HEL-835: day2 handoff, dual-target requirement checks with handoff

Co-authored-by: multica-agent <github@multica.ai>
This commit is contained in:
zonggong
2026-10-25 09:06:59 +08:00
co-authored by multica-agent
parent 65df25271e
commit c5167ca7c5
2 changed files with 41 additions and 0 deletions
+39
View File
@@ -0,0 +1,39 @@
{
"baseline_version": "1.10",
"read_commit": "b98ab6ab50d1631bd737dd9d4b549d9ea573d0db",
"executed_through": "65df252",
"requirement_ids": [
"CG-006", "CG-008", "CG-013", "CG-014", "CG-015", "CG-016", "CG-018", "CG-021",
"AM-001", "VIS-001", "VIS-011", "VIS-013",
"UX-001", "UX-003", "UX-005", "UX-006",
"GOV-001", "GOV-002", "GOV-003", "GOV-004", "GOV-005", "DEL-001"
],
"deliverable": "HEL-835 统一审核+隔离部署+两自然日有限真实验证:HEL-834 源码校验入库并独立复跑(225 pytest / 42 vitest / 23 浏览器合同 chromium CDP / 30 端到端 ControlledLLM);8350 独立实例部署运行(fc5e294→65df252),部署层 guard v1→v3 按两轮总管复核演进(唯一 call_id 结算、Scene 绑定分类、账本校验、done-delta 结算 ok;36/36 隔离证明),唯一权威账 guard-ledger.json 全程不清零;首日 9 次真实调用 + 夜间真实页面交互 2 轮(1 成功 1 取消)+ 3 条生活事件全部逐笔可溯;第二自然日真实隔天提问(近20轮排除生效 hits=[] 回答不编造)与已批合成 probe(隔离库 21 轮合成填充+手写 gist,真实模型 trigram 检索命中窗口外目标记忆并注入,回复精确引用记忆细节)完成。",
"acceptance": [
"guard v3 隔离证明 36/36(同秒乱序结算/毒化边界不可绕过/账本有效性/取消关闭关上游/done-then-close ok)",
"服务器 guard SHA256 与仓库逐字节一致(v3: ee6c533d…),双目标需求检查含本 handoff 通过(59项)",
"隔天真实提问:hits=[] mode=none(近20轮排除生效),回答依赖近期上下文且诚实不编造;2 条新记忆压缩 done",
"合成 probe:mode=trigram 命中目标 id=1(窗口外),回复精确引用仅存在于记忆的『蓝灰色封面』;底色指纹不变;probe 进程同轮启动并停止",
"最终权威账:主聊 6/6、gist 8/8、人格 1/1 全部用尽且零超限零拒绝外调用;所有真实调用可溯到作业/消息行与 SSE 证据"
],
"deferred": [
"自然积累的真实长期记忆隔天取回未验(需真实累计>20轮后另卡承接,不改生产检索窗口)",
"真实视频口型、R4 主动联系、R5 语音、R6 天气素材完整备份、R7 授权观察仍由主卡 HEL-797 追踪;HEL-831 三天观察保持原实例",
"8350 后续真实对话需新授权额度(当前 6/6、8/8 已满,guard 会拒绝并零上游);PR#10 待用户验收后处置,禁止合并且未合并",
"guard v2 时期结算 outcome 语义失真(成功被标 cancelled)已由 v3 修正,历史 calls 原样保留不回写"
],
"requirement_changes": [],
"verification": [
"docs/r3/:e2e-HEL-835(30/30)、guard-test-HEL-835(36/36)、browser-HEL-835(23/23 chromium)、pytest-rerun(225 passed)、web-test(42 passed)、DEPLOY-HEL-835.md、PROBE-runbook-HEL-835.md、probe-seed-HEL-835.py",
"服务器 /opt/cybergril-hel835-data/evidence/:day1-summary.json(含 day2 审计)、01–09 全部原始 SSE/确认/账本/probe 日志;guard-ledger.json 唯一权威账",
"PR#10 head 已 API 核对同步至最终提交;base agent/agent/de8874a3803b",
"真实模型调用总计:首日 9 + 夜间 5(1主聊成功+1取消+3生活)+ 隔日 4(1主聊+2gist+1判定)+ probe 1 = 19 次,全部计入权威账"
],
"product_target": "origin/agent/agent/de8874a3803b",
"branch": "hel-835-unified-review",
"status": "in_progress",
"deployed": true,
"deploy_endpoint": "http://192.168.200.12:8350",
"real_model_calls_total": 19,
"pr_url": "https://gitea.xbay.cc/leefer/CyberGril/pulls/10"
}
+2
View File
@@ -0,0 +1,2 @@
需求结构检查通过:59项;来源和覆盖完整。仍须人工核对授权语义与实际体验。
需求结构检查通过:59项;来源和覆盖完整。仍须人工核对授权语义与实际体验。