Files
aiagent/backend/app/models/agent_execution_log.py
renjianbo ab1589921a fix: 修复35个安全与功能缺陷,补全知识进化/数字孪生/行为采集模块
## 安全修复 (12项)
- Webhook接口添加全局Token认证,过滤敏感请求头
- 修复JWT Base64 padding公式,防止签名验证绕过
- 数据库密码/飞书Token从源码移除,改为环境变量
- 工作流引擎添加路径遍历防护 (_resolve_safe_path)
- eval()添加模板长度上限检查
- 审批API添加认证依赖
- 前端v-html增强XSS转义,console.log仅开发模式输出
- 500错误不再暴露内部异常详情

## Agent运行时修复 (7项)
- 删除_inject_knowledge_context中未定义db变量的finally块
- 工具执行添加try/except保护,异常不崩溃Agent
- LLM重试计入budget计数器
- self_review异常时passed=False
- max_iterations截断标记success=False
- 工具参数JSON解析失败时记录警告日志
- run()开始时重置_llm_invocations计数器

## 配置与基础设施
- DEBUG默认False,SQL_ECHO独立配置项
- init_db()补全13个缺失模型导入
- 新增WEBHOOK_AUTH_TOKEN/SQL_ECHO配置项
- 新增.env.example模板文件

## 前端修复 (12项)
- 登录改用URLSearchParams替代FormData
- 401拦截器通过Pinia store统一清理状态
- SSE流超时从60s延长至300s
- final/error事件时清除streamTimeout
- localStorage聊天记录添加24h TTL
- safeParseArgCount替代模板中裸JSON.parse
- fetchUser 401时同时清除user对象

## 新增模块
- 知识进化: knowledge_extractor/retriever/tasks
- 数字孪生: shadow_executor/comparison模型
- 行为采集: behavior_middleware/collector/fingerprint_engine
- 代码审查: code_review_agent/document_review_agent
- 反馈学习: feedback_learner
- 瓶颈检测/优化引擎/成本估算/需求估算
- 速率限制器 (rate_limiter)
- Alembic迁移 015-020

## 文档
- 商业化落地计划
- 8篇docs文档 (架构/API/部署/开发/贡献等)
- Docker Compose生产配置

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-05-10 19:50:20 +08:00

58 lines
2.7 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""
Agent 执行日志模型 — 结构化记录每次 Agent 执行的完整信息
用于知识自进化系统的数据基础
"""
from sqlalchemy import Column, String, Text, Integer, DateTime, JSON, Float, Boolean
from sqlalchemy.dialects.mysql import CHAR
from app.core.database import Base
import uuid
from datetime import datetime
class AgentExecutionLog(Base):
"""Agent 每次执行的完整结构化日志"""
__tablename__ = "agent_execution_logs"
id = Column(CHAR(36), primary_key=True, default=lambda: str(uuid.uuid4()), comment="日志ID")
agent_id = Column(String(36), nullable=True, index=True, comment="Agent ID")
agent_name = Column(String(200), nullable=True, comment="Agent 名称")
goal_id = Column(String(36), nullable=True, index=True, comment="关联 Goal ID")
task_id = Column(String(36), nullable=True, index=True, comment="关联 Task ID")
user_id = Column(String(36), nullable=True, index=True, comment="用户 ID")
session_id = Column(String(100), nullable=True, comment="会话标识")
# 输入/输出
input_text = Column(Text, nullable=True, comment="用户输入文本")
output_text = Column(Text, nullable=True, comment="Agent 输出文本")
output_truncated = Column(Boolean, default=False, comment="输出是否被截断")
# 执行结果
success = Column(Boolean, default=True, comment="是否成功")
error_message = Column(Text, nullable=True, comment="错误信息")
# 性能指标
latency_ms = Column(Integer, nullable=True, comment="总耗时(ms)")
iterations_used = Column(Integer, default=0, comment="ReAct 迭代次数")
tool_calls_made = Column(Integer, default=0, comment="工具调用总次数")
# 结构化明细JSON
tool_chain = Column(JSON, nullable=True, comment="工具调用链: [{tool_name, input, output, duration_ms}]")
llm_calls = Column(JSON, nullable=True, comment="LLM调用明细: [{model, prompt_tokens, completion_tokens, latency_ms}]")
steps = Column(JSON, nullable=True, comment="执行步骤详情(精简版)")
# 模型信息
model = Column(String(100), nullable=True, comment="使用的模型")
provider = Column(String(50), nullable=True, comment="模型提供商")
# 用户反馈(后续补充)
user_rating = Column(Integer, nullable=True, comment="用户评分(1-5)")
user_feedback = Column(Text, nullable=True, comment="用户反馈文本")
# 知识提取标记
knowledge_extracted = Column(Boolean, default=False, comment="是否已提取知识")
created_at = Column(DateTime, default=datetime.now, comment="创建时间")
def __repr__(self):
return f"<AgentExecutionLog(id={self.id}, agent={self.agent_name}, success={self.success})>"