206 lines
7.3 KiB
Python
206 lines
7.3 KiB
Python
from collections.abc import Callable
|
||
|
||
from django.utils import timezone
|
||
|
||
from agent_core.orchestrator import run_agent
|
||
from agent_core.results import AgentResult
|
||
from apps.audit.services import create_audit_log, create_notification_record
|
||
from apps.scenarios.services import get_scenario
|
||
|
||
from .models import Conversation
|
||
|
||
|
||
def create_conversation_for_batch(batch_id: str, product_name: str) -> Conversation:
|
||
"""
|
||
为资料包创建主会话。
|
||
|
||
会话标题固定优先使用解析出的产品名称，
|
||
缺失时回退到批次号，确保前台始终有稳定标题。
|
||
"""
|
||
conversation = Conversation.objects.create(
|
||
conversation_id=_generate_conversation_id(),
|
||
title=product_name or f"未命名资料包-{batch_id}",
|
||
product_name=product_name,
|
||
batch_id=batch_id,
|
||
task_status=Conversation.STATUS_PENDING,
|
||
node_results=_build_initial_node_results(),
|
||
)
|
||
return conversation
|
||
|
||
|
||
def create_knowledge_conversation() -> Conversation:
|
||
"""
|
||
创建未绑定资料包的知识库问答会话。
|
||
|
||
该会话用于用户尚未上传资料时直接向 RAG 知识库提问，
|
||
因此 batch_id 与 product_name 保持为空，Agent Core 通过空范围执行全局检索。
|
||
"""
|
||
return Conversation.objects.create(
|
||
conversation_id=_generate_conversation_id(),
|
||
title="知识库问答会话",
|
||
product_name="",
|
||
batch_id="",
|
||
task_status=Conversation.STATUS_PENDING,
|
||
node_results=_build_knowledge_node_results(),
|
||
)
|
||
|
||
|
||
def execute_conversation_agent(
|
||
*,
|
||
conversation: Conversation,
|
||
message: str,
|
||
document_ids: list[int],
|
||
detail_url_builder: Callable[[int], str] | None = None,
|
||
) -> tuple[AgentResult, object]:
|
||
"""
|
||
在服务层串起会话执行、审计留痕与通知落库。
|
||
|
||
View 只负责收集请求参数和渲染结果，不直接承载 Agent Core 编排。
|
||
"""
|
||
scenario = get_scenario("document_review")
|
||
try:
|
||
result = run_agent(
|
||
scenario,
|
||
message,
|
||
options={
|
||
"conversation_id": conversation.conversation_id,
|
||
"batch_id": conversation.batch_id,
|
||
"product_name": conversation.product_name,
|
||
"document_ids": document_ids,
|
||
},
|
||
)
|
||
except Exception as exc:
|
||
result = AgentResult(status="failed", error=str(exc), answer="")
|
||
|
||
audit_log = create_audit_log(
|
||
"document_review",
|
||
"注册审核智能体",
|
||
message,
|
||
result,
|
||
batch_id=conversation.batch_id,
|
||
conversation_id=conversation.conversation_id,
|
||
product_name=conversation.product_name,
|
||
)
|
||
_apply_agent_result_to_conversation(conversation, result)
|
||
detail_url = detail_url_builder(audit_log.id) if detail_url_builder else ""
|
||
_persist_notification_records(result, web_detail_url=detail_url)
|
||
return result, audit_log
|
||
|
||
|
||
def execute_conversation_export(*, batch, conversation: Conversation) -> dict:
|
||
"""
|
||
在服务层串起 Word 导出、会话摘要更新和审计留痕。
|
||
|
||
View 只负责提示成功/失败消息，不直接承载导出编排细节。
|
||
"""
|
||
from .export_service import (
|
||
generate_registration_export,
|
||
update_conversation_with_export_report,
|
||
)
|
||
|
||
upstream_summary = (
|
||
(conversation.latest_summary or {}).get("upstream_structured_output")
|
||
or (conversation.latest_summary or {}).get("structured_output")
|
||
or {}
|
||
)
|
||
export_report = generate_registration_export(
|
||
batch=batch,
|
||
conversation=conversation,
|
||
upstream_summary=upstream_summary,
|
||
)
|
||
update_conversation_with_export_report(conversation, export_report)
|
||
audit_log = create_audit_log(
|
||
"document_review",
|
||
"Word 回填导出",
|
||
"生成 Word 导出文件",
|
||
AgentResult(
|
||
answer=export_report.get("summary", ""),
|
||
structured_output=export_report,
|
||
status="success",
|
||
conversation_id=conversation.conversation_id,
|
||
batch_id=conversation.batch_id,
|
||
product_name=conversation.product_name,
|
||
node_results=conversation.node_results,
|
||
),
|
||
batch_id=conversation.batch_id,
|
||
conversation_id=conversation.conversation_id,
|
||
product_name=conversation.product_name,
|
||
)
|
||
return {
|
||
"export_report": export_report,
|
||
"audit_log": audit_log,
|
||
}
|
||
|
||
|
||
def _generate_conversation_id() -> str:
|
||
return f"conv-{Conversation.objects.count() + 1:03d}"
|
||
|
||
|
||
def _build_initial_node_results() -> list[dict]:
|
||
return [
|
||
{"code": "package_import", "label": "资料包导入", "status": "已完成"},
|
||
{"code": "overview", "label": "目录汇总", "status": "处理中"},
|
||
{"code": "completeness", "label": "法规完整性检查", "status": "待处理"},
|
||
{"code": "field_extraction", "label": "字段抽取", "status": "待处理"},
|
||
{"code": "consistency", "label": "一致性核查", "status": "待处理"},
|
||
{"code": "risk", "label": "风险预警", "status": "待处理"},
|
||
{"code": "word_export", "label": "Word 回填导出", "status": "待处理"},
|
||
{"code": "feishu_notify", "label": "飞书通知", "status": "待处理"},
|
||
]
|
||
|
||
|
||
def _build_knowledge_node_results() -> list[dict]:
|
||
return [
|
||
{"code": "knowledge_retrieval", "label": "知识库检索", "status": "待处理"},
|
||
{"code": "answer_generation", "label": "问答生成", "status": "待处理"},
|
||
{"code": "risk", "label": "风险预警", "status": "待处理"},
|
||
]
|
||
|
||
|
||
def _persist_notification_records(result: AgentResult, *, web_detail_url: str = "") -> None:
|
||
payload = result.notification_payload or {}
|
||
owners = payload.get("owners") or []
|
||
if not owners:
|
||
return
|
||
resolved_detail_url = payload.get("web_detail_url") or web_detail_url
|
||
resolved_message_status = payload.get("message_status") or (
|
||
"sent" if result.status == "success" else "failed"
|
||
)
|
||
resolved_receipt = payload.get("receipt") or {"status": result.status}
|
||
for owner in owners:
|
||
create_notification_record(
|
||
batch_id=payload.get("batch_id", ""),
|
||
conversation_id=payload.get("conversation_id", ""),
|
||
product_name=payload.get("product_name", ""),
|
||
trigger_source="agent_execution",
|
||
notify_reason=payload.get("notify_reason", "task_completed"),
|
||
owner_role=owner.get("owner_role", ""),
|
||
feishu_user_id=owner.get("feishu_user_id", ""),
|
||
message_status=resolved_message_status,
|
||
web_detail_url=resolved_detail_url,
|
||
receipt=resolved_receipt,
|
||
)
|
||
|
||
|
||
def _apply_agent_result_to_conversation(conversation: Conversation, result: AgentResult) -> None:
|
||
conversation.task_status = result.status
|
||
if result.node_results:
|
||
conversation.node_results = result.node_results
|
||
conversation.latest_summary = {
|
||
"answer": result.answer,
|
||
"status": result.status,
|
||
"error": result.error,
|
||
"structured_output": result.structured_output,
|
||
"notification_payload": result.notification_payload,
|
||
}
|
||
conversation.last_run_at = timezone.now()
|
||
conversation.save(
|
||
update_fields=[
|
||
"task_status",
|
||
"node_results",
|
||
"latest_summary",
|
||
"last_run_at",
|
||
"updated_at",
|
||
]
|
||
)
|