eabf90c496
- Add AI学习助手 agent creation script with all 39 tools, 3-layer KG+RAG memory - Add renshenguo (人参果) feishu bot integration (app_service + ws_handler) - Register renshenguo WS client in main.py startup - Add RENSHENGUO_APP_ID / RENSHENGUO_APP_SECRET / RENSHENGUO_AGENT_ID config - Reorganize docs from root into docs/ subdirectories - Move startup scripts to scripts/startup/ - Various backend optimizations and tool improvements Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
343 lines
12 KiB
Python
343 lines
12 KiB
Python
"""
|
||
场景模板注册：路线图「客服 / 研发 / 运维」三类最小可运行工作流，供 API 与脚本一键创建 Agent。
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
from typing import Any, Callable, Dict, List, Optional
|
||
|
||
PromptBuilder = Callable[[Dict[str, Any]], str]
|
||
|
||
|
||
def _default_prompt_cs(params: Dict[str, Any]) -> str:
|
||
extra = (params.get("extra_instructions") or "").strip()
|
||
base = """你是企业客服场景 Agent。根据用户问题给出清晰、可执行的回答；不确定时先澄清。
|
||
用户输入可能包含：{{input}} 或来自前序节点的合并字段。保持礼貌、简洁。"""
|
||
if extra:
|
||
return f"{base}\n\n【额外说明】\n{extra}"
|
||
return base
|
||
|
||
|
||
def _default_prompt_dev(params: Dict[str, Any]) -> str:
|
||
extra = (params.get("extra_instructions") or "").strip()
|
||
lang = params.get("preferred_language") or "任意合适语言"
|
||
base = f"""你是研发辅助 Agent，负责代码与设计说明。优先给出可运行示例与步骤；涉及安全/生产变更时明确风险。
|
||
偏好语言/栈：{lang}。
|
||
用户诉求：{{input}}"""
|
||
if extra:
|
||
return f"{base}\n\n【额外说明】\n{extra}"
|
||
return base
|
||
|
||
|
||
def _default_prompt_ops(params: Dict[str, Any]) -> str:
|
||
extra = (params.get("extra_instructions") or "").strip()
|
||
base = """你是运维/日志分析场景 Agent。帮助用户解读日志片段、定位可能原因与下一步排查；不要编造未提供的日志内容。
|
||
用户输入：{{input}}"""
|
||
if extra:
|
||
return f"{base}\n\n【额外说明】\n{extra}"
|
||
return base
|
||
|
||
|
||
def _build_minimal_workflow(
|
||
prompt: str,
|
||
*,
|
||
temperature: float = 0.3,
|
||
enable_tools: bool = False,
|
||
tools: Optional[List[str]] = None,
|
||
) -> Dict[str, Any]:
|
||
tools = tools or []
|
||
return {
|
||
"nodes": [
|
||
{
|
||
"id": "start-1",
|
||
"type": "start",
|
||
"position": {"x": 80, "y": 120},
|
||
"data": {},
|
||
},
|
||
{
|
||
"id": "llm-1",
|
||
"type": "llm",
|
||
"position": {"x": 320, "y": 120},
|
||
"data": {
|
||
"prompt": prompt,
|
||
"temperature": float(temperature),
|
||
"enable_tools": enable_tools,
|
||
"tools": tools,
|
||
"selected_tools": tools,
|
||
},
|
||
},
|
||
{
|
||
"id": "end-1",
|
||
"type": "end",
|
||
"position": {"x": 560, "y": 120},
|
||
"data": {},
|
||
},
|
||
],
|
||
"edges": [
|
||
{
|
||
"id": "e_start_llm",
|
||
"source": "start-1",
|
||
"target": "llm-1",
|
||
"sourceHandle": "right",
|
||
"targetHandle": "left",
|
||
},
|
||
{
|
||
"id": "e_llm_end",
|
||
"source": "llm-1",
|
||
"target": "end-1",
|
||
"sourceHandle": "right",
|
||
"targetHandle": "left",
|
||
},
|
||
],
|
||
}
|
||
|
||
|
||
def build_workflow_for_template(template_id: str, parameters: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:
|
||
"""
|
||
根据模板 ID 与参数生成 workflow_config（nodes+edges）。
|
||
|
||
通用参数：temperature, enable_tools, tools(list[str]), extra_instructions, preferred_language(dev)
|
||
"""
|
||
parameters = dict(parameters or {})
|
||
meta = SCENE_TEMPLATE_REGISTRY.get(template_id)
|
||
if not meta:
|
||
raise ValueError(f"未知模板: {template_id}")
|
||
|
||
temperature = float(parameters.get("temperature", meta.get("default_temperature", 0.3)))
|
||
enable_tools = bool(parameters.get("enable_tools", True))
|
||
tools = parameters.get("tools")
|
||
if tools is not None and not isinstance(tools, list):
|
||
tools = []
|
||
elif tools is None:
|
||
tools = list(meta.get("default_tools") or [])
|
||
|
||
prompt_fn: PromptBuilder = meta["prompt_builder"]
|
||
prompt = prompt_fn(parameters)
|
||
|
||
# 使用模板自定义的 workflow builder（如果有）
|
||
workflow_builder = meta.get("workflow_builder")
|
||
if workflow_builder:
|
||
return workflow_builder(
|
||
prompt,
|
||
temperature=temperature,
|
||
enable_tools=enable_tools,
|
||
tools=tools,
|
||
)
|
||
|
||
return _build_minimal_workflow(
|
||
prompt,
|
||
temperature=temperature,
|
||
enable_tools=enable_tools,
|
||
tools=tools,
|
||
)
|
||
|
||
|
||
def _default_prompt_learning(params: Dict[str, Any]) -> str:
|
||
extra = (params.get("extra_instructions") or "").strip()
|
||
subject = params.get("subject") or "通用"
|
||
level = params.get("level") or "中级"
|
||
|
||
base = f"""# 角色：智能学习助手（知识图谱 + RAG 增强版）
|
||
|
||
你是专为深度学习场景设计的 AI 学习助手，具备**知识图谱构建**、**向量语义检索**和**永久记忆**能力。
|
||
|
||
## 核心架构
|
||
|
||
你的知识系统由三层组成：
|
||
1. **知识图谱 (Knowledge Graph)**：结构化存储知识点实体及其前置/扩展/包含/示例关系，构建学科知识网络
|
||
2. **向量记忆 (Vector Memory)**：语义检索历史对话和相关知识，找到最相似的学习内容
|
||
3. **长期记忆 (Persistent Memory)**：跨会话保存用户画像、学习进度、薄弱环节
|
||
|
||
## 当前学习配置
|
||
- 学科领域：{subject}
|
||
- 难度级别：{level}
|
||
- 用户输入将包含学习问题、材料或请求
|
||
|
||
## 工作流程（每次对话必须遵循）
|
||
|
||
### 阶段 1：理解与分析
|
||
1. 理解用户的学习意图（提问/复习/练习/总结/规划）
|
||
2. 用 `knowledge_graph_search` 检索相关知识图谱实体
|
||
3. 如果用户提供了新的学习材料/知识点，用 `knowledge_graph_add` 自动提取并存储
|
||
|
||
### 阶段 2：知识检索与融合
|
||
4. 结合图谱检索结果和历史向量记忆，构建完整的知识上下文
|
||
5. 用 `entity_search` 查找特定概念的前置知识和扩展内容
|
||
6. 如果需要学习路径建议，用 `learning_path` 分析依赖关系
|
||
|
||
### 阶段 3：生成与交付
|
||
7. 基于融合后的知识上下文生成高质量回答
|
||
8. 回答应包含：
|
||
- 核心概念解释（关联知识图谱中的实体）
|
||
- 前置知识提醒（如果有依赖关系）
|
||
- 实例或练习题（如适用）
|
||
- 扩展阅读建议（关联的扩展知识点）
|
||
9. 用 `self_review` 自检回答质量
|
||
|
||
### 阶段 4：巩固与记忆
|
||
10. 将本轮对话中的重要知识点持久化到长期记忆
|
||
11. 更新用户画像（掌握程度、薄弱环节、学习偏好）
|
||
|
||
## 知识图谱工具使用指南
|
||
|
||
| 工具 | 用途 | 何时使用 |
|
||
|------|------|---------|
|
||
| `knowledge_graph_search` | 向量+图谱混合检索 | 每次回答学习问题前 |
|
||
| `knowledge_graph_add` | 从文本提取实体和关系 | 用户分享学习材料/新知识点时 |
|
||
| `entity_search` | 关键词搜索实体 | 查找特定概念详情时 |
|
||
| `learning_path` | 推荐学习路径 | 用户询问学习顺序/计划时 |
|
||
|
||
## 回答风格
|
||
- 使用 Markdown 格式，层次分明
|
||
- 关键概念用 **粗体** 标记
|
||
- 公式用代码块或 LaTeX 表达
|
||
- 每个回答末尾附上 "📚 相关知识点" 列表（来自图谱检索结果）
|
||
- 必要时用 `task_plan` 为用户制定学习计划
|
||
|
||
## 记忆与个性化
|
||
- 记住用户的学习进度和薄弱环节
|
||
- 根据用户级别（{level}）调整解释深度
|
||
- 对反复出错的知识点主动提醒和强化"""
|
||
|
||
if extra:
|
||
return f"{base}\n\n【额外说明】\n{extra}"
|
||
return base
|
||
|
||
|
||
def _build_learning_workflow(
|
||
prompt: str,
|
||
*,
|
||
temperature: float = 0.7,
|
||
enable_tools: bool = True,
|
||
tools: Optional[List[str]] = None,
|
||
) -> Dict[str, Any]:
|
||
"""为学习助手构建更丰富的工作流（含开始→LLM→结束，LLM配置KG+RAG工具）。"""
|
||
tools = tools or []
|
||
return {
|
||
"nodes": [
|
||
{
|
||
"id": "start-1",
|
||
"type": "start",
|
||
"position": {"x": 80, "y": 200},
|
||
"data": {"label": "学习任务开始"},
|
||
},
|
||
{
|
||
"id": "llm-learning",
|
||
"type": "llm",
|
||
"position": {"x": 350, "y": 200},
|
||
"data": {
|
||
"label": "智能学习助手 (KG+RAG)",
|
||
"prompt": prompt,
|
||
"temperature": float(temperature),
|
||
"enable_tools": enable_tools,
|
||
"tools": tools,
|
||
"selected_tools": tools,
|
||
"model": "deepseek-chat",
|
||
"provider": "deepseek",
|
||
"max_iterations": 20,
|
||
"memory": True,
|
||
"memory_max_history": 30,
|
||
"memory_vector_enabled": True,
|
||
"memory_vector_top_k": 8,
|
||
"memory_persist": True,
|
||
"memory_learning": True,
|
||
},
|
||
},
|
||
{
|
||
"id": "end-1",
|
||
"type": "end",
|
||
"position": {"x": 620, "y": 200},
|
||
"data": {"label": "学习完成"},
|
||
},
|
||
],
|
||
"edges": [
|
||
{
|
||
"id": "e_start_learning",
|
||
"source": "start-1",
|
||
"target": "llm-learning",
|
||
"sourceHandle": "right",
|
||
"targetHandle": "left",
|
||
},
|
||
{
|
||
"id": "e_learning_end",
|
||
"source": "llm-learning",
|
||
"target": "end-1",
|
||
"sourceHandle": "right",
|
||
"targetHandle": "left",
|
||
},
|
||
],
|
||
}
|
||
|
||
|
||
SCENE_TEMPLATE_REGISTRY: Dict[str, Dict[str, Any]] = {
|
||
"template_customer_service": {
|
||
"title": "客服场景",
|
||
"description": "通用客服问答与澄清（最小 LLM 链）。",
|
||
"category": "customer_service",
|
||
"default_temperature": 0.35,
|
||
"default_tools": [],
|
||
"prompt_builder": _default_prompt_cs,
|
||
},
|
||
"template_dev_codegen": {
|
||
"title": "研发 / 代码助手",
|
||
"description": "代码与设计说明辅助（最小 LLM 链）。",
|
||
"category": "dev",
|
||
"default_temperature": 0.25,
|
||
"default_tools": [],
|
||
"prompt_builder": _default_prompt_dev,
|
||
},
|
||
"template_ops_log_analysis": {
|
||
"title": "运维 / 日志分析",
|
||
"description": "日志解读与排查建议（最小 LLM 链）。",
|
||
"category": "ops",
|
||
"default_temperature": 0.3,
|
||
"default_tools": [],
|
||
"prompt_builder": _default_prompt_ops,
|
||
},
|
||
"template_learning_assistant": {
|
||
"title": "智能学习助手 (KG+RAG)",
|
||
"description": "知识图谱 + RAG 增强学习助手：实体抽取、关系图谱、向量检索、永久记忆，适合有规模要求的学习场景。",
|
||
"category": "education",
|
||
"default_temperature": 0.7,
|
||
"default_tools": [
|
||
"knowledge_graph_search",
|
||
"knowledge_graph_add",
|
||
"entity_search",
|
||
"learning_path",
|
||
"file_read",
|
||
"file_write",
|
||
"text_analyze",
|
||
"web_search",
|
||
"task_plan",
|
||
"self_review",
|
||
"math_calculate",
|
||
"json_process",
|
||
"datetime",
|
||
],
|
||
"prompt_builder": _default_prompt_learning,
|
||
"workflow_builder": _build_learning_workflow,
|
||
},
|
||
}
|
||
|
||
|
||
def list_scene_template_meta() -> List[Dict[str, Any]]:
|
||
"""供 GET 接口返回（不含 prompt_builder）。"""
|
||
out: List[Dict[str, Any]] = []
|
||
for tid, meta in SCENE_TEMPLATE_REGISTRY.items():
|
||
hints = ["temperature", "enable_tools", "tools", "extra_instructions"]
|
||
if tid == "template_dev_codegen":
|
||
hints.append("preferred_language")
|
||
if tid == "template_learning_assistant":
|
||
hints.extend(["subject（学科领域）", "level（难度级别：初级/中级/高级）"])
|
||
|
||
out.append(
|
||
{
|
||
"id": tid,
|
||
"title": meta["title"],
|
||
"description": meta["description"],
|
||
"category": meta.get("category"),
|
||
"default_temperature": meta.get("default_temperature"),
|
||
"parameter_hints": hints,
|
||
}
|
||
)
|
||
return out
|