docs(release): v1.8.0 发布说明 + 本体导入设计约束落地记录 + 对标分析

- 新增 docs/ONTOLOGY_IMPORT.md:本体 schema 导入的硬约束(fail-fast 闸门、整事务、
  幂等键、IRI 双空间映射、不做范围对账删除),并记录实现落点与逐条偏差
- 新增 docs/auto-ontology-benchmark-20261106.md:NVIDIA auto-ontology 对标分析与
  Top-8 路线图,本版已落地项逐条标注
- docs/README.md 同步 API 清单与 MCP 工具表(54 个:只读 27 / 写入 27)
- 项目 CLAUDE.md 补「本体 schema 导入」小节:闸门顺序、事务与串行理由、幂等表、
  RDF Schema 命名空间年份坑
- README.md 镜像标签更新到 1.8.0,本地构建镜像补上 Linux 静态二进制交叉编译步骤,
  健康检查端点改回 /api/v1/version
- scripts/ 收拢本体 v2 迁移与分级校正脚本(一次性运维工具,不参与构建)

Co-Authored-By: Claude Code <[email protected]>
This commit is contained in:
adminandClaude Code committed 2026-11-06 13:53:25 +08:00
1 parent d14440a3ed
commit aef8b83e24
24 files changed
+8460 -20

No files matched your search

+139
View File
@@ -0,0 +1,139 @@
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
P1-④ 实例级细粒度分级:敏感实例加 object_markings 覆盖层(自愈/幂等)。
两层分级模型(server/classification.go + handlers_objects.go):
1) objects.classification 列 = 类型默认级(recompute 物化),由 classificationCond 强制。
2) object_markings 表 = 实例级「额外覆盖层」,由 visibilityCond(CanRead) 强制。
两者都要过:实例 marking 要求主体 clearance 包含其全部 marking。
-> 给实例打 ["L4"] 即在列值之上再叠加一层「仅 L4+ 可见」限制。
规则:
- Standard (列默认 L0/公开) 正向命中敏感词 -> ["L3"] (员工可见, 公开/connector 不可见)
- PolicyDocument (列默认 L3) 正向命中敏感词 -> ["L4"] (仅 manager/ingest 可见)
- UsageControlPolicy(列默认 L3) 全部 -> ["L4"] (访问控制策略, 默认受限)
敏感词排除「非」前缀否定语境(如「非涉密」不计为正向命中)。
自愈:
- 正向命中 -> 确保 object_markings = [目标级](已是则跳过)。
- 未命中 -> 若当前有 object_markings(上一轮误标),PUT [] 移除覆盖层,回落到列默认级;
当前本就无标记(由列默认级治理)则不动。
"""
import sys, json, argparse
sys.path.insert(0, "scripts")
import raise_employee_clearance as r
SENSITIVE = ["涉密", "机密", "秘密", "保密", "内部", "受限", "敏感", "军事",
"国防", "绝密", "商密", "密级", "secret", "internal", "confidential",
"restricted", "classified", "sensitive"]
CONFIG = {
"Standard": {"scan": True, "level": "L3", "reason": "内部/涉密标准(实际扫描0命中, 保持L0公开)"},
"PolicyDocument": {"scan": True, "level": "L4", "reason": "仅国家秘密等真涉密文件升L4; 含'非涉密'否定语境的不升"},
# UsageControlPolicy 不在此处处理:用户确认降到 L3(经理以上也可治理),
# 由 scripts/revert_ucp_level.py 维护(实例级移除L4覆盖层, 回落类型默认L3)。
}
def list_instances(typ):
out, offset = [], 0
while True:
code, data = r.api("GET", f"/objects?type={typ}&limit=1000&offset={offset}")
if code != 200 or not data:
break
arr = data.get("items", data) if isinstance(data, dict) else data
if not arr:
break
out.extend(arr)
if len(arr) < 1000:
break
offset += 1000
return out
def text_of(o):
parts = [str(o.get(k, "") or "") for k in
("name", "title", "title_cn", "title_en", "standard_code",
"doc_no", "publisher", "issuer", "scope", "classification")]
pj = o.get("properties_json") or o.get("properties")
if isinstance(pj, dict):
parts += [str(v) for v in pj.values()]
elif isinstance(pj, str):
parts.append(pj)
return " ".join(parts).lower()
def positive_hit(txt):
hits = []
for kw in SENSITIVE:
k = kw.lower()
start = 0
found = False
while True:
i = txt.find(k, start)
if i < 0:
break
if i > 0 and txt[i - 1] == "非": # 否定语境,不算正向命中
start = i + len(k)
continue
found = True
start = i + len(k)
if found:
hits.append(kw)
return hits
def get_current_markings(typ, pk):
code, data = r.api("GET", f"/access/objects/{typ}/{pk}/markings")
if code == 200 and isinstance(data, dict):
return data.get("markings", []) or []
return []
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--dry-run", action="store_true", help="只盘点不写入")
args = ap.parse_args()
plan, applied, reverted = [], 0, 0
for typ, cfg in CONFIG.items():
insts = list_instances(typ)
for o in insts:
pk = o.get("pk")
name = o.get("name") or o.get("title_cn") or o.get("title") or pk
txt = text_of(o)
cur = get_current_markings(typ, pk)
hits = positive_hit(txt) if cfg["scan"] else ["(全部)"]
if hits:
desired = [cfg["level"]]
note = "升" + str(hits)
else:
# 未命中:仅当本就被误升到目标级(覆盖层)时才回落移除;
# 其余(如 recompute 物化的列默认级标记)保持原样不动。
desired = [] if cur == [cfg["level"]] else cur
note = "回落(移除误标)" if desired == [] else "保持"
if cur == desired:
continue
plan.append((typ, pk, name, desired, cur, note))
if args.dry_run:
continue
code, _ = r.api("PUT", f"/access/objects/{typ}/{pk}/markings",
{"markings": desired})
if code == 200:
if desired:
applied += 1
else:
reverted += 1
else:
print(f" ✗ 写入失败 {typ}/{pk}: HTTP {code}", file=sys.stderr)
print(f"[{typ}] 实例总数={len(insts)} 目标级={cfg['level']} ({cfg['reason']})")
print(f"\n=== 待处理 {len(plan)} 条(升={sum(1 for p in plan if p[3])} 移除误标={sum(1 for p in plan if not p[3])})===")
for typ, pk, name, desired, cur, note in plan[:60]:
print(f" {typ}/{pk} {cur} -> {desired or '[]'} {note} {name!r}")
if len(plan) > 60:
print(f" ... 其余 {len(plan)-60} 条省略")
if args.dry_run:
print("\n[DRY-RUN] 未写入。去掉 --dry-run 执行。")
else:
print(f"\n已写入:加覆盖层 {applied} 条,移除误标 {reverted} 条(幂等)。")
if __name__ == "__main__":
main()