sycommon-python-lib 0.2.5a26__tar.gz → 0.2.5a27__tar.gz

{sycommon_python_lib-0.2.5a26 → sycommon_python_lib-0.2.5a27}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sycommon-python-lib
-Version: 0.2.5a26
+Version: 0.2.5a27
 Summary: Add your description here
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown
@@ -9,7 +9,7 @@ Requires-Dist: aiohttp==3.13.5
 Requires-Dist: aiomysql==0.3.2
 Requires-Dist: anyio==4.13.0
 Requires-Dist: decorator==5.3.1
-Requires-Dist: deepagents==0.6.4
+Requires-Dist: deepagents==0.6.6
 Requires-Dist: elasticsearch==9.4.1
 Requires-Dist: fastapi==0.136.3
 Requires-Dist: jinja2==3.1.6

{sycommon_python_lib-0.2.5a26 → sycommon_python_lib-0.2.5a27}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "sycommon-python-lib"
-version = "0.2.5a26"
+version = "0.2.5a27"
 description = "Add your description here"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -10,7 +10,7 @@ dependencies = [
     "aiomysql==0.3.2",
     "anyio==4.13.0",
     "decorator==5.3.1",
-    "deepagents==0.6.4",
+    "deepagents==0.6.6",
     "elasticsearch==9.4.1",
     "fastapi==0.136.3",
     "jinja2==3.1.6",
@@ -54,6 +54,11 @@ package-data = {command = ["templates/**/*", "templates/**/.*"], sycommon = ["ll
 requires = ["setuptools"]
 build-backend = "setuptools.build_meta"
 
+[dependency-groups]
+dev = [
+    "pytest>=9.0.3",
+]
+
 [project.scripts]
 sycommon = "command.cli:main"
 sycli = "sycli.cli:main"

{sycommon_python_lib-0.2.5a26 → sycommon_python_lib-0.2.5a27}/src/sycommon/agent/chat_events.py

@@ -62,6 +62,9 @@ class EventType(str, Enum):
     PARALLEL_AGENTS = "parallel_agents"  # 并行 Agent
     AGENT_COMPLETE = "agent_complete"  # Agent 任务完成
 
+    # 评分相关
+    RUBRIC_EVALUATION = "rubric_evaluation"  # Rubric 评估结果
+
     # 生命周期
     DONE = "done"  # 完成
     ERROR = "error"  # 错误
@@ -368,6 +371,39 @@ class ChatEventBuilder:
         """
         return ChatEvent(EventType.CANCELLED, {"message": message})
 
+    # ============== 评分相关 ==============
+
+    @staticmethod
+    def rubric_evaluation(
+        iteration: int,
+        result: str,
+        criteria: list,
+        explanation: str = "",
+        agent: str = DEFAULT_AGENT_NAME,
+    ) -> ChatEvent:
+        """
+        构建 Rubric 评估事件
+
+        Args:
+            iteration: 评估迭代次数
+            result: 评估结果（satisfied / needs_revision / failed / max_iterations_reached / grader_error）
+            criteria: 各评分标准的判定列表
+            explanation: 评估说明
+            agent: Agent 名称
+
+        Returns:
+            ChatEvent: 评估事件
+        """
+        data = {
+            "iteration": iteration,
+            "result": result,
+            "criteria": criteria,
+            "agent": agent,
+        }
+        if explanation:
+            data["explanation"] = explanation
+        return ChatEvent(EventType.RUBRIC_EVALUATION, data)
+
 
 # ============== 便捷函数（直接返回字典，兼容旧代码） ==============
 

{sycommon_python_lib-0.2.5a26 → sycommon_python_lib-0.2.5a27}/src/sycommon/agent/deep_agent.py

@@ -117,10 +117,28 @@ class AgentConfig(BaseModel):
     # 调试模式
     debug: bool = False
 
+    # Rubric 评分配置
+    rubric_max_iterations: int = 3  # rubric 评估最大迭代次数（0 = 不启用）
+
     class Config:
         arbitrary_types_allowed = True
 
 
+class _RubricEvalCollector:
+    """收集 RubricMiddleware 通过 on_evaluation 回调产生的评估结果"""
+
+    def __init__(self):
+        self.evaluations: list = []
+
+    def on_evaluation(self, evaluation):
+        """RubricMiddleware 每次评估完成后调用"""
+        self.evaluations.append(evaluation)
+
+    @property
+    def latest(self):
+        return self.evaluations[-1] if self.evaluations else None
+
+
 class DeepAgent:
     """数字员工封装类"""
 
@@ -131,12 +149,14 @@ class DeepAgent:
         config: Dict[str, Any],
         sandbox_backend: HTTPSandboxBackend,
         recovery_manager: Optional[SandboxRecoveryManager] = None,
+        rubric_eval_collector: Optional[_RubricEvalCollector] = None,
     ):
         self.user_id = user_id
         self.agent = agent
         self.config = config
         self.sandbox_backend = sandbox_backend
         self.recovery_manager = recovery_manager
+        self.rubric_eval_collector = rubric_eval_collector
 
     async def cleanup(self):
         """清理资源"""
@@ -149,6 +169,7 @@ class DeepAgent:
     async def chat(
         self,
         message: str,
+        rubric: Optional[str] = None,
         cancel_event: Optional[asyncio.Event] = None,
         on_event: Optional[Callable[[ChatEvent], Awaitable[None]]] = None,
     ) -> AsyncGenerator[ChatEvent, None]:
@@ -156,6 +177,7 @@ class DeepAgent:
 
         Args:
             message: 用户消息
+            rubric: 评分标准（可选），传入后 agent 完成时会自动评估是否满足标准
             cancel_event: 取消事件（用于取消当前任务）
             on_event: 事件回调，每个事件触发后调用（可用于 DB 持久化等）
 
@@ -172,8 +194,11 @@ class DeepAgent:
         total_output_tokens = 0
 
         try:
+            input_state = {"messages": [HumanMessage(content=message)]}
+            if rubric:
+                input_state["rubric"] = rubric
             async for chunk in self._astream_with_retry(
-                {"messages": [HumanMessage(content=message)]},
+                input_state,
                 max_retries=3,
                 base_delay=1.0,
             ):
@@ -650,6 +675,8 @@ async def create_deep_agent(
 
     # 创建 agent
     from deepagents import create_deep_agent as _create_deep_agent
+    from deepagents import RubricMiddleware
+    from deepagents.middleware.filesystem import FilesystemMiddleware
 
     # 创建 compact_conversation 工具 middleware
     # 自动压缩由 deepagents 内置的 SummarizationMiddleware 处理（通过 model.profile 获取阈值）
@@ -657,6 +684,27 @@ async def create_deep_agent(
         model, config.model_name, sandbox_backend,
     )
 
+    # 🔑 Rubric 评估收集器：通过 on_evaluation 回调收集评估结果
+    rubric_eval_collector = None
+
+    # 构建 middleware 列表
+    middleware_list = [
+        ToolResultTruncationMiddleware(),
+        TokenTrackingMiddleware(model_name=config.model_name, user_id=user_id),
+        summarization_mw,
+    ]
+
+    # RubricMiddleware：复用 FilesystemMiddleware 的工具（底层走 HTTPSandboxBackend HTTP 远程调用）
+    if config.rubric_max_iterations > 0 and sandbox_backend:
+        rubric_eval_collector = _RubricEvalCollector()
+        _rubric_fs_middleware = FilesystemMiddleware(backend=sandbox_backend)
+        middleware_list.append(RubricMiddleware(
+            model=model,
+            tools=_rubric_fs_middleware.tools,
+            max_iterations=config.rubric_max_iterations,
+            on_evaluation=rubric_eval_collector.on_evaluation,
+        ))
+
     agent_kwargs = {
         "model": model,
         "tools": config.tools or [get_current_date],
@@ -664,11 +712,7 @@ async def create_deep_agent(
         "checkpointer": checkpointer,
         "system_prompt": config.system_prompt,
         "debug": config.debug,
-        "middleware": [
-            ToolResultTruncationMiddleware(),
-            TokenTrackingMiddleware(model_name=config.model_name, user_id=user_id),
-            summarization_mw,
-        ],
+        "middleware": middleware_list,
     }
     if skills_path:
         agent_kwargs["skills"] = [skills_path]
@@ -685,6 +729,7 @@ async def create_deep_agent(
         config=agent_config,
         sandbox_backend=sandbox_backend,
         recovery_manager=recovery_manager,
+        rubric_eval_collector=rubric_eval_collector,
     )
 
 

{sycommon_python_lib-0.2.5a26 → sycommon_python_lib-0.2.5a27}/src/sycommon/agent/sandbox/file_ops.py

@@ -510,9 +510,15 @@ class FileOperationsMixin:
             })
             # GrepResponse: {"matches": [...], "partial_error": "...", "error": "..."}
             matches = [GrepMatch(**item) for item in result.get("matches", [])]
+            # deepagents GrepResult 只有 error 和 matches 字段，
+            # 将 partial_error 合并到 error 中（部分失败但有结果时不设 error）
+            partial_error = result.get("partial_error")
+            error = result.get("error")
+            if not error and partial_error and not matches:
+                error = partial_error
             return GrepResult(
                 matches=matches,
-                error=result.get("error"),
+                error=error,
             )
         except Exception as e:
             SYLogger.error(f"[Sandbox] grep 搜索失败: {e}")
@@ -537,9 +543,13 @@ class FileOperationsMixin:
             }, timeout=timeout)
             # GrepResponse: {"matches": [...], "partial_error": "...", "error": "..."}
             matches = [GrepMatch(**item) for item in result.get("matches", [])]
+            partial_error = result.get("partial_error")
+            error = result.get("error")
+            if not error and partial_error and not matches:
+                error = partial_error
             return GrepResult(
                 matches=matches,
-                error=result.get("error"),
+                error=error,
             )
         except Exception as e:
             SYLogger.error(f"[Sandbox] 异步 grep 搜索失败: {e}")

{sycommon_python_lib-0.2.5a26 → sycommon_python_lib-0.2.5a27}/src/sycommon/llm/llm_with_token_tracking.py

@@ -312,15 +312,6 @@ class LLMWithTokenTracking(BaseChatModel):
             return self.llm.profile
         return None
 
-    @property
-    def profile(self):
-        p = self._resolve_model_profile()
-        return p
-
-    @profile.setter
-    def profile(self, value):
-        pass
-
     def bind_tools(self, tools, *, tool_choice=None, **kwargs):
         """绑定工具 - 委托给底层 LLM，返回绑定后的 Runnable"""
         return self.llm.bind_tools(tools, tool_choice=tool_choice, **kwargs)

{sycommon_python_lib-0.2.5a26 → sycommon_python_lib-0.2.5a27}/src/sycommon_python_lib.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sycommon-python-lib
-Version: 0.2.5a26
+Version: 0.2.5a27
 Summary: Add your description here
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown
@@ -9,7 +9,7 @@ Requires-Dist: aiohttp==3.13.5
 Requires-Dist: aiomysql==0.3.2
 Requires-Dist: anyio==4.13.0
 Requires-Dist: decorator==5.3.1
-Requires-Dist: deepagents==0.6.4
+Requires-Dist: deepagents==0.6.6
 Requires-Dist: elasticsearch==9.4.1
 Requires-Dist: fastapi==0.136.3
 Requires-Dist: jinja2==3.1.6

{sycommon_python_lib-0.2.5a26 → sycommon_python_lib-0.2.5a27}/src/sycommon_python_lib.egg-info/requires.txt

@@ -3,7 +3,7 @@ aiohttp==3.13.5
 aiomysql==0.3.2
 anyio==4.13.0
 decorator==5.3.1
-deepagents==0.6.4
+deepagents==0.6.6
 elasticsearch==9.4.1
 fastapi==0.136.3
 jinja2==3.1.6