AstrBot 4.13.2__py3-none-any.whl → 4.14.1__py3-none-any.whl

astrbot/core/pipeline/process_stage/method/agent_sub_stages/internal.py

@@ -1,55 +1,36 @@
 """本地 Agent 模式的 LLM 调用 Stage"""
 
 import asyncio
-import json
-import os
+import base64
 from collections.abc import AsyncGenerator
+from dataclasses import replace
 
 from astrbot.core import logger
-from astrbot.core.agent.message import Message, TextPart
+from astrbot.core.agent.message import Message
 from astrbot.core.agent.response import AgentStats
-from astrbot.core.agent.tool import ToolSet
-from astrbot.core.astr_agent_context import AstrAgentContext
-from astrbot.core.conversation_mgr import Conversation
-from astrbot.core.message.components import File, Image, Reply
+from astrbot.core.astr_main_agent import (
+    MainAgentBuildConfig,
+    MainAgentBuildResult,
+    build_main_agent,
+)
+from astrbot.core.message.components import File, Image
 from astrbot.core.message.message_event_result import (
     MessageChain,
     MessageEventResult,
     ResultContentType,
 )
 from astrbot.core.platform.astr_message_event import AstrMessageEvent
-from astrbot.core.provider import Provider
 from astrbot.core.provider.entities import (
     LLMResponse,
     ProviderRequest,
 )
-from astrbot.core.star.star_handler import EventType, star_map
-from astrbot.core.utils.file_extract import extract_file_moonshotai
-from astrbot.core.utils.llm_metadata import LLM_METADATAS
+from astrbot.core.star.star_handler import EventType
 from astrbot.core.utils.metrics import Metric
 from astrbot.core.utils.session_lock import session_lock_manager
 
-from .....astr_agent_context import AgentContextWrapper
-from .....astr_agent_hooks import MAIN_AGENT_HOOKS
-from .....astr_agent_run_util import AgentRunner, run_agent, run_live_agent
-from .....astr_agent_tool_exec import FunctionToolExecutor
+from .....astr_agent_run_util import run_agent, run_live_agent
 from ....context import PipelineContext, call_event_hook
 from ...stage import Stage
-from ...utils import (
-    CHATUI_EXTRA_PROMPT,
-    EXECUTE_SHELL_TOOL,
-    FILE_DOWNLOAD_TOOL,
-    FILE_UPLOAD_TOOL,
-    KNOWLEDGE_BASE_QUERY_TOOL,
-    LIVE_MODE_SYSTEM_PROMPT,
-    LLM_SAFETY_MODE_SYSTEM_PROMPT,
-    PYTHON_TOOL,
-    SANDBOX_MODE_PROMPT,
-    TOOL_CALL_PROMPT,
-    TOOL_CALL_PROMPT_SKILLS_LIKE_MODE,
-    decoded_blocked,
-    retrieve_knowledge_base,
-)
 
 
 class InternalAgentSubStage(Stage):
@@ -111,419 +92,49 @@ class InternalAgentSubStage(Stage):
             "safety_mode_strategy", "system_prompt"
         )
 
+        self.computer_use_runtime = settings.get("computer_use_runtime")
         self.sandbox_cfg = settings.get("sandbox", {})
 
-        self.conv_manager = ctx.plugin_manager.context.conversation_manager
+        # Proactive capability configuration
+        proactive_cfg = settings.get("proactive_capability", {})
+        self.add_cron_tools = proactive_cfg.get("add_cron_tools", True)
 
-    def _select_provider(self, event: AstrMessageEvent):
-        """选择使用的 LLM 提供商"""
-        sel_provider = event.get_extra("selected_provider")
-        _ctx = self.ctx.plugin_manager.context
-        if sel_provider and isinstance(sel_provider, str):
-            provider = _ctx.get_provider_by_id(sel_provider)
-            if not provider:
-                logger.error(f"未找到指定的提供商: {sel_provider}。")
-            return provider
-        try:
-            prov = _ctx.get_using_provider(umo=event.unified_msg_origin)
-        except ValueError as e:
-            logger.error(f"Error occurred while selecting provider: {e}")
-            return None
-        return prov
-
-    async def _get_session_conv(self, event: AstrMessageEvent) -> Conversation:
-        umo = event.unified_msg_origin
-        conv_mgr = self.conv_manager
-
-        # 获取对话上下文
-        cid = await conv_mgr.get_curr_conversation_id(umo)
-        if not cid:
-            cid = await conv_mgr.new_conversation(umo, event.get_platform_id())
-        conversation = await conv_mgr.get_conversation(umo, cid)
-        if not conversation:
-            cid = await conv_mgr.new_conversation(umo, event.get_platform_id())
-            conversation = await conv_mgr.get_conversation(umo, cid)
-        if not conversation:
-            raise RuntimeError("无法创建新的对话。")
-        return conversation
-
-    async def _apply_kb(
-        self,
-        event: AstrMessageEvent,
-        req: ProviderRequest,
-    ):
-        """Apply knowledge base context to the provider request"""
-        if not self.kb_agentic_mode:
-            if req.prompt is None:
-                return
-            try:
-                kb_result = await retrieve_knowledge_base(
-                    query=req.prompt,
-                    umo=event.unified_msg_origin,
-                    context=self.ctx.plugin_manager.context,
-                )
-                if not kb_result:
-                    return
-                if req.system_prompt is not None:
-                    req.system_prompt += (
-                        f"\n\n[Related Knowledge Base Results]:\n{kb_result}"
-                    )
-            except Exception as e:
-                logger.error(f"Error occurred while retrieving knowledge base: {e}")
-        else:
-            if req.func_tool is None:
-                req.func_tool = ToolSet()
-            req.func_tool.add_tool(KNOWLEDGE_BASE_QUERY_TOOL)
-
-    async def _apply_file_extract(
-        self,
-        event: AstrMessageEvent,
-        req: ProviderRequest,
-    ):
-        """Apply file extract to the provider request"""
-        file_paths = []
-        file_names = []
-        for comp in event.message_obj.message:
-            if isinstance(comp, File):
-                file_paths.append(await comp.get_file())
-                file_names.append(comp.name)
-            elif isinstance(comp, Reply) and comp.chain:
-                for reply_comp in comp.chain:
-                    if isinstance(reply_comp, File):
-                        file_paths.append(await reply_comp.get_file())
-                        file_names.append(reply_comp.name)
-        if not file_paths:
-            return
-        if not req.prompt:
-            req.prompt = "总结一下文件里面讲了什么？"
-        if self.file_extract_prov == "moonshotai":
-            if not self.file_extract_msh_api_key:
-                logger.error("Moonshot AI API key for file extract is not set")
-                return
-            file_contents = await asyncio.gather(
-                *[
-                    extract_file_moonshotai(file_path, self.file_extract_msh_api_key)
-                    for file_path in file_paths
-                ]
-            )
-        else:
-            logger.error(f"Unsupported file extract provider: {self.file_extract_prov}")
-            return
-
-        # add file extract results to contexts
-        for file_content, file_name in zip(file_contents, file_names):
-            req.contexts.append(
-                {
-                    "role": "system",
-                    "content": f"File Extract Results of user uploaded files:\n{file_content}\nFile Name: {file_name or 'Unknown'}",
-                },
-            )
-
-    def _modalities_fix(
-        self,
-        provider: Provider,
-        req: ProviderRequest,
-    ):
-        """检查提供商的模态能力，清理请求中的不支持内容"""
-        if req.image_urls:
-            provider_cfg = provider.provider_config.get("modalities", ["image"])
-            if "image" not in provider_cfg:
-                logger.debug(
-                    f"用户设置提供商 {provider} 不支持图像，将图像替换为占位符。"
-                )
-                # 为每个图片添加占位符到 prompt
-                image_count = len(req.image_urls)
-                placeholder = " ".join(["[图片]"] * image_count)
-                if req.prompt:
-                    req.prompt = f"{placeholder} {req.prompt}"
-                else:
-                    req.prompt = placeholder
-                req.image_urls = []
-        if req.func_tool:
-            provider_cfg = provider.provider_config.get("modalities", ["tool_use"])
-            # 如果模型不支持工具使用，但请求中包含工具列表，则清空。
-            if "tool_use" not in provider_cfg:
-                logger.debug(
-                    f"用户设置提供商 {provider} 不支持工具使用，清空工具列表。",
-                )
-                req.func_tool = None
-
-    def _sanitize_context_by_modalities(
-        self,
-        provider: Provider,
-        req: ProviderRequest,
-    ) -> None:
-        """Sanitize `req.contexts` (including history) by current provider modalities."""
-        if not self.sanitize_context_by_modalities:
-            return
-
-        if not isinstance(req.contexts, list) or not req.contexts:
-            return
-
-        modalities = provider.provider_config.get("modalities", None)
-        # if modalities is not configured, do not sanitize.
-        if not modalities or not isinstance(modalities, list):
-            return
-
-        supports_image = bool("image" in modalities)
-        supports_tool_use = bool("tool_use" in modalities)
-
-        if supports_image and supports_tool_use:
-            return
-
-        sanitized_contexts: list[dict] = []
-        removed_image_blocks = 0
-        removed_tool_messages = 0
-        removed_tool_calls = 0
-
-        for msg in req.contexts:
-            if not isinstance(msg, dict):
-                continue
-
-            role = msg.get("role")
-            if not role:
-                continue
-
-            new_msg: dict = msg
-
-            # tool_use sanitize
-            if not supports_tool_use:
-                if role == "tool":
-                    # tool response block
-                    removed_tool_messages += 1
-                    continue
-                if role == "assistant" and "tool_calls" in new_msg:
-                    # assistant message with tool calls
-                    if "tool_calls" in new_msg:
-                        removed_tool_calls += 1
-                    new_msg.pop("tool_calls", None)
-                    new_msg.pop("tool_call_id", None)
-
-            # image sanitize
-            if not supports_image:
-                content = new_msg.get("content")
-                if isinstance(content, list):
-                    filtered_parts: list = []
-                    removed_any_image = False
-                    for part in content:
-                        if isinstance(part, dict):
-                            part_type = str(part.get("type", "")).lower()
-                            if part_type in {"image_url", "image"}:
-                                removed_any_image = True
-                                removed_image_blocks += 1
-                                continue
-                        filtered_parts.append(part)
-
-                    if removed_any_image:
-                        new_msg["content"] = filtered_parts
-
-            # drop empty assistant messages (e.g. only tool_calls without content)
-            if role == "assistant":
-                content = new_msg.get("content")
-                has_tool_calls = bool(new_msg.get("tool_calls"))
-                if not has_tool_calls:
-                    if not content:
-                        continue
-                    if isinstance(content, str) and not content.strip():
-                        continue
-
-            sanitized_contexts.append(new_msg)
-
-        if removed_image_blocks or removed_tool_messages or removed_tool_calls:
-            logger.debug(
-                "sanitize_context_by_modalities applied: "
-                f"removed_image_blocks={removed_image_blocks}, "
-                f"removed_tool_messages={removed_tool_messages}, "
-                f"removed_tool_calls={removed_tool_calls}"
-            )
-
-        req.contexts = sanitized_contexts
-
-    def _plugin_tool_fix(
-        self,
-        event: AstrMessageEvent,
-        req: ProviderRequest,
-    ):
-        """根据事件中的插件设置，过滤请求中的工具列表"""
-        if event.plugins_name is not None and req.func_tool:
-            new_tool_set = ToolSet()
-            for tool in req.func_tool.tools:
-                mp = tool.handler_module_path
-                if not mp:
-                    continue
-                plugin = star_map.get(mp)
-                if not plugin:
-                    continue
-                if plugin.name in event.plugins_name or plugin.reserved:
-                    new_tool_set.add_tool(tool)
-            req.func_tool = new_tool_set
-
-    async def _handle_webchat(
-        self,
-        event: AstrMessageEvent,
-        req: ProviderRequest,
-        prov: Provider,
-    ):
-        """处理 WebChat 平台的特殊情况，包括第一次 LLM 对话时总结对话内容生成 title"""
-        from astrbot.core import db_helper
-
-        chatui_session_id = event.session_id.split("!")[-1]
-        user_prompt = req.prompt
-
-        session = await db_helper.get_platform_session_by_id(chatui_session_id)
-
-        if (
-            not user_prompt
-            or not chatui_session_id
-            or not session
-            or session.display_name
-        ):
-            return
-
-        llm_resp = await prov.text_chat(
-            system_prompt=(
-                "You are a conversation title generator. "
-                "Generate a concise title in the same language as the user’s input, "
-                "no more than 10 words, capturing only the core topic."
-                "If the input is a greeting, small talk, or has no clear topic, "
-                "(e.g., “hi”, “hello”, “haha”), return <None>. "
-                "Output only the title itself or <None>, with no explanations."
-            ),
-            prompt=(
-                f"Generate a concise title for the following user query:\n{user_prompt}"
-            ),
-        )
-        if llm_resp and llm_resp.completion_text:
-            title = llm_resp.completion_text.strip()
-            if not title or "<None>" in title:
-                return
-            logger.info(
-                f"Generated chatui title for session {chatui_session_id}: {title}"
-            )
-            await db_helper.update_platform_session(
-                session_id=chatui_session_id,
-                display_name=title,
-            )
-
-    async def _save_to_history(
-        self,
-        event: AstrMessageEvent,
-        req: ProviderRequest,
-        llm_response: LLMResponse | None,
-        all_messages: list[Message],
-        runner_stats: AgentStats | None,
-    ):
-        if (
-            not req
-            or not req.conversation
-            or not llm_response
-            or llm_response.role != "assistant"
-        ):
-            return
-
-        if not llm_response.completion_text and not req.tool_calls_result:
-            logger.debug("LLM 响应为空，不保存记录。")
-            return
-
-        # using agent context messages to save to history
-        message_to_save = []
-        skipped_initial_system = False
-        for message in all_messages:
-            if message.role == "system" and not skipped_initial_system:
-                skipped_initial_system = True
-                continue  # skip first system message
-            if message.role in ["assistant", "user"] and getattr(
-                message, "_no_save", None
-            ):
-                # we do not save user and assistant messages that are marked as _no_save
-                continue
-            message_to_save.append(message.model_dump())
-
-        # get token usage from agent runner stats
-        token_usage = None
-        if runner_stats:
-            token_usage = runner_stats.token_usage.total
-
-        await self.conv_manager.update_conversation(
-            event.unified_msg_origin,
-            req.conversation.cid,
-            history=message_to_save,
-            token_usage=token_usage,
-        )
+        self.conv_manager = ctx.plugin_manager.context.conversation_manager
 
-    def _get_compress_provider(self) -> Provider | None:
-        if not self.llm_compress_provider_id:
-            return None
-        if self.context_limit_reached_strategy != "llm_compress":
-            return None
-        provider = self.ctx.plugin_manager.context.get_provider_by_id(
-            self.llm_compress_provider_id,
+        self.main_agent_cfg = MainAgentBuildConfig(
+            tool_call_timeout=self.tool_call_timeout,
+            tool_schema_mode=self.tool_schema_mode,
+            sanitize_context_by_modalities=self.sanitize_context_by_modalities,
+            kb_agentic_mode=self.kb_agentic_mode,
+            file_extract_enabled=self.file_extract_enabled,
+            file_extract_prov=self.file_extract_prov,
+            file_extract_msh_api_key=self.file_extract_msh_api_key,
+            context_limit_reached_strategy=self.context_limit_reached_strategy,
+            llm_compress_instruction=self.llm_compress_instruction,
+            llm_compress_keep_recent=self.llm_compress_keep_recent,
+            llm_compress_provider_id=self.llm_compress_provider_id,
+            max_context_length=self.max_context_length,
+            dequeue_context_length=self.dequeue_context_length,
+            llm_safety_mode=self.llm_safety_mode,
+            safety_mode_strategy=self.safety_mode_strategy,
+            computer_use_runtime=self.computer_use_runtime,
+            sandbox_cfg=self.sandbox_cfg,
+            add_cron_tools=self.add_cron_tools,
+            provider_settings=settings,
+            subagent_orchestrator=conf.get("subagent_orchestrator", {}),
+            timezone=self.ctx.plugin_manager.context.get_config().get("timezone"),
         )
-        if provider is None:
-            logger.warning(
-                f"未找到指定的上下文压缩模型 {self.llm_compress_provider_id}，将跳过压缩。",
-            )
-            return None
-        if not isinstance(provider, Provider):
-            logger.warning(
-                f"指定的上下文压缩模型 {self.llm_compress_provider_id} 不是对话模型，将跳过压缩。"
-            )
-            return None
-        return provider
-
-    def _apply_llm_safety_mode(self, req: ProviderRequest) -> None:
-        """Apply LLM safety mode to the provider request."""
-        if self.safety_mode_strategy == "system_prompt":
-            req.system_prompt = (
-                f"{LLM_SAFETY_MODE_SYSTEM_PROMPT}\n\n{req.system_prompt or ''}"
-            )
-        else:
-            logger.warning(
-                f"Unsupported llm_safety_mode strategy: {self.safety_mode_strategy}.",
-            )
-
-    def _apply_sandbox_tools(self, req: ProviderRequest, session_id: str) -> None:
-        """Add sandbox tools to the provider request."""
-        if req.func_tool is None:
-            req.func_tool = ToolSet()
-        if self.sandbox_cfg.get("booter") == "shipyard":
-            ep = self.sandbox_cfg.get("shipyard_endpoint", "")
-            at = self.sandbox_cfg.get("shipyard_access_token", "")
-            if not ep or not at:
-                logger.error("Shipyard sandbox configuration is incomplete.")
-                return
-            os.environ["SHIPYARD_ENDPOINT"] = ep
-            os.environ["SHIPYARD_ACCESS_TOKEN"] = at
-        req.func_tool.add_tool(EXECUTE_SHELL_TOOL)
-        req.func_tool.add_tool(PYTHON_TOOL)
-        req.func_tool.add_tool(FILE_UPLOAD_TOOL)
-        req.func_tool.add_tool(FILE_DOWNLOAD_TOOL)
-        req.system_prompt += f"\n{SANDBOX_MODE_PROMPT}\n"
 
     async def process(
         self, event: AstrMessageEvent, provider_wake_prefix: str
     ) -> AsyncGenerator[None, None]:
-        req: ProviderRequest | None = None
-
         try:
-            provider = self._select_provider(event)
-            if provider is None:
-                logger.info("未找到任何对话模型（提供商），跳过 LLM 请求处理。")
-                return
-            if not isinstance(provider, Provider):
-                logger.error(
-                    f"选择的提供商类型无效({type(provider)})，跳过 LLM 请求处理。"
-                )
-                return
-
             streaming_response = self.streaming_response
             if (enable_streaming := event.get_extra("enable_streaming")) is not None:
                 streaming_response = bool(enable_streaming)
 
-            # 检查消息内容是否有效，避免空消息触发钩子
             has_provider_request = event.get_extra("provider_request") is not None
             has_valid_message = bool(event.message_str and event.message_str.strip())
-            # 检查是否有图片或其他媒体内容
             has_media_content = any(
                 isinstance(comp, Image | File) for comp in event.message_obj.message
             )
@@ -536,161 +147,50 @@ class InternalAgentSubStage(Stage):
                 logger.debug("skip llm request: empty message and no provider_request")
                 return
 
-            api_base = provider.provider_config.get("api_base", "")
-            for host in decoded_blocked:
-                if host in api_base:
-                    logger.error(
-                        f"Provider API base {api_base} is blocked due to security reasons. Please use another ai provider."
-                    )
-                    return
-
             logger.debug("ready to request llm provider")
 
-            # 通知等待调用 LLM（在获取锁之前）
             await call_event_hook(event, EventType.OnWaitingLLMRequestEvent)
 
             async with session_lock_manager.acquire_lock(event.unified_msg_origin):
                 logger.debug("acquired session lock for llm request")
-                if event.get_extra("provider_request"):
-                    req = event.get_extra("provider_request")
-                    assert isinstance(req, ProviderRequest), (
-                        "provider_request 必须是 ProviderRequest 类型。"
-                    )
 
-                    if req.conversation:
-                        req.contexts = json.loads(req.conversation.history)
-
-                else:
-                    req = ProviderRequest()
-                    req.prompt = ""
-                    req.image_urls = []
-                    if sel_model := event.get_extra("selected_model"):
-                        req.model = sel_model
-                    if provider_wake_prefix and not event.message_str.startswith(
-                        provider_wake_prefix
-                    ):
-                        return
-
-                    req.prompt = event.message_str[len(provider_wake_prefix) :]
-                    # func_tool selection 现在已经转移到 astrbot/builtin_stars/astrbot 插件中进行选择。
-                    # req.func_tool = self.ctx.plugin_manager.context.get_llm_tool_manager()
-                    for comp in event.message_obj.message:
-                        if isinstance(comp, Image):
-                            image_path = await comp.convert_to_file_path()
-                            req.image_urls.append(image_path)
-
-                            req.extra_user_content_parts.append(
-                                TextPart(text=f"[Image Attachment: path {image_path}]")
-                            )
-                        elif isinstance(comp, File):
-                            file_path = await comp.get_file()
-                            file_name = comp.name or os.path.basename(file_path)
-                            req.extra_user_content_parts.append(
-                                TextPart(
-                                    text=f"[File Attachment: name {file_name}, path {file_path}]"
-                                )
-                            )
-
-                    conversation = await self._get_session_conv(event)
-                    req.conversation = conversation
-                    req.contexts = json.loads(conversation.history)
-
-                    event.set_extra("provider_request", req)
-
-                # fix contexts json str
-                if isinstance(req.contexts, str):
-                    req.contexts = json.loads(req.contexts)
-
-                # apply file extract
-                if self.file_extract_enabled:
-                    try:
-                        await self._apply_file_extract(event, req)
-                    except Exception as e:
-                        logger.error(f"Error occurred while applying file extract: {e}")
+                build_cfg = replace(
+                    self.main_agent_cfg,
+                    provider_wake_prefix=provider_wake_prefix,
+                    streaming_response=streaming_response,
+                )
 
-                if not req.prompt and not req.image_urls:
-                    if not event.get_group_id() and req.extra_user_content_parts:
-                        req.prompt = "<attachment>"
-                    else:
-                        return
+                build_result: MainAgentBuildResult | None = await build_main_agent(
+                    event=event,
+                    plugin_context=self.ctx.plugin_manager.context,
+                    config=build_cfg,
+                )
 
-                # call event hook
-                if await call_event_hook(event, EventType.OnLLMRequestEvent, req):
+                if build_result is None:
                     return
 
-                # apply knowledge base feature
-                await self._apply_kb(event, req)
-
-                # truncate contexts to fit max length
-                # NOW moved to ContextManager inside ToolLoopAgentRunner
-                # if req.contexts:
-                #     req.contexts = self._truncate_contexts(req.contexts)
-                #     self._fix_messages(req.contexts)
-
-                # session_id
-                if not req.session_id:
-                    req.session_id = event.unified_msg_origin
-
-                # check provider modalities, if provider does not support image/tool_use, clear them in request.
-                self._modalities_fix(provider, req)
-
-                # filter tools, only keep tools from this pipeline's selected plugins
-                self._plugin_tool_fix(event, req)
-
-                # sanitize contexts (including history) by provider modalities
-                self._sanitize_context_by_modalities(provider, req)
+                agent_runner = build_result.agent_runner
+                req = build_result.provider_request
+                provider = build_result.provider
 
-                # apply llm safety mode
-                if self.llm_safety_mode:
-                    self._apply_llm_safety_mode(req)
-
-                # apply sandbox tools
-                if self.sandbox_cfg.get("enable", False):
-                    self._apply_sandbox_tools(req, req.session_id)
+                api_base = provider.provider_config.get("api_base", "")
+                for host in decoded_blocked:
+                    if host in api_base:
+                        logger.error(
+                            "Provider API base %s is blocked due to security reasons. Please use another ai provider.",
+                            api_base,
+                        )
+                        return
 
                 stream_to_general = (
                     self.unsupported_streaming_strategy == "turn_off"
                     and not event.platform_meta.support_streaming_message
                 )
 
-                # run agent
-                agent_runner = AgentRunner()
-                logger.debug(
-                    f"handle provider[id: {provider.provider_config['id']}] request: {req}",
-                )
-                astr_agent_ctx = AstrAgentContext(
-                    context=self.ctx.plugin_manager.context,
-                    event=event,
-                )
-
-                # inject model context length limit
-                if provider.provider_config.get("max_context_tokens", 0) <= 0:
-                    model = provider.get_model()
-                    if model_info := LLM_METADATAS.get(model):
-                        provider.provider_config["max_context_tokens"] = model_info[
-                            "limit"
-                        ]["context"]
-
-                # ChatUI 对话的标题生成
-                if event.get_platform_name() == "webchat":
-                    asyncio.create_task(self._handle_webchat(event, req, provider))
-
-                    # 注入 ChatUI 额外 prompt
-                    # 比如 follow-up questions 提示等
-                    req.system_prompt += f"\n{CHATUI_EXTRA_PROMPT}\n"
-
-                # 注入基本 prompt
-                if req.func_tool and req.func_tool.tools:
-                    tool_prompt = (
-                        TOOL_CALL_PROMPT
-                        if self.tool_schema_mode == "full"
-                        else TOOL_CALL_PROMPT_SKILLS_LIKE_MODE
-                    )
-                    req.system_prompt += f"\n{tool_prompt}\n"
+                if await call_event_hook(event, EventType.OnLLMRequestEvent, req):
+                    return
 
                 action_type = event.get_extra("action_type")
-                if action_type == "live":
-                    req.system_prompt += f"\n{LIVE_MODE_SYSTEM_PROMPT}\n"
 
                 event.trace.record(
                     "astr_agent_prepare",
@@ -703,24 +203,6 @@ class InternalAgentSubStage(Stage):
                     },
                 )
 
-                await agent_runner.reset(
-                    provider=provider,
-                    request=req,
-                    run_context=AgentContextWrapper(
-                        context=astr_agent_ctx,
-                        tool_call_timeout=self.tool_call_timeout,
-                    ),
-                    tool_executor=FunctionToolExecutor(),
-                    agent_hooks=MAIN_AGENT_HOOKS,
-                    streaming=streaming_response,
-                    llm_compress_instruction=self.llm_compress_instruction,
-                    llm_compress_keep_recent=self.llm_compress_keep_recent,
-                    llm_compress_provider=self._get_compress_provider(),
-                    truncate_turns=self.dequeue_context_length,
-                    enforce_max_turns=self.max_context_length,
-                    tool_schema_mode=self.tool_schema_mode,
-                )
-
                 # 检测 Live Mode
                 if action_type == "live":
                     # Live Mode: 使用 run_live_agent
@@ -840,3 +322,52 @@ class InternalAgentSubStage(Stage):
                     f"Error occurred while processing agent request: {e}"
                 )
             )
+
+    async def _save_to_history(
+        self,
+        event: AstrMessageEvent,
+        req: ProviderRequest,
+        llm_response: LLMResponse | None,
+        all_messages: list[Message],
+        runner_stats: AgentStats | None,
+    ):
+        if (
+            not req
+            or not req.conversation
+            or not llm_response
+            or llm_response.role != "assistant"
+        ):
+            return
+
+        if not llm_response.completion_text and not req.tool_calls_result:
+            logger.debug("LLM 响应为空，不保存记录。")
+            return
+
+        message_to_save = []
+        skipped_initial_system = False
+        for message in all_messages:
+            if message.role == "system" and not skipped_initial_system:
+                skipped_initial_system = True
+                continue
+            if message.role in ["assistant", "user"] and getattr(
+                message, "_no_save", None
+            ):
+                continue
+            message_to_save.append(message.model_dump())
+
+        token_usage = None
+        if runner_stats:
+            token_usage = runner_stats.token_usage.total
+
+        await self.conv_manager.update_conversation(
+            event.unified_msg_origin,
+            req.conversation.cid,
+            history=message_to_save,
+            token_usage=token_usage,
+        )
+
+
+# we prevent astrbot from connecting to known malicious hosts
+# these hosts are base64 encoded
+BLOCKED = {"dGZid2h2d3IuY2xvdWQuc2VhbG9zLmlv", "a291cmljaGF0"}
+decoded_blocked = [base64.b64decode(b).decode("utf-8") for b in BLOCKED]