容适配。",
       "vllm_rerank": "也支持 Jina AI, Cohere, PPIO 等提供商。",
       "default": ""
     }
diff --git a/dashboard/src/utils/providerUtils.js b/dashboard/src/utils/providerUtils.js
index 4bfe3ea6e..7c992f531 100644
--- a/dashboard/src/utils/providerUtils.js
+++ b/dashboard/src/utils/providerUtils.js
@@ -21,6 +21,8 @@ export function getProviderIcon(type) {
     'nvidia': 'https://cdn.jsdelivr.net/npm/@lobehub/icons-static-svg@latest/icons/nvidia-color.svg',
     'siliconflow': 'https://cdn.jsdelivr.net/npm/@lobehub/icons-static-svg@latest/icons/siliconcloud.svg',
     'moonshot': 'https://cdn.jsdelivr.net/npm/@lobehub/icons-static-svg@latest/icons/kimi.svg',
+    'kimi': 'https://cdn.jsdelivr.net/npm/@lobehub/icons-static-svg@latest/icons/kimi.svg',
+    'kimi-code': 'https://cdn.jsdelivr.net/npm/@lobehub/icons-static-svg@latest/icons/kimi.svg',
     'ppio': 'https://cdn.jsdelivr.net/npm/@lobehub/icons-static-svg@latest/icons/ppio.svg',
     'dify': 'https://cdn.jsdelivr.net/npm/@lobehub/icons-static-svg@latest/icons/dify-color.svg',
     "coze": "https://cdn.jsdelivr.net/npm/@lobehub/icons-static-svg@1.66.0/icons/coze.svg",
@@ -50,9 +52,11 @@ export function getProviderIcon(type) {
  * @returns {string} 提供商描述
  */
 export function getProviderDescription(template, name, tm) {
-  if (name == 'OpenAI') {
+  if (name === 'OpenAI') {
     return tm('providers.description.openai', { type: template.type });
-  } else if (name == 'vLLM Rerank') {
+  } else if (template.provider === 'kimi-code') {
+    return tm('providers.description.kimi_code');
+  } else if (name === 'vLLM Rerank') {
     return tm('providers.description.vllm_rerank', { type: template.type });
   }
   return tm('providers.description.default', { type: template.type });
diff --git a/tests/test_anthropic_kimi_code_provider.py b/tests/test_anthropic_kimi_code_provider.py
new file mode 100644
index 000000000..ea22e376c
--- /dev/null
+++ b/tests/test_anthropic_kimi_code_provider.py
@@ -0,0 +1,81 @@
+import httpx
+
+import astrbot.core.provider.sources.anthropic_source as anthropic_source
+import astrbot.core.provider.sources.kimi_code_source as kimi_code_source
+
+
+class _FakeAsyncAnthropic:
+    def __init__(self, **kwargs):
+        self.kwargs = kwargs
+
+    async def close(self):
+        return None
+
+
+def test_anthropic_provider_injects_custom_headers_into_http_client(monkeypatch):
+    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"],
+            "custom_headers": {
+                "User-Agent": "custom-agent/1.0",
+                "X-Test-Header": 123,
+            },
+        },
+        provider_settings={},
+    )
+
+    assert provider.custom_headers == {
+        "User-Agent": "custom-agent/1.0",
+        "X-Test-Header": "123",
+    }
+    assert isinstance(provider.client.kwargs["http_client"], httpx.AsyncClient)
+    assert provider.client.kwargs["http_client"].headers["User-Agent"] == "custom-agent/1.0"
+    assert provider.client.kwargs["http_client"].headers["X-Test-Header"] == "123"
+
+
+def test_kimi_code_provider_sets_defaults_and_preserves_custom_headers(monkeypatch):
+    monkeypatch.setattr(anthropic_source, "AsyncAnthropic", _FakeAsyncAnthropic)
+
+    provider = kimi_code_source.ProviderKimiCode(
+        provider_config={
+            "id": "kimi-code",
+            "type": "kimi_code_chat_completion",
+            "key": ["test-key"],
+            "custom_headers": {"X-Trace-Id": "trace-1"},
+        },
+        provider_settings={},
+    )
+
+    assert provider.base_url == kimi_code_source.KIMI_CODE_API_BASE
+    assert provider.get_model() == kimi_code_source.KIMI_CODE_DEFAULT_MODEL
+    assert provider.custom_headers == {
+        "User-Agent": kimi_code_source.KIMI_CODE_USER_AGENT,
+        "X-Trace-Id": "trace-1",
+    }
+    assert provider.client.kwargs["http_client"].headers["User-Agent"] == (
+        kimi_code_source.KIMI_CODE_USER_AGENT
+    )
+    assert provider.client.kwargs["http_client"].headers["X-Trace-Id"] == "trace-1"
+
+
+def test_kimi_code_provider_restores_required_user_agent_when_blank(monkeypatch):
+    monkeypatch.setattr(anthropic_source, "AsyncAnthropic", _FakeAsyncAnthropic)
+
+    provider = kimi_code_source.ProviderKimiCode(
+        provider_config={
+            "id": "kimi-code",
+            "type": "kimi_code_chat_completion",
+            "key": ["test-key"],
+            "custom_headers": {"User-Agent": "   "},
+        },
+        provider_settings={},
+    )
+
+    assert provider.custom_headers == {
+        "User-Agent": kimi_code_source.KIMI_CODE_USER_AGENT,
+    }

From 6a42ad793480c9836cab75e3970f19103e8bf78f Mon Sep 17 00:00:00 2001
From: Soulter <37870767+Soulter@users.noreply.github.com>
Date: Fri, 20 Mar 2026 13:42:58 +0800
Subject: [PATCH 4/6] fix(openai): improve logging for proxy and API base
 configuration (#6669)

fix: #6558
---
 .../provider/sources/openai_embedding_source.py    | 14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)

diff --git a/astrbot/core/provider/sources/openai_embedding_source.py b/astrbot/core/provider/sources/openai_embedding_source.py
index 04397b182..2b62d865c 100644
--- a/astrbot/core/provider/sources/openai_embedding_source.py
+++ b/astrbot/core/provider/sources/openai_embedding_source.py
@@ -19,17 +19,15 @@ class OpenAIEmbeddingProvider(EmbeddingProvider):
         self.provider_config = provider_config
         self.provider_settings = provider_settings
         proxy = provider_config.get("proxy", "")
+        provider_id = provider_config.get("id", "unknown_id")
         http_client = None
         if proxy:
-            logger.info(f"[OpenAI Embedding] 使用代理: {proxy}")
+            logger.info(f"[OpenAI Embedding] {provider_id} Using proxy: {proxy}")
             http_client = httpx.AsyncClient(proxy=proxy)
-        api_base = provider_config.get("embedding_api_base", "").strip()
-        if not api_base:
-            api_base = "https://api.openai.com/v1"
-        else:
-            api_base = api_base.removesuffix("/")
-            if not api_base.endswith("/v1"):
-                api_base = f"{api_base}/v1"
+        api_base = provider_config.get(
+            "embedding_api_base", "https://api.openai.com/v1"
+        ).strip()
+        logger.info(f"[OpenAI Embedding] {provider_id} Using API Base: {api_base}")
         self.client = AsyncOpenAI(
             api_key=provider_config.get("embedding_api_key"),
             base_url=api_base,

From 735bd43648b094b62dd56eabc5d1cc7b2faf7760 Mon Sep 17 00:00:00 2001
From: letr <123731298+letr007@users.noreply.github.com>
Date: Fri, 20 Mar 2026 13:45:21 +0800
Subject: [PATCH 5/6] fix(dashboard): simplify persona selector layout for
 mobile screens (#5907)

---
 .../folder/BaseFolderItemSelector.vue         | 93 ++++++++++++++++---
 1 file changed, 81 insertions(+), 12 deletions(-)

diff --git a/dashboard/src/components/folder/BaseFolderItemSelector.vue b/dashboard/src/components/folder/BaseFolderItemSelector.vue
index 4b4b4f1c4..9e5c5130f 100644
--- a/dashboard/src/components/folder/BaseFolderItemSelector.vue
+++ b/dashboard/src/components/folder/BaseFolderItemSelector.vue
@@ -14,19 +14,23 @@
         </div>
 
         <!-- 选择对话框 -->
-        <v-dialog v-model="dialog" max-width="1000px" min-width="800px">
+        <v-dialog
+            v-model="dialog"
+            :max-width="isCompactLayout ? '96vw' : '1000px'"
+            :min-width="isCompactLayout ? undefined : '800px'"
+        >
             <v-card class="selector-dialog-card">
-                <v-card-title class="dialog-title d-flex align-center py-4 px-5">
+                <v-card-title class="dialog-title d-flex align-center" :class="isCompactLayout ? 'py-3 px-4' : 'py-4 px-5'">
                     <v-icon class="mr-3" color="primary">mdi-account-circle</v-icon>
                     <span>{{ labels.dialogTitle || '选择项目' }}</span>
                 </v-card-title>
 
                 <v-divider />
 
-                <v-card-text class="pa-0" style="height: 600px; max-height: 80vh; overflow: hidden;">
+                <v-card-text class="pa-0 selector-content">
                     <div class="selector-layout">
                         <!-- 左侧文件夹树 -->
-                        <div class="folder-sidebar">
+                        <div v-if="!isCompactLayout" class="folder-sidebar">
                             <div class="sidebar-header pa-3 pb-2">
                                 <span class="text-caption text-medium-emphasis font-weight-medium">
                                     <v-icon size="small" class="mr-1">mdi-folder-multiple</v-icon>
@@ -58,6 +62,20 @@
 
                         <!-- 右侧项目列表 -->
                         <div class="items-panel">
+                            <div v-if="isCompactLayout" class="mobile-folder-bar px-4 py-2">
+                                <v-btn icon="mdi-arrow-left" size="small" variant="text"
+                                    :disabled="currentFolderId === null" @click="navigateToParentFolder" />
+                                <v-btn size="small" variant="tonal" color="primary" prepend-icon="mdi-home"
+                                    @click="navigateToFolder(null)">
+                                    {{ labels.rootFolder || '根目录' }}
+                                </v-btn>
+                                <span class="text-caption text-medium-emphasis text-truncate mobile-folder-label">
+                                    {{ currentFolderLabel }}
+                                </span>
+                            </div>
+
+                            <v-divider v-if="isCompactLayout" />
+
                             <!-- 面包屑导航 -->
                             <div class="breadcrumb-bar px-4 py-3">
                                 <v-breadcrumbs :items="breadcrumbItems" density="compact" class="pa-0">
@@ -245,6 +263,18 @@ export default defineComponent({
         };
     },
     computed: {
+        isCompactLayout(): boolean {
+            return this.$vuetify.display.smAndDown;
+        },
+
+        currentFolderLabel(): string {
+            if (this.currentFolderId === null) {
+                return this.labels.rootFolder || '根目录';
+            }
+            const currentFolder = this.breadcrumbPath[this.breadcrumbPath.length - 1];
+            return currentFolder?.name || this.labels.rootFolder || '根目录';
+        },
+
         displayValue(): string {
             if (this.displayValueFormatter) {
                 return this.displayValueFormatter(this.modelValue);
@@ -332,6 +362,20 @@ export default defineComponent({
             this.$emit('navigate', folderId);
         },
 
+        navigateToParentFolder() {
+            if (this.currentFolderId === null) {
+                return;
+            }
+
+            if (this.breadcrumbPath.length <= 1) {
+                this.navigateToFolder(null);
+                return;
+            }
+
+            const parent = this.breadcrumbPath[this.breadcrumbPath.length - 2];
+            this.navigateToFolder(parent?.folder_id ?? null);
+        },
+
         findFolderInTree(folderId: string): FolderTreeNode | null {
             const findNode = (nodes: FolderTreeNode[]): FolderTreeNode | null => {
                 for (const node of nodes) {
@@ -414,6 +458,13 @@ export default defineComponent({
 .selector-layout {
     display: flex;
     height: 100%;
+    min-width: 0;
+}
+
+.selector-content {
+    height: 600px;
+    max-height: 80vh;
+    overflow: hidden;
 }
 
 .folder-sidebar {
@@ -450,6 +501,18 @@ export default defineComponent({
 
 .items-content {
     background-color: transparent;
+    min-width: 0;
+}
+
+.mobile-folder-bar {
+    display: flex;
+    align-items: center;
+    gap: 8px;
+}
+
+.mobile-folder-label {
+    min-width: 0;
+    flex: 1;
 }
 
 .tree-list {
@@ -517,22 +580,28 @@ export default defineComponent({
     background-color: rgba(var(--v-theme-primary), 0.08);
 }
 
-@media (max-width: 600px) {
+@media (max-width: 960px) {
     .selector-layout {
         flex-direction: column;
         height: auto;
-        max-height: 500px;
+        max-height: none;
     }
 
-    .folder-sidebar {
-        width: 100%;
-        border-right: none;
-        border-bottom: 1px solid rgba(var(--v-border-color), var(--v-border-opacity));
-        max-height: 150px;
+    .selector-content {
+        max-height: 76vh;
     }
 
     .items-list {
-        max-height: 300px;
+        min-height: 0;
+    }
+
+    .breadcrumb-bar {
+        overflow-x: auto;
+    }
+
+    .breadcrumb-bar :deep(.v-breadcrumbs) {
+        flex-wrap: nowrap;
+        min-width: max-content;
     }
 }
 </style>

From 39131d2e12bd163827d21dcd33bf3bf6ef40fcbb Mon Sep 17 00:00:00 2001
From: Yufeng He <40085740+he-yufeng@users.noreply.github.com>
Date: Fri, 20 Mar 2026 13:53:56 +0800
Subject: [PATCH 6/6] fix: Follow-up logic persists after /stop trigger (#6656)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

/stop 设置 agent_stop_requested 标记，但 runner 直到当前工具调用
超时才从 _ACTIVE_AGENT_RUNNERS 注销。在此窗口期内，用户发的新消息
被 try_capture_follow_up() 当作 follow-up 吞掉。

在 follow-up 捕获前检查 stop 标记：一旦用户请求停止，就不再把后续
消息注入到正在终止的 agent 上下文中。

Fixes #6626
---
 astrbot/core/pipeline/process_stage/follow_up.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/astrbot/core/pipeline/process_stage/follow_up.py b/astrbot/core/pipeline/process_stage/follow_up.py
index 6c1a4fa06..79ec16a85 100644
--- a/astrbot/core/pipeline/process_stage/follow_up.py
+++ b/astrbot/core/pipeline/process_stage/follow_up.py
@@ -172,6 +172,9 @@ def try_capture_follow_up(event: AstrMessageEvent) -> FollowUpCapture | None:
     if not active_sender_id or active_sender_id != sender_id:
         return None
 
+    if runner_event.get_extra("agent_stop_requested"):
+        return None
+
     ticket = runner.follow_up(message_text=_event_follow_up_text(event))
     if not ticket:
         return None