v1.0.10 — Attribution & Data Scope / 审计归因与数据范围版

[1.0.10] - 2026-09-15

KeelBase 1.0.10 — Attribution & Data Scope / 审计归因与数据范围版
维护线第十个增量（基线 = v1.0.9 tag）：权限-2 通用数据范围 + 角色数据范围配置化（roles/permissions 表 + 动态角色规则注册表，CASL 由配置构建且 fail-safe 回退内置）· §22.19 审计归因层（AU-1 真实客户端 IP / AU-2 AI 审计补 IP / AU-5 沙盘补审计去逃逸口 / AU-6 入口来源 source）· 证据交付物层 D-1/D-2/D-3（离线自包含 HTML 报告：单动作 + 期间 + 人读决策说明）· CE-2 wire 契约收口（PC-1…PC-7：CASL 决策 / 审计查询行 / org 数据范围 / Agent 对象 / MCP 投影 / 导航对账 / Application Model 机读）· 生成器防覆盖守卫（--force 不覆盖非生成物 + 撞名拒绝）· Web 修复（i18n 响应式 / AI 入口改流式）· Redis 真接线（@keyv/redis）· 副作用哈希链写入串行化。注：本次发布随历史匿名化——竞品名从全史提交消息与历史文件内容移除（双端 force-push，tag 重指），v1.0.10 建在匿名化后的历史上。

Added / 新增

• 权限-2 通用数据范围 + 角色数据范围配置化 — src/authz/：roles / permissions / role_permissions / user_role 实体 + RoleRuleRegistry（装配期载入内存、同步读取；reload 失败 fail-safe 回退内置规则，绝不 fail-open）+ DataScopeService（角色配置范围 → ScopeDescriptor：own / own_dept_and_below，缺配置回退更紧默认）；CASL 从注册表构建（既有调用点与 spec 零改动）；events/todos 加 dept_id（写入盖章）
  Data scope：按角色配置"本人 / 本部门及以下"，改配置即生效、双向可逆
• §22.19 审计归因层（AU-1 / AU-2 / AU-5 / AU-6） — AU-1 真实客户端 IP（src/config/trust-proxy.ts，默认 1 跳、可配 CIDR，绝不盲信 XFF）；AU-2 AI 审计补 ip 列（对齐操作审计）；AU-5 沙盘真改状态补 AI 审计（effect_revoke + source='sandbox'，fail-closed 不静默丢审计）；AU-6 ActorContext.source（web/admin/headless/mcp/sandbox）+ tool_call 行补 provider
• 证据交付物层 D-1 / D-2 / D-3 — 离线自包含 HTML 报告（零外链/零依赖，仅 Node 内置）：D-2 单动作证据报告（verify-evidence.mjs --format=html）+ D-1 期间审计报告（render-period-report.mjs）+ D-3 人读「决策说明」（explainAuthorization 单一真源，随审计响应与 /2 导出的 compliance[].decisionNote 下发）
• CE-2 wire 契约收口（PC-1…PC-7） — CASL 权限决策 / 审计查询行与聚合端点 / org·orgId 数据范围 membership / Agent 一级对象 / MCP tools-list 治理投影 / AI 导航↔客户端路由双向对账 / Application Model 机读化

Fixed / 修复

• 迁移 ↔ 实体漂移消除 — 索引/FK/唯一约束对齐 TypeORM 派生名（消除 migration:generate 稳定 churn）
• 生成器防覆盖 — --force 不再覆盖非生成器产物（手写/既有模块）；目标文件撞名时拒绝（防静默覆盖旗舰/重复注册）
• Web — 语言切换后表头/筛选/状态标签不重算（i18n 常量改 computed）；三个 AI 入口改走流式（修复非流式下写操作无法弹确认框）
• Redis 缓存真接线 — 换 @keyv/redis 的 createKeyv（移除不兼容的 ioredis-yet）；未接线时显式告警而非静默走内存
• 副作用哈希链写入串行化 — 修并发分叉（镜像主审计链双保险）；单条撤销幂等守卫（已撤销/已请求补偿不重复触发外部补偿）
• 部署 — 补 docker/init 建 governance 库（全新部署下治理台/sidecar 起不来）；prod 镜像含 src/ + tsconfig（修复容器内 create-admin/seed-demo 必然失败）

Migration / 迁移

• 1820000000000-AddRolesPermissions（roles/permissions/role_permissions/user_role + pg 白名单单源）
• 1821000000000-AddDeptIdToTodosEvents（events/todos 加 dept_id）
• 1822000000000-AddAiAuditIp（ai_audit_logs 加 ip，链外列）

Release Precheck（2026-09-15）

• 全量测试：后端单测 + 覆盖率（statements≥85 门槛 + 安全模块分档门控）· 后端 e2e · Web-Admin-Vue vitest · Flutter · 生成器 node --test · Gate 1 Golden Application · 端点-文档一致性 · Release Gate 24/24 PASS（确定性） —— 全过
• 覆盖率：达标且不低于 v1.0.9
• 四层 code review：本版评审阻塞项已修复（见 fix(precheck): R-2 与运行态扫查批提交）；本次发版会话补跑全量测试与 Release Gate 全绿。阿里 OCR 层对 v1.0.9..HEAD 区间（88 提交）两次运行未产出报告——工具在区间内已删/更名路径上停滞，叠发版窗口 GitHub 网络中断；如实记录，不虚报
• 阻塞项修复：迁移↔实体漂移（f5ebd48c）、data-scope e2e 过期断言（63fb39f0）

v1.0.9 — Consulting→Build & Enterprise Proof / 业务访谈直生成与两主张同证

Highlights

• S4 Consulting→Build — a Business Spec intermediate layer + deterministic mapper + CLI + CI job: a business interview yields a Business Spec, which deterministically maps to a Module Protocol and then to a generated module (entity/DTO/API/pages/permissions/audit), verified in CI (compile / unit / migration self-consistency).
• Enterprise Proof × S5 — the Build claim and the Trust claim are now proven in a single run (verify-consulting-to-build.sh).
• T5 cross-entry decision consistency — the governance sidecar emits structured authorization (auto/confirm/block); external MCP tool denials are now audited; the allow snapshot uses a single constructor shared by SSE / non-stream / MCP.
• FE-1 frontend runtime neutrality — envelope/error/refresh converge into neutral adapters; Web / Flutter / Taro consume /app/capabilities and /app/provenance; a frontend boundary gate forbids cross-repo coupling.
• Run-level batch revoke (§4 G1) and decision vocabulary unification (approve|decline, wire Schema v2).
• NC-3 first-run readiness — GET /app/readiness (five dimensions, each with an executable next step) plus an admin-console card.

Also

• Fixed: app-version latestVersion single-sourced; NODE_ENV env-file fallback; postgres migration single-source + drift guard; evidence-root v3 fixes; run-row approval boundary; production dependency vulnerabilities.
• 15 release-precheck blockers fixed (four-layer review: OpenCodeReview over 186 files, Claude multi-dimensional, Standards+Spec, AI Code Economy).
• History cleanup (non-functional): benchmark run-artifact filenames/content work-hour timestamps removed from all history; tags re-pointed; HEAD tree unchanged.

Full details: CHANGELOG.md

---

亮点

• S4 Consulting→Build（业务访谈直生成） —— Business Spec 中间层 + 确定性映射器 + CLI + CI job：业务访谈产出 Business Spec，确定性映射为 Module Protocol 并生成模块（实体 / DTO / API / 页面 / 权限 / 审计），CI 校验（编译 / 单测 / 迁移自洽）。
• Enterprise Proof × S5 合流 —— Build 主张与 Trust 主张一次运行同证（verify-consulting-to-build.sh）。
• T5 跨入口决策一致 —— 治理 sidecar 输出结构化依据（auto/confirm/block）；外部 MCP 工具拒绝补审计；放行快照在 SSE / 非流式 / MCP 共用单一构造。
• FE-1 前端 Runtime-Neutrality —— 信封 / 错误 / 刷新收敛为 neutral adapter；Web / Flutter / Taro 消费 /app/capabilities 与 /app/provenance；前端边界门禁禁止跨仓耦合。
• run 级批量撤销（§4 G1） 与 决策词汇统一（approve|decline，wire Schema v2）。
• NC-3 首次运行就绪清单 —— GET /app/readiness（五维 + 每维可执行下一步）+ 管理台首屏卡。

其它

• 修复：app-version latestVersion 单源；NODE_ENV 环境文件回退；postgres 迁移单源 + 漂移守卫；证据根 v3 修复；审批 run 行边界；生产依赖漏洞。
• 修复本版四层评审 15 处阻塞项（OpenCodeReview 186 文件 + Claude 多维 + Standards+Spec + Code Economy）。
• 历史清理（非功能性）：benchmark 运行产物文件名 / 内容的工作时段时刻从全史移除；tag 重指；HEAD 树零变。

完整明细见 CHANGELOG.md

KeelBase v1.0.8 — Conformance & Revoke Contract / 契约常绿与撤销契约收口版

KeelBase v1.0.8 — Conformance & Revoke Contract
KeelBase v1.0.8 —— 契约常绿与撤销契约收口版

Maintenance-line eighth increment (baseline = v1.0.7 tag). Full changes in CHANGELOG [1.0.8].
维护线第八个增量（基线 = v1.0.7 tag）。完整变更见 CHANGELOG [1.0.8]。

Highlights

亮点

CE-1 Conformance ever-green: wire Schema v1 freeze (tool/SSE/confirmation/trace/side-effect/audit-payload/evidence/governance-policy/delegation + ai-tool surface + D7/E1, registry 24) + protocol-vector corpus (B1/B2) with canonical golden-sample CI drift detection + C1/C2 semantic-change discipline & single-source terminology gate
CE-1 契约常绿收口：wire Schema v1 冻结（registry 24）+ 协议语料金样本 CI 漂移检测 + 语义变更纪律与术语单一真源闸

Revoke Contract deepening (revoke-contract.spec, A–F matrix): G1 conversation-level batch revoke (BE + AI Action Center UI), G4 acceptance e2e, KB-5 run-level single-run aggregation (multiple same-round write tools authorized once)
撤销契约深化：会话级批量撤销（G1 BE+FE）+ G4 验收固化 + KB-5 单轮 run 聚合一次授权

Deferred review fixes: s5 revoke now asks before acting (no silent revocation of real AI side effects); FP-8 proxy records the proxy_call side-effect anchor even on empty success; proxy timeout read at call time; Trust-journey completion count hardened against lost-update / spam
deferred 修复：s5 撤销前确认 + FP-8 proxy 空体副作用锚 + proxy 超时调用期读 + 旅程计数防刷

Review fixes (code-review two-axis): H1 run-count copy bilingualized (no hard-coded Chinese in English UI); H2 run aggregation gates members with HS-2 before inclusion (no authorize-then-deny order); insights spec made date-deterministic
评审修复：H1 双语硬编码清除 + H2 run 门控顺序收敛 + insights 去日期脆弱

History time normalization: v1.0.7-era work-hour commits re-timestamped to off-hours; v1.0.7 tag re-pointed to the content-equivalent commit (zero leak, zero content change)
历史时间归一：v1.0.7 工作时段时间戳拨到晚间，tag 指向内容等价 commit（泄漏清零）

Release Precheck

发布前检查

Backend unit cov + e2e (302), Web-Admin-Vue vitest + coverage, Flutter (623), generator, docs-endpoint consistency, wire-schema ever-green all pass; coverage & security tier gates green; code review (Standards+Spec) run, blockers fixed.
后端 cov + e2e(302)、Vue vitest+coverage、Flutter(623)、生成器、文档一致性、wire-schema 常绿全绿；覆盖率与安全分档门控通过；code-review（双轴）已跑并修复阻塞项。

KeelBase v1.0.7 — Trust Journey & AI Proof Cards / Trust 旅程与对抗评测 Proof 收口版

KeelBase v1.0.7 — Trust Journey & AI Proof Cards
KeelBase v1.0.7 —— Trust 旅程与对抗评测 Proof 收口版

Maintenance-line seventh increment (baseline v1.0.6). Full changes in CHANGELOG [1.0.7].
维护线第七个增量（基线 v1.0.6）。完整变更见 CHANGELOG [1.0.7]。

Highlights

亮点

Trust Journey "see → act" closed loop: console entry → one-click run (Ask → human confirm → privilege-violation denied → high-risk blocked) → real Copilot approve → persisted → revoke → view evidence
Trust 旅程「从看到做」闭环：控制台直达 → 一键连跑（Ask → 人工确认 → 越权拒绝 → 高风险阻断）→ 真实 Copilot 批准落库 → 撤销 → 看证据

Operable Trust Sandbox: result panel/dialog, self data cleanup, cross-visitor journey stats on admin console
Trust 沙盘可运营：结果面板/弹窗、数据自清理、跨访客旅程统计上管理台

revokeClass revocation-capability contract (four tiers, registration validation, honest UI rendering)
revokeClass 撤销能力语义契约（四档声明 + 注册强校验 + UI 诚实渲染）

Proof Cards T1/T2/T3: strangers can reproduce verification from public docs (Stranger Kit)
Proof Cards T1/T2/T3：陌生人按公开文档可复现验证（Stranger Kit）

KB-1..8 antithesis-review wrap-up (bilingual glossary / not-claims / threat model, each tied to spec & tests)
KB-1..8 对抗评测收口（词汇表/不承诺清单/威胁模型双语化 + 挂 spec/测试引用）

Review-blocker fixes: Trust Sandbox cleanup authz containment, s5 honest revoke copy, governed dead-link removal
评审阻塞修复：Trust 沙盘 cleanup 越权收敛、s5 撤销诚实文案、治理详情去死链

AI gate fix & dependency security updates (vitest 3.2.7, Dependabot security-first)
AI 门控修复、依赖安全更新（vitest 3.2.7、Dependabot security-first）

Release Precheck

发布前检查

Backend unit 2197 + e2e, Vue 400, Flutter, generator, docs-endpoint consistency all green; coverage & security tier gates passed. Code review (Standards+Spec) run, blockers fixed.
后端单测 2197 + e2e、Vue 400、Flutter、生成器、文档-端点一致性全绿；覆盖率与安全分档门控通过；code-review（Standards+Spec）已跑并修复阻塞项。

KeelBase 1.0.6 — AI Action Center North-Star & Cross-entry Decision Consistency / AI 行为中心北极星与跨入口决策一致版

KeelBase 1.0.6 — AI Action Center North-Star & Cross-entry Decision Consistency / AI 行为中心北极星与跨入口决策一致版

The sixth patch since 1.0: AI Action Center (business-user "My AI Activity" Trust surface) + §22.17 ① evidence root v3 / G-1~G-3 / P-③ (per-action cross-chain evidence root, event-time authorization snapshots, chain-outside columns into the payload, side-effect hash chain, policy history with cross-version replay) + Policy Evidence (grant snapshots carry the policy fingerprint revision) + governance policy visual editing (gate-mode tiers / R4 two-person approval) — the evidence line is now a single complete, undeniable, reproducible explanation of one AI action + cross-entry decision consistency (T5) + executable DX errors (NC-2) (reason/impact/next-step end to end) + side-effect status reverse query (B-path reconciliation).

1.0 后第六个补丁：AI Action Center「我的 AI 行为」业务用户 Trust 面 + §22.17 ① 证据根 v3 / G-1~G-3 / P-③ 策略历史回放 + Policy Evidence（授权放行快照携带策略版本指纹）+ 治理策略可视化编辑（门控档位化 / R4 双人审批）——证据线「一条 AI 行为完整、不可抵赖地解释与复现」闭环 + 跨入口决策一致性（T5）+ 可执行 DX 错误（NC-2：reason/impact/next-step 前后端闭环）+ 副作用状态服务身份反向查询（B 路径对账）。

Highlights / 亮点

• AI Action Center — new user-scoped GET /ai/my/tool-effects (own AI write side-effect list, status executed/revoked + target enrichment, data-minimized, owner-isolated) + workbench "My AI Activity" page with per-row revoke / full evidence / object history.
  AI Action Center — 新本人端点 GET /ai/my/tool-effects（本人 AI 写副作用清单，executed/revoked + 目标富化，数据最小化）+ 工作台「我的 AI 行为」页（行内撤销 / 完整证据 / 对象历史）。
• Evidence root v3 / G-1~G-3 / P-③ — per-action cross-chain evidence root (GET /ai/governance/evidence-root/:resultType/:resultId, offline-verifiable: grant snapshot with policy.revision + Decision Evidence + same-conversation audit chain + op-audit chain + side effect, bound by a cross-chain root anchor) + REST write authorization snapshots (G-1) + HS-11 payload v2 covering chain-outside columns (G-2) + side-effect hash chain (G-3) + policy history with cross-version replay (P-③).
  证据根 v3 / G-1~G-3 / P-③ — 单动作跨链证据根（GET /ai/governance/evidence-root/:resultType/:resultId，离线可验：授权快照含 policy.revision + Decision Evidence + 同会话审计链 + 操作审计链 + 副作用，跨链根锚绑定）+ REST 写授权依据快照（G-1）+ HS-11 payload v2 链外列入链（G-2）+ 副作用哈希链（G-3）+ 策略历史跨版本回放（P-③）。
• Policy Evidence — the governance policy is content-fingerprint versioned (order-insensitive; no migration); every grant-time authorization snapshot cites authorization.policy = {revision, updatedAt} — "why allowed" points at the exact policy revision in force.
  Policy Evidence — 治理策略内容指纹版本化（与键序无关、无迁移）；授权放行快照携带 authorization.policy = {revision, updatedAt}——「当时为什么允许」指向确切策略版本。
• Governance policy visual editing — per-tool gate mode (auto/confirm/approval) tiered control in the console, R4 two-person approval escalation for any R3-and-below write tool, diff-save only overrides.
  治理策略可视化编辑 — 策略中心每工具门控档位（auto/confirm/approval）+ 任意 R3 及以下写工具可升档 R4 双人审批 + 差异保存只写覆盖。
• Cross-entry decision consistency (T5) — MCP tools/call deny now writes an AI audit row with structured reasons (parity with REST/SSE) and confirmation-gated calls are marked; non-streaming B-path proxy tools set source=bridge; deterministic cross-entry e2e registered in CI and the release gate.
  跨入口决策一致性（T5） — MCP 工具 deny 写审计 + 结构化 reasons（对齐 REST/SSE）+ confirmation 门控审计标注；非流式 B 路径补 source=bridge；cross-entry 确定性 e2e 进 CI + release-gate。
• Executable DX errors (NC-2) — error codes carry reason/impact/nextStep (zh/en); AI failures surface LLM_UNAVAILABLE (502) with a "check your key / config" hint instead of a bare 500; the Web console renders an actionable error card and the login page shows email-verification guidance inline — errors become guidance end to end.
  可执行 DX 错误（NC-2） — 错误码带 reason/impact/nextStep（中英）；AI 不可用返回 LLM_UNAVAILABLE（502）带「检查 API Key/配置」指引不再笼统 500；前端错误卡渲染 + 登录页邮箱验证引导——报错变向导。
• Side-effect status reverse query — service-identity GET /external/effects/:resultType/:resultId for Java/B-path reconciliation (local entity revoked from target soft-delete; proxy_call carries an honest revokeHint — confirm on the Java side).
  副作用状态反向查询 — 服务身份 GET /external/effects/:resultType/:resultId（B 路径 Java 对账；本地实体 revoked=目标软删；proxy_call 诚实 revokeHint「撤销态在 Java 侧」）。

KeelBase 1.0.5 — AI Follow-up & Audit Evidence Hardening / AI 主动跟进与审计证据加固版

KeelBase 1.0.5 — AI Follow-up & Audit Evidence Hardening / AI 主动跟进与审计证据加固版

The fifth patch since 1.0: AI Follow-up Agent (AI proactively finds "who to follow up") + hot-reload for AI Bridge proxy tools (no restart needed) + AI audit evidence semantics wrap-up (authorization snapshot at the moment an event is allowed · privilege-escalation attempts as first-class events · approval semantics · lifecycle el-steps) + Security Showcase productized as adversarial proof (canary: defense drift → fail-loud turns red) + codebase health audit phase 1/2 (authorization subdomain extraction that cuts the runtime import cycle) + pre-release four-layer review fixes and full regression.

1.0 后第五个补丁：AI Follow-up Agent（AI 主动发现「该跟进谁」）+ AI Bridge 代理工具免重启热更新 + AI 审计证据语义收尾（事件时点放行授权快照 · 越权尝试一级事件 · 审批语义 · 生命周期 el-steps）+ Security Showcase 对抗性证明产品化（canary：防线漂移即 fail-loud 变红）+ 全库健康体检阶段 1/2（授权子域下沉切断 import 环）+ 发布前四层评审修复与全量回归。

Highlights / 亮点

• AI Follow-up Agent (A1) — detect_idle_customers read-only tool (no follow-up in 30 days incl. never contacted, derived from activity without migration) → AI summarizes recommendations → user confirms → create_followup_task (R3 confirm + revocable) → audited. Real end-to-end verification across providers (DeepSeek/GLM/Kimi, golden-crm 8/8).
  AI Follow-up Agent（A1） — detect_idle_customers 只读工具（30 天无跟进含从未联系，activity 派生无迁移）→ AI 汇总建议 → 用户确认 → create_followup_task（R3 确认 + 可撤销）→ 审计。多 provider 真实端到端验证（DeepSeek/GLM/Kimi golden-crm 8/8）。
• Hot-reload for ai_proxy_tools without restart (AI Bridge B path) — ToolRegistry.unregister + ProxyToolRegistryService.reload + SettingsService.onChange; config changes take effect the moment they are written.
  ai_proxy_tools 免重启热更新（AI Bridge B 路径） — ToolRegistry.unregister + ProxyToolRegistryService.reload + SettingsService.onChange，写配置即生效。
• AI audit evidence semantics wrap-up (A-3/A-5/A-7/A-8) — authorization snapshot at the time a tool-call event is allowed (object format) + privilege-escalation attempts as first-class business events (escalation / high-risk block / generic block, tri-split) + correct semantics for approval decide/review into audit + governance drawer lifecycle el-steps + A-8 denied view / behavior-replay filters.
  AI 审计证据语义收尾（A-3/A-5/A-7/A-8） — 放行 tool_call 事件时点授权快照（对象格式）+ 越权尝试一级事件业务化（越权/高风险阻断/通用阻断三分）+ approval decide/review 入审计正确语义 + 治理抽屉生命周期 el-steps + A-8 denied 视图/行为回放筛选。
• Security Showcase (A2 productized) — run deterministic adversarial scenarios from the admin console in one click (injection HS-8 / CASL escalation / R5 block / R3 confirm), reusing the real defense logic with no LLM; canary: defense drift (a sample no longer hits / CASL now allows / risk tier changed) throws and turns red.
  Security Showcase（A2 产品化） — 管理台一键运行确定性对抗场景（注入 HS-8 / CASL 越权 / R5 阻断 / R3 确认），复用真实防护逻辑无 LLM；canary：防线漂移（样本不再命中 / CASL 放行 / 风险级变更）即抛错变红。
• Codebase health audit phase 1/2 — health report committed; AuthorizationExplainerService authorization subdomain extraction (ai.service net −131 lines, cuts the audit/auth reverse runtime import cycle).
  代码库健康体检阶段 1/2 — 体检报告入库；AuthorizationExplainerService 授权子域下沉（ai.service 净减 131 行，切断 audit/auth 反向运行时 import 环）。
• Evidence & protocol system — protocol-conformance in CI (22/22) + docs/evidence three-tier evidence index + offline evidence-package review archive.
  证据与协议体系 — protocol-conformance 进 CI（22/22）+ docs/evidence 证据三层索引 + 证据包离线复核留档。
• Others: bilingual demo seed across modules · workbench version badge · feature-flag nav hiding · keelbase init auto SPDX header on generated modules · CodeQL security cleanup · README governance landscape.
  其它：全模块双语演示 seed · 工作台版本徽标 · feature-flag 导航隐藏 · keelbase init 生成模块自动 SPDX 头 · CodeQL 安全清理 · README 治理全景图。

Fixed / 修复（含发布前四层评审）

• Authorization snapshot on allow now written only on the succeeded branch (fixes rejected/timeouts being miscounted as A-8/blocked).
  AI 放行授权快照仅写「成功执行」分支（修复被拒/超时被 A-8/blocked 误计）。
• Governance drawer revoke state reachable (B4 effect adds targetSoftDeleted) + authorize node status fix.
  治理抽屉撤销态可达（B4 effect 补 targetSoftDeleted）+ authorize 节点状态修正。
• AI timeline "escalation/denied" filter semantics (isError + authorization) + added "AI executed".
  AI 时间线「越权/被拒」筛选语义（isError+authorization）+ 补「AI 执行」。
• Security Showcase false-green fix (canary) + bilingual reason/trace + error handling.
  Security Showcase 假绿修复（canary）+ 双语 reason/trace + 错误处理。
• detect_idle_customers counts statistics before truncation / createdAt tie-break / limit validation.
  detect_idle_customers count 截断前统计 / createdAt tie-break / limit 校验。
• Workbench card click · flow detail 500 · notification queue hang when redis is unavailable.
  工作台卡片点击 · 流程详情 500 · redis 不可用通知队列挂起。

Release Precheck / 发布前评审（2026-09-03）

Four-layer code review (Aliyun OCR 16 + Claude multi-dimension 19 + code-review skill 7 + Code Economy WARN) → 6 blocker/high-value classes fixed; backend unit tests 1997 · e2e 260 · vitest 312 · Flutter 623 · release-gate 12/12 · coverage Stmts 90.15 (security-module gate ≥85 all passed, no regression vs 1.0.4).

四层 code review（阿里 OCR 16 + Claude 多维 19 + code-review skill 7 + Code Economy WARN）→ 修复 6 类阻塞/高价值项；后端单测 1997 · e2e 260 · vitest 312 · Flutter 623 · release-gate 12/12 · 覆盖率 Stmts 90.15（安全模块门控 ≥85 全过，较 1.0.4 无降级）。

Assets & docs: official site three-entry demo (/user/ /admin/ /mobile/); full changelog in CHANGELOG.

Assets 与文档：官方站点三入口 demo（/user/ /admin/ /mobile/）；详细变更见 CHANGELOG。

Post-release: CI supply-chain vulnerabilities cleared (fast-uri/qs/@xmldom, a9c58e5) + Docker CN-build node-gyp headers mirror build-arg (07c5009); v1.0.5 tag moved forward two commits, main CI all green.

发布后补充：CI supply-chain 依赖漏洞清零（fast-uri/qs/@xmldom，a9c58e5）+ Docker 中国区构建 node-gyp headers 镜像 build-arg（07c5009）；v1.0.5 tag 前移至两提交，main CI 全绿。

KeelBase 1.0.4 — Business Action Ledger & Governance Moat 2.x / 业务行为取证与治理护城河版

KeelBase 1.0.4 — Business Action Ledger & Governance Moat 2.x / 业务行为取证与治理护城河版

The fourth patch since 1.0: §22.16 Business Action Ledger fully landed (A-1–A-8) + Governance Moat 2.1/2.2/2.3 + business-language UX across all four ends + unified product language (Business-safe AI Runtime + DNA + Trust Manifesto) + EASY-5 first-run onboarding preset (the first cut of v1.1 P0-6) + Apache-2.0 open-source trust packaging.

1.0 后第四个补丁：§22.16 业务行为取证系统完整落地（A-1~A-8）+ 治理护城河 2.1/2.2/2.3 + 四端业务语言化 + 产品语言统一（Business-safe AI Runtime + DNA + Trust Manifesto）+ EASY-5 首启引导 preset（v1.1 P0-6 第一刀）+ Apache-2.0 开源可信包装。

• Business Action Ledger (§22.16 A-1–A-8) — field-level trace + decision rationale + business-event normalization + per-entity behavior-history ledger + three-layer audit explainer + cross-system identity chain with authorization basis + compliance evidence pack v2 + dedicated privilege-escalation view + FLOW approval chains into audit — the full business narrative of "who, on what authorization, did what, and how each level approved" is presentable live.
  业务行为取证系统（§22.16 A-1~A-8）：字段级留痕 + 决策依据 + 业务事件归一化 + 实体行为史账本 + 审计解释器三层展开 + 跨系统身份链与授权依据 + 合规证据包 v2 + 越权专门视图 + FLOW 审批链入审计——「谁、基于什么授权、做了什么、每级怎么批的」完整业务叙事可现场演示。
• Governance Moat 2.1/2.2/2.3 — protocol-conformance certification suite (verify-protocol-conformance, 22/22 tamper vectors) / MCP governance-contract exposure / real-time policy push (sidecar applies within seconds) / policy template library (finance / government / general) / compliance mapping tables (Intelligent-Agent-Interconnect guideline / EU AI Act / MLPS 等保) / offline machine verification of evidence packs.
  治理护城河 2.1/2.2/2.3：协议合规认证套件（verify-protocol-conformance 22/22 篡改向量）/ MCP 治理契约透出 / 策略实时推送（sidecar 秒级生效）/ 策略模板库（金融/政务/通用）/ 合规映射表（《智能体互联》/ EU AI Act / 等保）/ 证据包离线机器验证。
• Business-language UX across all four ends — AI audit / behavior replay / execution trace / risk center / approvals / security review / workbench Action Detail / Flutter main app — all tool actions now speak business language, technical detail folded away.
  四端业务语言化：AI 审计/行为回放/执行轨迹/风险中心/审批/安全审查/工作台 Action Detail/Flutter 主 App 全部工具动作改业务语言，技术细节折叠。
• EASY-5 first-run onboarding preset (first cut of v1.1 P0-6) — runtime FeatureFlags preset (full/small/lite) + Settings persistence + admin first-run three-card onboarding modal.
  EASY-5 首启引导 preset（v1.1 P0-6 第一刀）：FeatureFlags 运行时 preset（full/small/lite）+ Settings 持久化 + 管理台首启三卡片弹窗。
• Unified product language (P0-4) — Business-safe AI Runtime positioning + KeelBase engineering DNA + Trust Manifesto + glossary v0.1.
  产品语言统一（P0-4）：Business-safe AI Runtime 定位 + KeelBase 工程 DNA + Trust Manifesto + 词汇表 v0.1。
• Apache-2.0 open-source trust packaging — SPDX license headers on 1515 repo files + DCO sign-off convention + public versioning plan + gitleaks/CodeQL CI scanning.
  Apache-2.0 开源可信包装：全仓 1515 文件 SPDX 许可头 + DCO 签名约定 + 公开版本计划 + gitleaks/CodeQL CI 扫描。

Verification / 验证：Release Gate PASS · backend unit tests 1947 · e2e 259 · Web-Admin vitest 304 · generator/CLI 81 · security-module coverage gate 6/6 met (operation-audit 83.5→97.7) · four-layer review (Aliyun OCR + Claude built-in + code-review skill + Code Economy) blockers cleared.

验证：Release Gate PASS · 后端单测 1947 · e2e 259 · Web-Admin vitest 304 · 生成器/CLI 81 · 安全模块覆盖率门控 6/6 达标（operation-audit 83.5→97.7）· 四层评审（阿里 OCR + Claude 自带 + code-review skill + Code Economy）阻塞项清零。

Details in CHANGELOG.

详见 CHANGELOG。

KeelBase 1.0.3 — Trust & Governance Hardening / 治理加固与产品证明版

KeelBase 1.0.3 — Trust & Governance Hardening / 治理加固与产品证明版

The third patch since 1.0: Moat 2.0 (standalone governance console + sidecar zero-code adoption + one control plane across systems) + audit visualization (field-level audit / hash-chain view / evidence-pack export) + critical-path performance E-3 + product-proof pack (Trust-proof pack six scenarios / privilege-escalation V-2 / demo reset V-4) + third-party pre-release review hardening + demo videos.

1.0 后第三个补丁：护城河 2.0（独立治理台 + sidecar 零代码接入 + 多系统单控制面）+ 审计可视化（字段级审计 / 哈希链视图 / 证据包导出）+ 关键路径性能 E-3 + 产品证明包（Trust 证明包六场景 / 越权 V-2 / 演示复位 V-4）+ 发布前三方评审加固 + 演示视频。

• Trust proof pack — verify-trust-proof.mjs verifies six scenarios in one command (normal success / unauthorized 403 / R5 block / human confirmation / revocation / Java bootstrap, measured 15/15) + delete_customer R5-blocking tool + deterministic DemoProvider availability.
  Trust 证明包：verify-trust-proof.mjs 六场景一键验证（正常成功 / 越权 403 / R5 阻断 / 人工确认 / 撤销 / Java 引导，实测 15/15）+ delete_customer R5 阻断工具 + DemoProvider 确定性可用。
• Governance Moat 2.0 — standalone governance console / sidecar zero-code adoption / tool gating (S-2) / one control plane across systems (MOAT-3) / cross-service revocation and approval callbacks.
  治理护城河 2.0：独立治理台 / sidecar 零代码接入 / 工具门控（S-2）/ 多系统单控制面（MOAT-3）/ 跨服务撤销·审批回调。
• Audit visualization E-1/E-2/D4 — field-level change audit / hash-chain view / behavior replay / HMAC evidence-pack export.
  审计可视化 E-1/E-2/D4：字段级变更审计 / 哈希链视图 / 行为回放 / HMAC 证据包导出。
• Performance E-3 + AI-23 content-safety deepening + frontend RBAC permission points + CLI doctor/next-steps + weapp pass.
  性能 E-3 + AI-23 内容安全深度化 + 前端 RBAC 权限点 + CLI doctor/next-steps + weapp 专项。
• Third-party review hardening (Aliyun OCR + Claude built-in + code-review skill) — 20+ items incl. hash-chain row locks / sqlite concurrency for operation audit / side-effect type mapping / audit trends / sidecar authentication / migration allow-list and consistency.
  三方评审加固（阿里 ocr + Claude 自带 + code-review skill）：哈希链锁行 / 操作审计 sqlite 并发 / 副作用类型映射 / 审计趋势 / sidecar 鉴权 / 迁移白名单与一致性等 20+ 项。

Verification / 验证：Release Gate 12/12 PASS · backend unit tests 1882 · e2e 258 · Web-Admin vitest 296 · security-module coverage gate 6/6 met.

验证：Release Gate 12/12 PASS · 后端单测 1882 · e2e 258 · Web-Admin vitest 296 · 安全模块覆盖率门控 6/6 达标。

Details in CHANGELOG.

详见 CHANGELOG。

KeelBase Demo Videos 官方演示视频

Official bilingual demo videos (Chinese & English, ~4min each): AI golden flow (analyze → confirm → create → audit → revoke) + overreach denial + evidence/admin-console + Build (keelbase init) + AI Bridge + private deploy.

中英双语官方演示视频（各约 4 分钟）：AI 黄金流程（分析→确认→创建→审计→撤销）+ 越权拒绝 + 管理台安全验证 + Build（keelbase init）+ AI Bridge + 私有部署。

KeelBase v1.0.2 — Governance & Product-Proof Release / 治理深化与产品证明版

KeelBase 1.0.2 — Governance & Product-Proof Release / 治理深化与产品证明版

The second patch since 1.0 (180 commits after 1.0.1). v1.1 is still not triggered (the four product-proof-phase acceptance criteria are not yet met).

1.0 后第二个补丁（1.0.1 后 180 提交）。1.1 仍未触发（产品证明期 4 项验收未达）。

• M2 KeelBase Guard — Agent Registry / Policy Center / Risk Center / Guard Overview / security-governance navigation group / Agent↔audit linkage.
  M2 KeelBase Guard：Agent Registry / Policy Center / Risk Center / Guard Overview / 安全治理导航组 / Agent↔审计联动。
• Trust UX (D1/D2/D4) — AI CRM business closed loop (streaming Copilot + inline confirmation + governance drill-down), Action Detail page, human-language audit labels, Agent delegation-chain fields.
  Trust UX (D1/D2/D4)：AI CRM 业务闭环（Copilot 流式 + 内联确认 + 治理钻取）、Action Detail 页、人类语言审计标签、Agent 委托链字段。
• Product-proof preparation — onboarding-30min (invoices example), one-click Golden Flow demo, OIDC enterprise SSO, AR-2 MCP verification.
  产品证明准备：onboarding-30min（invoices 示例）、一键 Golden Flow demo、OIDC 企业 SSO、AR-2 MCP 验证。
• Cross-end & tooling — Taro i18n, bilingual README, React CI, role switching, Gitee issue sync, generator DX (idempotency / anchor fixes).
  跨端与工具：Taro i18n、README 双语、React CI、角色切换、Gitee issue 同步、生成器 DX（幂等/锚点修复）。
• Fixes — event postgres / generator overwriting flagship tools / wiring anchors / P0 deployment + security hardening (external review) / deployment domains no longer hard-coded.
  修复：事件 postgres、生成器覆盖旗舰工具、接线锚点、P0 部署+安全硬化（外部评审）、部署域名非硬编码。

Release gate deterministic 11/11 PASS · backend 1786 unit tests + e2e 258 + vitest 281 + CLI 69 all green · coverage 91.81/77.28/89.52/92.41.

Release gate 确定性 11/11 PASS · 后端 1786 单测 + e2e 258 + vitest 281 + CLI 69 全绿 · 覆盖率 91.81/77.28/89.52/92.41。