refactor: revise LLM message schema and fix the reload logic when using dataclass-based LLM Tool registration

[Soulter]

更好地描述 LLM Message 模型。

Inspired by elegant https://github.com/MoonshotAI/kosong

在这个 PR 中，

1. 我们基于 Pydantic 并且参考 MoonshotAI/kosong 引入了新的表示 LLM Message 的数据模型。由于 kosong 需要 Python 版本 >= 3.13，因此我们无法直接以库的形式引入 kosong。

   from astrbot.core.agent.message import (
       AssistantMessageSegment,
       UserMessageSegment,
       TextPart,
   )
   from astrbot.core.agent.tool import ToolSet
   
   provider = self.context.get_using_provider(event.unified_msg_origin)
   conv_mgr = self.context.conversation_manager
   curr_cid = await conv_mgr.get_curr_conversation_id(event.unified_msg_origin)
   if provider and curr_cid:
       user_msg = UserMessageSegment(content=[TextPart(text="hi")])
       llm_resp = await provider.text_chat(
           contexts=[user_msg],
           func_tool=ToolSet(tools=[HelloWorldTool()]),
       )
       await conv_mgr.add_message_pair(
           cid=curr_cid,
           user_message=user_msg,
           assistant_message=AssistantMessageSegment(
               content=[TextPart(text=llm_resp.completion_text)]
           ),
       )
       print(llm_resp.completion_text)

2. 重构了以类的方式注册的 Tool 的 Tool 运行方法。现在我们将重载 call 来添加 Tool 的运行方法。

   from mcp.types import CallToolResult
   from pydantic import Field
   from pydantic.dataclasses import dataclass
   
   from astrbot.api import FunctionTool
   from astrbot.core.agent.run_context import ContextWrapper
   from astrbot.core.astr_agent_context import AstrAgentContext
   
   
   @dataclass
   class HelloWorldTool(FunctionTool):
       name: str = "hello_world"  # 工具名称
       description: str = "Say hello to the world."  # 工具描述
       parameters: dict = Field(
           default_factory=lambda: {
               "type": "object",
               "properties": {
                   "greeting": {
                       "type": "string",
                       "description": "The greeting message.",
                   },
               },
               "required": ["greeting"],
           }
       )  # 工具参数定义，见 OpenAI 官网或 https://json-schema.org/understanding-json-schema/
   
       async def call(
           self, context: ContextWrapper[AstrAgentContext], **kwargs
       ) -> str | CallToolResult:
           # event 在 context.context.event 中可用
           greeting = kwargs.get("greeting", "Hello")
           return f"{greeting}, World!"  # 也支持 mcp.types.CallToolResult 类型

   相比原来的 run 方法，call 方法将 event 实参改为了 ContextWrapper[AstrAgentContext] 类型的 context 方法。优点是更加标准化、独立。并且，我们引入了 Pydantic 进行数据校验，现在将会自动对 parameters 进行数据校验以符合 https://json-schema.org/draft/2020-12 。同时，run 方法我们仍然做了兼容，不过从此开始我们推荐使用 call 方法。

3. 修复了以类注册 Tool 后，重载插件的 Bug。

4. [Bug]插件主动发送的AI消息无法进入对话历史，导致上下文断裂 #3216 为插件开发引入了更好用的添加 LLM 消息上下文到对话历史记录的方法 add_message_pair。可以参考第一个代码片段。

5. 为 Astr Agent SDK 做准备。

Compatibility & Breaking Changes / 兼容性与破坏性变更

• 这是一个破坏性变更 (Breaking Change)。/ This is a breaking change.
• 这不是一个破坏性变更。/ This is NOT a breaking change.

---

Checklist / 检查清单

• 😊 如果 PR 中有新加入的功能，已经通过 Issue / 邮件等方式和作者讨论过。/ If there are new features added in the PR, I have discussed it with the authors through issues/emails, etc.
• 👀 我的更改经过了良好的测试，并已在上方提供了“验证步骤”和“运行截图”。/ My changes have been well-tested, and "Verification Steps" and "Screenshots" have been provided above.
• 🤓 我确保没有引入新依赖库，或者引入了新依赖库的同时将其添加到了 requirements.txt 和 pyproject.toml 文件相应位置。/ I have ensured that no new dependencies are introduced, OR if new dependencies are introduced, they have been added to the appropriate locations in requirements.txt and pyproject.toml.
• 😮 我的更改没有引入恶意代码。/ My changes do not introduce malicious code.

[Copilot]

Pull Request Overview

This PR refactors the agent tool system to improve type safety and modularity by introducing a new message model, restructuring tool abstractions, and consolidating context handling. The changes modernize type annotations using union syntax (| instead of Optional) and relocate message-related classes to a dedicated module.

Key changes include:

• Introduction of astrbot/core/agent/message.py with Pydantic-based message models (Message, AssistantMessageSegment, ToolCallMessageSegment, etc.)
• Refactored FunctionTool to use ToolSchema base class with proper JSON schema validation
• Created MCPTool as a specialized subclass of FunctionTool for MCP service calls
• Moved tool_call_timeout from AstrAgentContext to ContextWrapper for better encapsulation
• Updated all provider sources to handle optional prompts and convert Message objects to dicts

Reviewed Changes

Copilot reviewed 19 out of 19 changed files in this pull request and generated 9 comments.

Show a summary per file

File	Description
astrbot/core/utils/pip_installer.py	Updated type hints to use modern union syntax (str | None)
astrbot/core/provider/sources/openai_source.py	Added Message import, made prompt optional, added type guards for tool calls
astrbot/core/provider/sources/gemini_source.py	Made prompt optional, added Message conversion logic, removed unreachable code
astrbot/core/provider/sources/coze_source.py	Added Message-to-dict conversion before processing contexts
astrbot/core/provider/sources/anthropic_source.py	Made prompt optional, added Message conversion logic
astrbot/core/provider/provider.py	Updated abstract method signatures to support list[Message] contexts, added conversion helper
astrbot/core/provider/func_tool_manager.py	Refactored MCP tool filtering to use isinstance(f, MCPTool) instead of origin field
astrbot/core/provider/entities.py	Moved message segment classes to agent.message module, renamed Anthropic Message import to avoid collision
astrbot/core/pipeline/process_stage/method/llm_request.py	Refactored tool execution logic to detect overridden call method, moved event to context
astrbot/core/pipeline/context_utils.py	Added call_local_llm_tool helper function for executing local tools
astrbot/core/pipeline/context.py	Exported call_local_llm_tool from context_utils
astrbot/core/conversation_mgr.py	Added add_message_pair method for recording user-assistant conversations
astrbot/core/astr_agent_context.py	Moved tool_call_timeout to ContextWrapper, added event field
astrbot/core/agent/tool.py	Refactored to use ToolSchema base class, added JSON schema validation, introduced abstract call method
astrbot/core/agent/runners/tool_loop_agent_runner.py	Updated to use new message classes, removed redundant event clearing
astrbot/core/agent/run_context.py	Moved tool_call_timeout from AstrAgentContext, removed unused AstrMessageEvent import
astrbot/core/agent/message.py	New file containing Pydantic message models with ContentPart registry system
astrbot/core/agent/mcp_client.py	Added MCPTool class implementing the call method for MCP services
astrbot/core/agent/hooks.py	Removed unused @dataclass decorator from BaseAgentRunHooks

[LIghtJUNction]

一些细节修正

[Copilot]

Pull Request Overview

Copilot reviewed 19 out of 19 changed files in this pull request and generated 11 comments.

Comments suppressed due to low confidence (2)

astrbot/core/pipeline/process_stage/method/llm_request.py:148

• Corrected spelling of 'deligate' to 'delegate'.

                    text=f"error when deligate task to {tool.agent.name}",

astrbot/core/pipeline/process_stage/method/llm_request.py:170

• Corrected spelling of 'deligate' to 'delegate'.

                text=f"error when deligate task to {tool.agent.name}",

[Copilot]

Pull Request Overview

Copilot reviewed 20 out of 20 changed files in this pull request and generated 7 comments.