Jeason
b7a27b12bd
Explorer 的 system prompt 明确告知 sandbox 规则 — "每条 SQL 必须包含聚合函数或 LIMIT"，减少 LLM 生成违规 SQL 浪费轮次 LLM 客户端单例 — 所有组件共享一个 openai.OpenAI 实例，不再各建各的 sanitize 顺序修复 — 小样本抑制放在 float round 之前，避免被 round 干扰 quick_detect 从 O(n²) 改为 O(n) — 按列聚合一次，加去重，不再对每行重复算整列统计 历史上下文实际生效 — get_context_for 的结果现在会注入到 Explorer 的初始 prompt 里，多轮分析时 LLM 能看到之前的发现
137 lines
4.5 KiB
Python
137 lines
4.5 KiB
Python
"""
|
||
交互式 CLI —— 四层架构自适应分析
|
||
用法: python cli.py [数据库路径]
|
||
"""
|
||
import os
|
||
import sys
|
||
|
||
sys.path.insert(0, os.path.dirname(__file__))
|
||
|
||
from core.config import DB_PATH, LLM_CONFIG, MAX_EXPLORATION_ROUNDS, PLAYBOOK_DIR
|
||
from agent import DataAnalysisAgent
|
||
|
||
|
||
def print_help():
|
||
print("""
|
||
可用命令:
|
||
<问题> 分析一个问题
|
||
rounds=<N> <问题> 设置探索轮数
|
||
report [主题] 整合所有分析，生成综合报告
|
||
schema 查看数据库 Schema
|
||
playbooks 查看已加载的预设剧本
|
||
regen 重新生成预设剧本
|
||
history 查看分析历史
|
||
audit 查看 SQL 审计日志
|
||
clear 清空分析历史
|
||
help 显示帮助
|
||
quit / q 退出
|
||
""")
|
||
|
||
|
||
def main():
|
||
db_path = sys.argv[1] if len(sys.argv) > 1 else DB_PATH
|
||
|
||
if not os.path.exists(db_path):
|
||
print(f"❌ 数据库不存在: {db_path}")
|
||
sys.exit(1)
|
||
|
||
if not LLM_CONFIG["api_key"]:
|
||
print("⚠️ 未配置 LLM_API_KEY")
|
||
sys.exit(1)
|
||
|
||
agent = DataAnalysisAgent(db_path)
|
||
|
||
print("=" * 60)
|
||
print(" 🤖 数据分析 Agent —— 四层架构")
|
||
print("=" * 60)
|
||
print(f"\n🔗 LLM: {LLM_CONFIG['model']} @ {LLM_CONFIG['base_url']}")
|
||
print(f"🔄 最大探索轮数: {MAX_EXPLORATION_ROUNDS}")
|
||
print(f"💾 数据库: {db_path}")
|
||
print(f"📋 预设剧本: {len(agent.playbook_mgr.playbooks)} 个")
|
||
print(f"\n💬 输入分析问题（help 查看命令）\n")
|
||
|
||
while True:
|
||
try:
|
||
user_input = input("📊 > ").strip()
|
||
except (EOFError, KeyboardInterrupt):
|
||
print("\n👋 再见！")
|
||
break
|
||
|
||
if not user_input:
|
||
continue
|
||
|
||
cmd = user_input.lower()
|
||
|
||
if cmd in ("quit", "exit", "q"):
|
||
print("👋 再见！")
|
||
break
|
||
elif cmd == "help":
|
||
print_help()
|
||
elif cmd == "schema":
|
||
print(agent.get_schema())
|
||
elif cmd == "history":
|
||
print(agent.get_history())
|
||
elif cmd == "audit":
|
||
print(agent.get_audit())
|
||
elif cmd == "clear":
|
||
agent.clear_history()
|
||
print("✅ 历史已清空")
|
||
elif cmd.startswith("report"):
|
||
topic = user_input[6:].strip()
|
||
try:
|
||
report = agent.full_report(question=topic)
|
||
print("\n" + report)
|
||
print("\n" + "~" * 60)
|
||
except Exception as e:
|
||
print(f"\n❌ 报告整合出错: {e}")
|
||
import traceback
|
||
traceback.print_exc()
|
||
elif cmd == "playbooks":
|
||
if not agent.playbook_mgr.playbooks:
|
||
print("（无预设剧本，输入 regen 让 AI 自动生成）")
|
||
else:
|
||
for i, pb in enumerate(agent.playbook_mgr.playbooks, 1):
|
||
print(f" {i}. 📋 {pb.name} — {pb.description} ({len(pb.preset_queries)} 条预设)")
|
||
elif cmd == "regen":
|
||
if os.path.isdir(PLAYBOOK_DIR):
|
||
for f in os.listdir(PLAYBOOK_DIR):
|
||
if f.startswith("auto_") and f.endswith(".json"):
|
||
os.remove(os.path.join(PLAYBOOK_DIR, f))
|
||
agent.playbook_mgr.playbooks.clear()
|
||
print("🤖 AI 正在重新生成预设剧本...")
|
||
generated = agent.playbook_mgr.auto_generate(agent.schema_text, save_dir=PLAYBOOK_DIR)
|
||
if generated:
|
||
print(f"✅ 生成 {len(generated)} 个剧本：")
|
||
for pb in generated:
|
||
print(f" 📋 {pb.name} — {pb.description}")
|
||
else:
|
||
print("⚠️ 生成失败")
|
||
else:
|
||
# 解析 rounds=N
|
||
max_rounds = MAX_EXPLORATION_ROUNDS
|
||
question = user_input
|
||
if "rounds=" in question.lower():
|
||
parts = question.split("rounds=")
|
||
question = parts[0].strip()
|
||
try:
|
||
max_rounds = int(parts[1].strip().split()[0])
|
||
except (ValueError, IndexError):
|
||
pass
|
||
|
||
try:
|
||
report = agent.analyze(question, max_rounds=max_rounds)
|
||
print("\n" + report)
|
||
print("\n" + "~" * 60)
|
||
except Exception as e:
|
||
print(f"\n❌ 分析出错: {e}")
|
||
import traceback
|
||
traceback.print_exc()
|
||
|
||
print("\n📋 本次会话审计:")
|
||
print(agent.get_audit())
|
||
agent.close()
|
||
|
||
|
||
if __name__ == "__main__":
|
||
main()
|