106 lines
3.5 KiB
Python
106 lines
3.5 KiB
Python
#!/usr/bin/env python3
|
||
"""
|
||
卡若AI 记忆冲突检测
|
||
|
||
在写入长期记忆（记忆.md）前，扫描已有规则/路由/偏好是否与新内容冲突。
|
||
生成 structured/memory-conflict-report.json。
|
||
|
||
用法：
|
||
python memory_conflict_check.py "新写入的内容"
|
||
python memory_conflict_check.py --file /path/to/new_content.txt
|
||
"""
|
||
|
||
import json
|
||
import re
|
||
import sys
|
||
from datetime import datetime
|
||
from pathlib import Path
|
||
|
||
MEMORY_FILE = Path("/Users/karuo/Documents/个人/1、卡若：本人/记忆.md")
|
||
REPORT_FILE = Path(__file__).parent / "structured" / "memory-conflict-report.json"
|
||
|
||
CONFLICT_PATTERNS = [
|
||
(r"默认.*(?:使用|用|改为|切换)", "默认行为变更"),
|
||
(r"禁止|不得|不要|不允许", "禁止规则"),
|
||
(r"一律|统一|强制|必须", "强制规则"),
|
||
(r"路径.*(?:改为|迁移|移动)", "路径变更"),
|
||
(r"密码|token|key|密钥", "敏感信息"),
|
||
]
|
||
|
||
|
||
def load_memory():
|
||
if not MEMORY_FILE.exists():
|
||
return ""
|
||
return MEMORY_FILE.read_text(encoding="utf-8")
|
||
|
||
|
||
def extract_rules(text):
|
||
rules = []
|
||
for line in text.splitlines():
|
||
stripped = line.strip()
|
||
if not stripped or stripped.startswith("#"):
|
||
continue
|
||
for pattern, category in CONFLICT_PATTERNS:
|
||
if re.search(pattern, stripped, re.IGNORECASE):
|
||
rules.append({"line": stripped[:120], "category": category})
|
||
break
|
||
return rules
|
||
|
||
|
||
def check_conflicts(new_content, existing_rules):
|
||
conflicts = []
|
||
new_lines = [l.strip() for l in new_content.splitlines() if l.strip()]
|
||
for new_line in new_lines:
|
||
for rule in existing_rules:
|
||
if rule["category"] in ("禁止规则", "强制规则", "默认行为变更"):
|
||
new_keywords = set(re.findall(r"[\u4e00-\u9fff]+", new_line))
|
||
old_keywords = set(re.findall(r"[\u4e00-\u9fff]+", rule["line"]))
|
||
overlap = new_keywords & old_keywords
|
||
if len(overlap) >= 3:
|
||
conflicts.append({
|
||
"new": new_line[:120],
|
||
"existing": rule["line"][:120],
|
||
"category": rule["category"],
|
||
"overlap_keywords": list(overlap)[:8],
|
||
})
|
||
return conflicts
|
||
|
||
|
||
def main():
|
||
if len(sys.argv) < 2:
|
||
print("用法: python memory_conflict_check.py \"新内容\" 或 --file path")
|
||
sys.exit(1)
|
||
|
||
if sys.argv[1] == "--file":
|
||
new_content = Path(sys.argv[2]).read_text(encoding="utf-8")
|
||
else:
|
||
new_content = sys.argv[1]
|
||
|
||
existing_text = load_memory()
|
||
existing_rules = extract_rules(existing_text)
|
||
conflicts = check_conflicts(new_content, existing_rules)
|
||
|
||
report = {
|
||
"updated": datetime.now().isoformat(),
|
||
"existing_rules_count": len(existing_rules),
|
||
"new_content_lines": len([l for l in new_content.splitlines() if l.strip()]),
|
||
"conflicts_found": len(conflicts),
|
||
"conflicts": conflicts[:20],
|
||
"status": "conflict" if conflicts else "clean",
|
||
}
|
||
|
||
REPORT_FILE.parent.mkdir(parents=True, exist_ok=True)
|
||
REPORT_FILE.write_text(json.dumps(report, ensure_ascii=False, indent=2), encoding="utf-8")
|
||
|
||
if conflicts:
|
||
print(f"⚠️ 发现 {len(conflicts)} 个潜在冲突：")
|
||
for c in conflicts[:5]:
|
||
print(f" - [{c['category']}] 新: {c['new'][:60]}… vs 已有: {c['existing'][:60]}…")
|
||
print(f"详见 {REPORT_FILE}")
|
||
else:
|
||
print(f"✅ 无冲突，可安全写入。报告: {REPORT_FILE}")
|
||
|
||
|
||
if __name__ == "__main__":
|
||
main()
|