Files
agent_jrxml/agent/graph.py
T
panda 70614dff5e feat: comprehensive v2 upgrade — streaming, error KB, file upload, layout analysis
Major changes:
- Streaming: LLM统一 _BaseLLM 接口 (invoke + stream), generate/modify/correct
  节点使用 get_stream_writer() 实现逐字输出, UI 节点平铺展开自动折叠
- Prompt外部化: 7个prompt拆分到 prompts/*.md, loader.py 支持热重载
- 错误自增长: backend/error_kb.py — 指纹去重 + ChromaDB持久化,
  correct_jrxml→validate 通过时自动入库, retrieve同时搜索错误KB
- 文件上传: backend/file_parser.py — PDF/DOCX/图片/文本解析,
  侧边栏多文件上传, 文本自动注入下一条消息
- A4模板识别: backend/layout_analyzer.py — 三种模式(完整A4/行片段修改/行片段新建),
  PaddleOCR元素提取 + 行分组 + JRXML section匹配
- 会话历史下载: jrxml_versions版本追踪 + 侧边栏历史版本下载按钮
- 预览修复: route_after_save跳过预览/导出意图的验证循环
- Ctrl+C修复: JS注入拦截Streamlit裸c键清缓存

Docs: CLAUDE.md (完整项目文档), ROADMAP.md (改进路线图)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-19 15:02:53 +08:00

232 lines
6.6 KiB
Python

"""LangGraph JRXML 生成代理的状态图定义。"""
import os
from typing import Literal
from dotenv import load_dotenv
from langgraph.graph import StateGraph, END
from agent.state import AgentState
from agent.nodes import (
load_session_node,
process_input,
manage_context,
save_state_snapshot,
classify_intent,
retrieve,
generate,
modify_jrxml,
handle_consult,
handle_undo,
handle_reset,
save_session_node,
validate,
explain_error,
correct_jrxml,
finalize,
)
load_dotenv()
MAX_RETRY = int(os.getenv("MAX_RETRY", "3"))
# ============================================================
# 路由函数
# ============================================================
def route_by_intent(state: AgentState) -> Literal[
"retrieve", "modify_jrxml", "save_session",
"handle_consult", "handle_undo", "handle_reset"
]:
"""根据 classify_intent 的结果路由到对应的处理节点。"""
intent = state.get("intent", "initial_generation")
if intent == "initial_generation":
return "retrieve"
elif intent == "modify_report":
return "modify_jrxml"
elif intent in ("preview_report", "export_pdf", "export_jrxml"):
return "save_session"
elif intent == "consult_question":
return "handle_consult"
elif intent == "undo_modification":
return "handle_undo"
elif intent == "reset_session":
return "handle_reset"
else:
# 兜底:根据是否有报表判断
if state.get("current_jrxml"):
return "modify_jrxml"
return "retrieve"
def route_after_generate(state: AgentState) -> Literal["save_session"]:
return "save_session"
def route_after_modify(state: AgentState) -> Literal["save_session"]:
return "save_session"
def route_after_undo(state: AgentState) -> Literal["save_session"]:
return "save_session"
def route_after_save(state: AgentState) -> Literal["validate", "finalize"]:
# 预览/导出意图跳过验证,直接完成
intent = state.get("intent", "")
if intent in ("preview_report", "export_pdf", "export_jrxml"):
return "finalize"
return "validate"
def route_after_validate(state: AgentState) -> Literal["finalize", "explain_error"]:
if state.get("status") == "pass":
return "finalize"
return "explain_error"
def route_after_explain(state: AgentState) -> Literal["correct_jrxml"]:
return "correct_jrxml"
def route_after_correct(state: AgentState) -> Literal["validate", "finalize"]:
retry = state.get("retry_count", 0)
if retry >= MAX_RETRY:
return "finalize"
return "validate"
# ============================================================
# 图构建
# ============================================================
def build_graph() -> StateGraph:
workflow = StateGraph(AgentState)
# 现有节点
workflow.add_node("load_session", load_session_node)
workflow.add_node("process_input", process_input)
workflow.add_node("manage_context", manage_context)
workflow.add_node("save_session", save_session_node)
workflow.add_node("retrieve", retrieve)
workflow.add_node("generate", generate)
workflow.add_node("modify_jrxml", modify_jrxml)
workflow.add_node("validate", validate)
workflow.add_node("explain_error", explain_error)
workflow.add_node("correct_jrxml", correct_jrxml)
workflow.add_node("finalize", finalize)
# 新增节点:意图识别
workflow.add_node("save_state_snapshot", save_state_snapshot)
workflow.add_node("classify_intent", classify_intent)
workflow.add_node("handle_consult", handle_consult)
workflow.add_node("handle_undo", handle_undo)
workflow.add_node("handle_reset", handle_reset)
# ---- 入口和前置流程 ----
workflow.set_entry_point("load_session")
workflow.add_edge("load_session", "process_input")
workflow.add_edge("process_input", "manage_context")
workflow.add_edge("manage_context", "save_state_snapshot")
workflow.add_edge("save_state_snapshot", "classify_intent")
# ---- 意图路由 ----
workflow.add_conditional_edges(
"classify_intent",
route_by_intent,
{
"retrieve": "retrieve",
"modify_jrxml": "modify_jrxml",
"save_session": "save_session",
"handle_consult": "handle_consult",
"handle_undo": "handle_undo",
"handle_reset": "handle_reset",
},
)
# ---- 初始生成分支 ----
workflow.add_edge("retrieve", "generate")
workflow.add_conditional_edges(
"generate",
route_after_generate,
{"save_session": "save_session"},
)
# ---- 修改分支 ----
workflow.add_conditional_edges(
"modify_jrxml",
route_after_modify,
{"save_session": "save_session"},
)
# ---- 撤销分支 ----
workflow.add_conditional_edges(
"handle_undo",
route_after_undo,
{"save_session": "save_session"},
)
# ---- 保存后进入验证 ----
workflow.add_conditional_edges(
"save_session",
route_after_save,
{"validate": "validate"},
)
# ---- 验证 → 修正循环 ----
workflow.add_conditional_edges(
"validate",
route_after_validate,
{"finalize": "finalize", "explain_error": "explain_error"},
)
workflow.add_conditional_edges(
"explain_error",
route_after_explain,
{"correct_jrxml": "correct_jrxml"},
)
workflow.add_conditional_edges(
"correct_jrxml",
route_after_correct,
{"validate": "validate", "finalize": "finalize"},
)
# ---- 咨询 / 重置 → 直接结束 ----
workflow.add_edge("handle_consult", "finalize")
workflow.add_edge("handle_reset", "finalize")
# ---- 结束 ----
workflow.add_edge("finalize", END)
return workflow.compile()
# ============================================================
# 初始状态
# ============================================================
def create_initial_state() -> AgentState:
return AgentState(
conversation_history=[],
current_jrxml="",
user_input="",
status="",
error_msg="",
natural_explanation="",
retry_count=0,
user_modification_request="",
final_jrxml="",
stage="initial_generation",
retrieved_context="",
full_conversation_history=[],
compressed_history="",
current_token_count=0,
session_id="",
session_name="",
created_at="",
updated_at="",
intent="",
history_states=[],
jrxml_versions=[],
last_error_case={},
)