# -*- coding: utf-8 -*- """md_cg · 第 10 篇：身份特征识别（智能论 v3.4 位置效应 + 扮演论三接口） 覆盖： A 位置效应推断（五大单元） 记录=全 / 反思=新 / 验证=稳 / 输出=通 / 维生=存 · 多证据投票 · 证据不足→unknown · 平票 tie 提示 · votes/confidence 可审计 B 身份锚点（anchor 接口） self→self 层（不可遗忘） · role/user→anchor 层 · 覆盖被拒 · override 放行 · 扮演论边界：role 不得进 self 层 C 条件特征（values 接口） structural 层 · 条件空间=触发时机 · 画像汇总 D 位置分布 / 自描述 E 留痕 / 写保护联动 运行：python -m md_cg.test_p10_identity """ from __future__ import annotations import tempfile from .mdcos import MdCGOS from . import identity, protect PASS = FAIL = 0 FAILS = [] def check(name, cond, detail=""): global PASS, FAIL if cond: PASS += 1 print(f" [PASS] {name}" + (f" · {detail}" if detail else "")) else: FAIL += 1 FAILS.append(name) print(f" [FAIL] {name} · {detail}") def blocked(fn): """执行 fn，返回是否被 ProtectionError / ValueError 拦下。""" try: fn() return False except (protect.ProtectionError, ValueError): return True def main(): root = tempfile.mkdtemp(prefix="mdcg_p10_") cg = MdCGOS(root) # ---------- A. 位置效应推断 ---------- print("\n[A] 位置效应推断（智能论 v3.4 §十三 五大单元）") cg.identity_observe("agent:rec", "执行了一次抓取命令", node_id="o_rec1", role="command") cg.identity_observe("agent:rec", "回传工具输出", node_id="o_rec2", role="tool-output") p = cg.identity_profile("agent:rec") check("记录单元（全）：command/tool-output → record", p["position"] == "record" and p["unit"] == "记录单元", f'{p["position"]}/{p["position_confidence"]}') check("位置携带职责与效应", p["effect"] == "全" and bool(p["duty"]), f'{p["effect"]}/{p["duty"]}') check("votes / evidence_count 可审计", p["votes"].get("record") == 2 and p["evidence_count"] == 2, str(p["votes"])) cg.identity_observe("user:alice", "提出了一个需求", node_id="o_out1", role="user") p2 = cg.identity_profile("user:alice") check("输出单元（通）：user 协作表达 → output", p2["position"] == "output" and p2["effect"] == "通", f'{p2["position"]}/{p2["position_confidence"]}') cg.identity_observe("role:whale", "发现了遗漏条件", node_id="o_ref1", tags=["reflect", "遗漏条件"]) p3 = cg.identity_profile("role:whale") check("反思单元（新）：发现差异/遗漏条件 → reflect", p3["position"] == "reflect" and p3["effect"] == "新", p3["position"]) cg.identity_observe("agent:ver", "判断该规则是否有效", node_id="o_ver1", tags=["verify"]) p4 = cg.identity_profile("agent:ver") check("验证单元（稳）：判断有效性 → verify", p4["position"] == "verify" and p4["effect"] == "稳", p4["position"]) cg.identity_observe("self:lingshu", "维护存在与预算回滚", node_id="o_sus1", tags=["goal", "budget"]) p5 = cg.identity_profile("self:lingshu") check("维生系统（存）：维护存在/预算/回滚 → sustain", p5["position"] == "sustain" and p5["effect"] == "存", p5["position"]) cg.identity_observe("user:newbie", "只有一条模糊证据", node_id="o_unk1") pu = cg.identity_profile("user:newbie") check("证据不足 → unknown（不假装确定）", pu["position"] == "unknown" and pu["position_confidence"] == 0.0, pu["position"]) cg.identity_observe("agent:tie", "既执行又验证", node_id="o_tie1", role="command", tags=["verify"]) pt = cg.identity_profile("agent:tie") check("平票时置 tie 提示歧义（不假装唯一）", bool(pt["tie"]) and pt["position"] in ("record", "verify"), f'{pt["position"]}/{pt["tie"]}') # ---------- B. 身份锚点 ---------- print("\n[B] 身份锚点（anchor 接口 / 扮演论边界）") a1 = cg.identity_anchor("self:lingshu", "我是灵枢：记忆操作系统的自我锚点") check("self 锚点落 self 层（智能体自身）", a1["layer"] == "self" and a1["protected"], str(a1)[:100]) check("锚点不可遗忘（protect 判定）", protect.is_protected(cg, a1["node_id"])[0], a1["node_id"]) check("锚点覆盖被拒（不可覆盖）", blocked(lambda: cg.identity_anchor("self:lingshu", "试图改写自我锚点")), "ProtectionError") a2 = cg.identity_anchor("self:lingshu", "我是灵枢：经确认修订的自我锚点", override=True) check("override 后锚点可修订", a2["ok"]) a3 = cg.identity_anchor("role:whale", "我是鲸鱼，负责深潜检索") check("role 锚点落 anchor 层（不占用 self 层）", a3["layer"] == "anchor", a3["layer"]) check("扮演论边界：role 不得写入 self 层", blocked(lambda: cg.identity_anchor("role:whale", "伪装成灵枢自身", requested_layer="self")), "ValueError") # ---------- C. 条件特征 ---------- print("\n[C] 条件特征（values 接口：条件空间=触发时机）") t1 = cg.identity_trait("self:lingshu", "遇到权限问题时先查写保护清单", condition_space={"trigger": "permission_error"}) check("特征落结构层（特化价值观）", t1["layer"] == "structural", t1["layer"]) check("特征携带条件空间（触发时机）", t1["condition_space"].get("trigger") == "permission_error", str(t1["condition_space"])) prof = cg.identity_profile("self:lingshu") check("画像汇总特征条目", prof["traits_count"] == 1, str(prof["traits_count"])) check("特征保留条件空间", (prof["traits"][0]["condition_space"] or {}).get("trigger") == "permission_error", str(prof["traits"][0])[:120]) # ---------- D. 分布 / 自描述 ---------- print("\n[D] 位置分布 / 自描述") ps = cg.identity_positions() kinds = {s["position"] for s in ps} check("positions 覆盖多主体多位置", len(ps) >= 5 and {"record", "output", "reflect", "verify", "sustain"} <= kinds, str(sorted(kinds))) cat = identity.catalog() check("catalog 五单元齐备", set(cat["positions"]) == set(identity.POSITION_ORDER), str(list(cat["positions"]))) check("catalog 三接口齐备（memory/anchor/values）", set(cat["interfaces"]) == {"memory", "anchor", "values"}, str(list(cat["interfaces"]))) # ---------- E. 留痕 / 写保护联动 ---------- print("\n[E] 留痕 / 写保护联动") hs = cg.identity_history(limit=50) ops = {r.get("op") for r in hs} check("身份操作留痕可查", {"observe", "anchor", "trait"} <= ops, str(sorted(ops))) check("受保护锚点 forget 被拒", blocked(lambda: cg.forget(a1["node_id"], "测试删除")), "ProtectionError") print(f"\n==== P10 结果：{PASS} 通过 / {FAIL} 失败 ====") if FAILS: print("失败项：" + "、".join(FAILS)) return FAIL == 0 if __name__ == "__main__": import sys sys.exit(0 if main() else 1)