NianToB/task_plan.md

# YINIAN Desktop Next Phase Plan

## Goal

Move from M1 foundation closure to an M2 product slice that a business customer can actually pilot: authenticated desktop entry, stable workspace context, visible daily operations surface, skill lifecycle primitives, and a clear path to connect the real server without disturbing ClawX/OpenClaw basics.

## 2026-05-13 对话 Markdown 与隐藏提示泄漏修复

Status: complete

Goal:
- 防止业务回答准则作为可见内容污染真实对话。
- 增强会话内 Markdown 结构、表格展示，并把执行思考收成无框时间线。

Tasks:
- [x] 停止在 chat/cron 运行时消息里追加隐藏业务准则。
- [x] 历史里已有的隐藏准则在用户消息展示、标签和去重时自动剥离。
- [x] 为 Markdown 标题、列表、引用和表格增加工作台风格渲染。
- [x] 执行过程卡从边框卡片改为无框时间线。
- [x] 新对话空态删除品牌小标题和图标，只保留核心引导文案。
- [x] 补充测试并运行验证。

Verification:
- `pnpm vitest run tests/unit/chat-message.test.tsx tests/unit/chat-target-routing.test.ts tests/unit/cron-routes.test.ts tests/unit/business-guidance.test.ts`: passed, 4 files / 24 tests.
- `pnpm vitest run tests/unit/chat-page-execution-graph.test.tsx tests/unit/task-visualization.test.ts tests/unit/chat-store-history-retry.test.ts`: passed, 3 files / 18 tests.
- `pnpm vitest run tests/unit/chat-message.test.tsx tests/unit/chat-target-routing.test.ts tests/unit/cron-routes.test.ts tests/unit/chat-store-history-retry.test.ts`: passed, 4 files / 27 tests.
- `pnpm run typecheck`: passed.
- `pnpm run build:vite`: passed with existing dynamic-import/chunk-size warnings.
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`: passed, 1 test.
- `pnpm exec playwright test --config=playwright.legacy.config.ts tests/e2e/yinian-visual-smoke.spec.ts`: passed, 1 test.
- `pnpm exec playwright test --config=playwright.legacy.config.ts tests/e2e/chat-table-header-light.spec.ts`: passed, 1 test after fixing the test route to load `/chat`.
- `git diff --check`: passed.
- `pnpm run typecheck`: passed after the new-chat empty-state cleanup.

Notes:
- The first `chat-table-header-light.spec.ts` attempt failed because the legacy E2E started on the default new-chat route and never loaded seeded chat history. The test now sets `#/chat` before reload.

## 2026-05-13 桌面任务提醒与业务回答渲染

Status: complete

Goal:
- 将任务提醒从任务卡片状态升级为 Electron 桌面系统通知。
- 将业务型回答从纯 Markdown 增强为可扫读的结构化摘要 UI。

Tasks:
- [x] 新增 cron 完成事件到桌面通知的解析、去重和展示转换。
- [x] 主进程监听 Gateway cron 终态通知，弹出系统通知，点击回到任务中心并定位任务。
- [x] 任务中心支持 `?task=` 高亮目标任务。
- [x] 聊天消息识别“状态/依据/影响/下一步”等业务标签，渲染业务摘要面板。
- [x] 更新隐藏业务回答准则，引导 OpenClaw 输出稳定标签。
- [x] 补充单测并运行类型检查、构建和 Electron 冒烟。

Verification:
- `pnpm vitest run tests/unit/cron-desktop-reminder.test.ts tests/unit/chat-message.test.tsx tests/unit/tasks-page.test.tsx tests/unit/business-guidance.test.ts`: passed, 4 files / 20 tests.
- `pnpm run typecheck`: passed.
- `pnpm run build:vite`: passed with existing dynamic-import/chunk-size warnings.
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`: passed, 1 test.

Notes:
- First E2E attempt failed because `pnpm run build:vite` was running in parallel and temporarily removed `dist/index.html`; a clean rerun after build completion passed.

## 2026-05-13 聊天页字体与 UI 统一

Status: complete

Goal:
- 修正对话页字体、空态、消息气泡、执行过程和输入框与智念工作台整体 UI 不一致的问题。

Tasks:
- [x] 明确全局桌面中文系统字体栈，并让输入控件继承。
- [x] 将聊天外层从半透明玻璃感收敛为智念标准面板。
- [x] 将空态从卡片式居中块改成轻量无框状态。
- [x] 统一消息气泡、业务摘要、工具状态、文件/图片卡、执行过程卡的边框、圆角、底色与字号。
- [x] 统一聊天输入框、快捷任务 chip、知识库选择器和工具按钮样式。
- [x] 运行单测、类型检查、构建、主 E2E 和视觉截图验证。

Verification:
- `pnpm vitest run tests/unit/chat-message.test.tsx`: passed, 1 file / 8 tests.
- `pnpm run typecheck`: passed.
- `pnpm run build:vite`: passed with existing dynamic-import/chunk-size warnings.
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`: passed, 1 test.
- `pnpm exec playwright test --config=playwright.legacy.config.ts tests/e2e/yinian-visual-smoke.spec.ts`: passed, 1 test.
- Visual chat screenshot reviewed at `test-results/yinian-visual/02-chat.png`.

## 2026-05-13 商品中心 Web 应用接入

Status: complete

Goal:
- 在应用中心新增“商品中心”，内嵌 `https://ticket.nianxx.cn/`，用于酒店企业底价采购旅游资源。
- 当前版本不注入真实 SSO 凭证，但 URL 构造和宿主页面要留下后续服务端单点登录接入边界。

Tasks:
- [x] 调研目标站点和当前应用中心接入方式。
- [x] 新增商品中心应用配置、路由和内嵌 Web 宿主页。
- [x] 补充中英文文案和操作状态。
- [x] 增加应用中心/商品中心的单元或冒烟测试。
- [x] 运行验证并确认本地开发预览。

Notes:
- 目标站点当前页面标题为“一感旅游”，描述为景区商品采购、余额管理、订单退款、报表对账一体化工作台。
- `curl -I https://ticket.nianxx.cn/` 返回 `HTTP/2 200`，未看到 `X-Frame-Options` 或阻断性 `Content-Security-Policy`，可先按 iframe 内嵌处理。

Errors Encountered:
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts` 首次失败在找不到 `app-center-item-product-center`；原因是 Electron E2E 使用的 `dist` 仍是旧构建。下一步先运行 `pnpm run build:vite`，再重跑 E2E。
- 重建后 E2E 已进入 `product-center-page`，但中文 iframe title 定位失败；改为在 iframe 上增加稳定的 `data-testid="product-center-frame"`。

Verification:
- `pnpm vitest run tests/unit/app-center.test.tsx tests/unit/product-center-url.test.ts`: passed, 2 files / 3 tests.
- `pnpm run typecheck`: passed.
- `pnpm run build:vite`: passed with existing Vite dynamic-import/chunk-size warnings.
- `pnpm test`: passed, 106 files / 675 tests; existing MaxListeners warnings remain.
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`: passed, 1 test.

## 2026-05-13 旅游资源订购 Login-State Follow-Up

Status: complete

Goal:
- 将应用名称从“商品中心”改为“旅游资源订购”。
- 内置 Web 使用持久化会话容器，保持 `ticket.nianxx.cn` 登录态。

Tasks:
- [x] 更新 zh/en 展示名、标题和测试用例命名。
- [x] 确认目标站点登录态存储方式。
- [x] 将宿主容器切换为 Electron `<webview>` 并指定持久 partition。
- [x] 运行单测、类型检查、构建和 E2E 验证。

Errors Encountered:
- 首次 follow-up 单测失败在 `app-center.test.tsx`，原因是断言仍匹配“底价采购”，实际新文案已改为“底价订购”；已更新断言。

Verification:
- `pnpm vitest run tests/unit/app-center.test.tsx tests/unit/product-center-url.test.ts tests/unit/product-center-page.test.tsx`: passed, 3 files / 4 tests.
- `pnpm run typecheck`: passed.
- `pnpm run build:vite`: passed with existing Vite dynamic-import/chunk-size warnings.
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`: passed, 1 test.

## 2026-05-13 快速视频创作 Rename And English Coverage

Status: complete

Goal:
- 将“智念视频助手”展示名改为“快速视频创作”。
- 核实英文状态下应用中心两个内置应用都有英文覆盖，不暴露 i18n key。

Tasks:
- [x] 更新 zh/en app-center 文案，包括应用卡片、宿主页标题、导航和启动/失败状态。
- [x] 增加英文应用中心覆盖测试。
- [x] 运行针对性单测、类型检查、构建和 Electron 冒烟。

Verification:
- `pnpm vitest run tests/unit/app-center.test.tsx tests/unit/product-center-page.test.tsx tests/unit/product-center-url.test.ts`: passed, 3 files / 5 tests.
- `pnpm run typecheck`: passed.
- `pnpm run build:vite`: passed with existing Vite dynamic-import/chunk-size warnings.
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`: passed, 1 test.

## 2026-05-13 快速视频创作内置 Web 英文覆盖

Status: complete

Goal:
- 英文状态下，快速视频创作内置 Web 应用不再暴露主要中文导航、标题、按钮、表单标签和项目页文案。
- 桌面宿主需要把当前语言传给内置 Web，并让内置应用在跨页面导航后保持该语言状态。

Tasks:
- [x] 定位英文漏覆盖来源，区分桌面外壳 i18n 与 NianxxPlay 内置 Web 硬编码中文。
- [x] 桌面端 NianxxPlay 宿主 URL 注入 `zhinianLang`。
- [x] NianxxPlay 内置应用新增英文覆盖桥，并持久化嵌入语言状态。
- [x] 补充针对 URL 语言参数和英文覆盖字典的测试。
- [x] 重建内置 Web bundle，运行桌面验证。

Verification:
- `pnpm vitest run tests/unit/nianxx-play-url.test.ts tests/unit/app-center.test.tsx`: passed, 2 files / 4 tests.
- `/Users/inmanx/Documents/NianxxPlay`: `npm test`: passed, 4 files / 14 tests.
- `/Users/inmanx/Documents/NianxxPlay`: `npm run build`: passed with existing `ali-oss/urllib` dynamic dependency warning.
- `NIANXX_PLAY_SKIP_BUILD=1 pnpm run prepare:nianxx-play`: passed; refreshed `build/apps/nianxx-play`.
- `pnpm run typecheck`: passed.
- `pnpm run build:vite`: passed with existing Vite dynamic-import/chunk-size warnings.
- Playwright page probe on `http://127.0.0.1:3010/studio?zhinianEmbed=1&zhinianLang=en`: key English checks passed and first-screen continuous Chinese matches were empty.
- Playwright page probe on `/planning?zhinianEmbed=1&zhinianLang=en`: key English checks passed and continuous Chinese matches were empty.
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`: passed, 1 test.

## 2026-05-13 设置偏好深色 UI 修复

Status: complete

Goal:
- 修复设置/使用偏好在深色主题下的色彩层级、对比度和局部浅色残留。
- 顺带补上全局 app surface 和标题栏的深色背景兜底，减少其它深色页面出现同类问题。

Tasks:
- [x] 抓取当前深色偏好页截图并定位色彩问题。
- [x] 重构偏好页容器、分组和分段按钮的深色样式。
- [x] 调整全局深色背景与标题栏深色样式。
- [x] 运行类型检查/构建并重新截图验证。

Verification:
- `pnpm run typecheck`: passed.
- `pnpm exec vitest run tests/unit/title-bar.test.tsx tests/unit/launch-at-startup.test.ts tests/unit/app-center.test.tsx tests/unit/nianxx-play-url.test.ts`: passed, 4 files / 11 tests.
- `pnpm run build:vite`: passed with existing dynamic-import/chunk-size warnings.
- Electron visual screenshot: `test-results/settings-preferences-dark-electron-after.png`.
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`: passed, 1 test.

Notes:
- `pnpm exec playwright test --config=playwright.legacy.config.ts tests/e2e/settings-proxy.spec.ts` was attempted but failed before reaching Settings because the legacy setup helper still expects `main-layout` immediately after setup, while the current product flow is login-first.

## 2026-05-13 WhatsApp 二维码生成检查

Status: complete

Goal:
- 修复 WhatsApp 扫码连接时二维码生成不出来的问题，至少确保失败时不被前端吞掉，并让 WhatsApp WebSocket 走应用代理设置。

Tasks:
- [x] 检查前端扫码 Modal、Host API 路由、主进程 WhatsApp 登录管理器和 QR 渲染链路。
- [x] 用临时 HOME 拉起 WhatsApp 登录管理器，确认当前网络下直连 `web.whatsapp.com` WebSocket 超时/失败。
- [x] 给 WhatsApp 登录管理器接入应用代理设置，支持 HTTP/HTTPS 与 SOCKS 代理。
- [x] 修复 Host API unified envelope 非 2xx 时未抛错的问题，扫码启动失败时前端恢复状态并显示错误。
- [x] 补充 WhatsApp 网络/代理失败的本地化提示。
- [x] 补充代理解析和 Host API 非 2xx 单测。
- [x] 运行类型检查、相关单测和 Vite 构建。

Verification:
- `pnpm run typecheck`: passed.
- `pnpm exec vitest run tests/unit/host-api.test.ts tests/unit/whatsapp-proxy.test.ts tests/unit/channel-routes.test.ts`: passed, 3 files / 38 tests.
- `pnpm run build:vite`: passed with existing Vite dynamic-import/chunk-size warnings.
- 临时 HOME 直连实测仍无法连接 WhatsApp WebSocket（当前网络环境预期如此），但现在会返回明确错误：需要检查网络或开启高级设置里的代理。

## 2026-05-13 OpenClaw 内核升级

Status: complete

Goal:
- 将桌面端内置 OpenClaw 内核从 `2026.4.29` 升级到 npm stable latest，并确认当前运行时补丁、依赖打包和核心验证仍可用。

Tasks:
- [x] 查询当前依赖版本和 npm dist-tag。
- [x] 更新 `openclaw` 依赖与 lockfile。
- [x] 检查 OpenClaw 运行时补丁是否仍能匹配新版 bundle。
- [x] 运行类型检查、相关单测、构建/打包前置验证。
- [x] 记录升级结果和残余风险。

Verification:
- `node build/openclaw/openclaw.mjs --version`: `OpenClaw 2026.5.7 (eeef486)`.
- `pnpm run typecheck`: passed.
- `node scripts/assert-electron-runtime-deps.mjs`: passed.
- `pnpm exec vitest run tests/unit/openclaw-cli.test.ts tests/unit/openclaw-proxy.test.ts tests/unit/channel-config.test.ts tests/unit/channel-routes.test.ts tests/unit/whatsapp-proxy.test.ts tests/unit/host-api.test.ts`: passed, 6 files / 58 tests.
- `pnpm run build:vite`: passed with existing Vite dynamic-import/chunk-size warnings.
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`: passed, 1 test.

Notes:
- npm stable latest is `2026.5.7`; beta remains newer at `2026.5.12-beta.4`, but this upgrade intentionally stays on stable.
- `bundle-openclaw` now builds `build/openclaw` from `openclaw@2026.5.7` and critical YINIAN runtime boundaries are present in the bundled output.
- Several old bundle patch snippets no longer match because OpenClaw 2026.5.7 moved or upstreamed equivalent logic; verified equivalents include raw-model fast path, disabled main-session recovery, stuck-session active abort threshold, agents skill discovery disable, and Windows hidden process spawn handling.

## Current Baseline

- M1 auth/config/skill registry foundation implemented.
- Typecheck passed.
- Unit tests passed: 89 files, 572 tests.
- E2E passed: 26 passed, 1 skipped.
- Production flow: login first, then Today.
- E2E compatibility flow: legacy ClawX setup/main pages remain testable.

## Phase 1: Stabilize M1 For Handoff

Status: complete

Tasks:
- [x] Review all uncommitted files and separate intentional YINIAN changes from accidental drift.
- [x] Add or update a concise M1 implementation note in `docs/`.
- [x] Document environment switches:
  - `YINIAN_API_BASE_URL`
  - `CLAWX_LEGACY_AUTOSTART`
  - `CLAWX_E2E`
- [x] Confirm storage namespace and logout cleanup behavior are captured in docs.
- [x] Run final verification after any cleanup:
  - `pnpm run typecheck`
  - `pnpm run test`
  - `pnpm run test:e2e`

Exit Criteria:
- Working tree changes are explainable by module.
- M1 behavior can be handed to another engineer without reading the full chat.

Result:
- Completed in this session.
- Added `docs/M1_HANDOFF.md`.
- Updated `docs/START_HERE.md`.
- Fixed a channel health timestamp edge case where success/failure could share the same millisecond.
- Verification passed.

## Phase 2: Real Server Contract v0

Status: complete

Tasks:
- [x] Define the first stable backend API contract for:
  - SMS login
  - password login
  - restore/refresh session
  - hotel list/current workspace
  - config snapshot
  - skills manifest
- [x] Add mock fixtures that mirror the proposed server JSON exactly.
- [x] Add response normalization tests for missing/extra fields.
- [x] Decide refresh-token persistence boundary:
  - mock can use local store
  - HTTP should persist refresh token only when server contract is explicit
- [x] Add failure states for expired session and unavailable server.

Exit Criteria:
- Backend engineer can implement against a documented contract.
- UI/store can tolerate server errors without breaking the shell.

Result:
- Added `docs/SERVER_CONTRACT_V0.md`.
- Added `tests/fixtures/yinian-server-contract.ts`.
- HTTP mode now refreshes persisted sessions through `/auth/refresh`.
- HTTP mode normalizes contract v0 config and manifest responses.
- Verification passed.

## Phase 3: Today Page M2 Pilot Surface

Status: complete

Tasks:
- [x] Turn Today from foundation dashboard into B-end business cockpit.
- [x] Add real sections for:
  - pending guest requests
  - urgent exceptions
  - today arrival/departure summary
  - active skills and sync status
  - recent agent actions
- [x] Add empty/loading/error states that are operational, not marketing copy.
- [x] Merge entitlement and local registry states into clear labels:
  - enabled but not synced
  - installed
  - update available
  - disabled
- [x] Add tests for at least empty state, populated state, and workspace switch refresh.

Exit Criteria:
- A business operator can understand what needs attention within 10 seconds.

Result:
- Rebuilt `src/pages/Today/index.tsx` as an operations cockpit.
- Added `tests/unit/today-page.test.tsx`.
- Verification passed.

## Phase 4: Skills Manager M2

Status: complete

Tasks:
- [x] Expand local registry UI beyond list/sync:
  - per-skill status
  - version
  - last synced time
  - entitlement source
  - retry failed sync
- [x] Keep real bundle download out of scope unless explicitly promoted.
- [x] Define install/update/skip/fail result semantics in UI state mapping over shared registry/status types.
- [x] Add tests for:
  - empty registry
  - manifest changes
  - disabled entitlement
  - sync failure
  - workspace switch isolation

Exit Criteria:
- Skills page can explain exactly why a skill is available, unavailable, or stale.

Result:
- Rebuilt `src/pages/YinianSkills/index.tsx` into an operator-facing Skills Manager.
- Added `tests/unit/yinian-skills-page.test.tsx`.
- Verification passed.

## Phase 5: Design System Pass

Status: complete

Tasks:
- [x] Establish YINIAN visual language over the inherited ClawX shell:
  - calmer B-end business density
  - refined sidebar/topbar
  - consistent status chips, tables, and panels
  - no decorative landing-page treatment
- [x] Replace one-off page styling with reusable UI patterns where it reduces duplication.
- [x] Verify with typecheck, unit tests, Vite build, and Playwright E2E.

Exit Criteria:
- Login, Today, Skills, and tenant switcher feel like one product.

Result:
- Added lightweight YINIAN UI primitives in `src/components/yinian/ui.tsx`.
- Unified Login, Today, Skills, sidebar service context, main workspace background, and sidebar active states around a calmer navy/slate operations style.
- Verification passed.

## Phase 6: Pilot Packaging And QA

Status: complete

Tasks:
- [x] Prepare a pilot checklist:
  - install/start
  - mock login
  - server login when configured
  - switch workspace
  - sync skills
  - logout
  - gateway starts only after auth
- [x] Add a known-issues section.
- [x] Decide whether to commit a clean M1 checkpoint before M2.

Exit Criteria:
- The build can be handed to a stakeholder for a guided demo.

Result:
- Added `docs/PILOT_QA.md` with demo script, QA checklist, verification gate, packaging notes, and known issues.
- Updated `docs/START_HERE.md` and `docs/M1_HANDOFF.md` to point at the pilot QA baseline.
- No git commit was created because the working assumption remains "do not commit unless explicitly requested".

## Risks

- E2E compatibility now has explicit YINIAN bypass logic; keep it narrow and documented.
- Real server auth semantics may force storage changes, especially refresh token handling.
- ClawX legacy pages still exist; YINIAN navigation must avoid exposing implementation/admin surfaces to business customers too early.
- Skill registry is ready for OTA later, but real bundle download/signature verification remains intentionally out of scope.

## Next Recommended Action

Next, either create a clean checkpoint commit when requested, or start the next product phase: real server integration and first real business skill execution path.

## 2026-05-13 Task And Skill Management Clarification

Status: complete

Goal:
- Remove ambiguity between quick tasks, ability-pack tasks, and scheduled tasks.
- Restore ability-pack-owned quick-task management to the ability-pack/skills area.
- Preserve the earlier scheduled-task creation design, including push channel selection.
- Support adding skills into task editing content through `@` mentions.

Tasks:
- [x] Inspect current ability-pack, quick-task, scheduled-task, and task editor code paths.
- [x] Recover earlier scheduled-task and quick-task design from local git history if current code no longer contains it.
- [x] Implement UI/data-flow changes with minimal disruption to existing stores and APIs.
- [x] Add or update focused tests for quick-task management, scheduled-task push channels, and skill insertion.
- [x] Run typecheck and targeted tests.

Exit Criteria:
- Quick tasks that belong to ability packs are managed from the ability-pack UI.
- Scheduled-task creation/editing still exposes the important earlier fields, especially push channels.
- Task editing can insert a skill mention into the editor via `@`.

Result:
- Restored the ability-pack quick-task tab in `YinianSkills`, including edit/save support.
- Renamed the user-facing ability-pack concept from “快捷任务” to “快捷能力” to avoid confusion with scheduled/manual tasks.
- Changed Task Center's quick tab into a run-now surface and removed quick-task creation/editing from it.
- Brought scheduled-task delivery mode, channel, account, and target selection into the Task Center scheduled dialog.
- Added `@` skill insertion in manual/scheduled task textareas, inserting the existing `使用{name} skill` trigger text.
- Verification passed:
  - `pnpm run typecheck`
  - `pnpm vitest run tests/unit/yinian-skills-page.test.tsx tests/unit/tasks-page.test.tsx`
  - `pnpm test` (97 files / 638 tests; existing MaxListeners warnings remain)

## 2026-05-13 Task Center Separation Follow-up

Status: complete

Goal:
- Fully separate quick capabilities from Task Center.
- Keep Task Center focused on scheduled tasks and run history.
- Remove quick-capability template selection from scheduled-task creation.
- Fix `@` capability mention keyboard navigation.
- Make custom execution time friendlier than raw cron entry.

Tasks:
- [x] Remove quick-capability run/manage surfaces from Task Center.
- [x] Remove quick-capability template selection and binding creation from scheduled-task dialog.
- [x] Improve `@` mention keyboard selection behavior.
- [x] Replace raw custom cron entry with a structured custom schedule form while preserving cron payload output.
- [x] Update locale copy and focused tests.
- [x] Run typecheck and relevant tests.

Exit Criteria:
- Task Center no longer exposes quick capabilities as a task type.
- New scheduled tasks are authored directly with content plus optional `@` capability insertion.
- ArrowUp/ArrowDown/Enter can reliably choose a capability mention.
- Custom execution time is understandable through form controls and still saves a cron expression.

Result:
- `/tasks` now defaults to scheduled tasks and only exposes Scheduled Tasks plus Run History tabs.
- Scheduled-task creation/editing no longer offers quick-capability template selection or creates quick-capability bindings.
- Existing scheduled-task binding cleanup remains on edit/delete so stale local quick-capability associations are removed opportunistically.
- `@` mention suggestions now preserve active selection across keyup, support Enter insertion, and scroll the active option into view when the browser supports it.
- Custom execution time now uses structured controls for daily, weekly, monthly, and minute-interval schedules, with advanced Cron kept as a fallback.
- Verification passed:
  - `pnpm run typecheck`
  - `pnpm vitest run tests/unit/tasks-page.test.tsx tests/unit/task-center-store.test.ts`
  - `pnpm test` (97 files / 638 tests; existing MaxListeners warnings remain)
  - `pnpm run build:vite` (existing Vite chunk warnings remain)
  - `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`

Errors Encountered:
- The first targeted E2E run still saw `tasks-tab-quick` because Electron E2E was using stale `dist` output. Rebuilt with `pnpm run build:vite` and reran the same E2E successfully.
- jsdom does not implement `scrollIntoView`; guarded the mention-list scroll call so browser scrolling remains while tests and older DOM environments do not fail.

## 2026-05-13 Task Center Pinning Follow-up

Status: complete

Goal:
- Rename the scheduled-task surface to Task Center without changing any time-setting controls.
- Let each task card pin/unpin itself to the left sidebar.
- Let pinned sidebar tasks trigger immediately from the sidebar.

Tasks:
- [x] Update task-center copy that still says scheduled task where it names the product surface, leaving execution-time labels untouched.
- [x] Add persisted pinned task ids to task-center store.
- [x] Add pin/unpin controls to task cards.
- [x] Render pinned task quick triggers in the left sidebar and wire immediate execution.
- [x] Add/update focused tests.
- [x] Run typecheck and relevant test/build checks.

Exit Criteria:
- Time configuration controls and labels are unchanged.
- Task cards can be pinned to the sidebar.
- Sidebar pinned tasks can be clicked to run immediately.

Result:
- Renamed the scheduled-task product surface copy to “任务中心” / “Task Center” while preserving execution-time labels and controls.
- Added `yinian:pinned-task-ids` local persistence through `useTaskCenterStore`.
- Added task-card pin/unpin controls.
- Sidebar now renders pinned task quick triggers and calls the existing cron trigger API immediately on click.
- Sidebar-triggered runs are also recorded in task run history.
- Verification passed:
  - `pnpm run typecheck`
  - `pnpm vitest run tests/unit/tasks-page.test.tsx tests/unit/task-center-store.test.ts tests/unit/today-page.test.tsx`
  - `pnpm test` (97 files / 640 tests; existing MaxListeners warnings remain)
  - `pnpm run build:vite` (existing Vite chunk warnings remain)
  - `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`

## 2026-05-13 Task Center Density And Sidebar Structure

Status: complete

Goal:
- Convert Task Center from card-grid management to dense list management.
- Preserve important task actions: enable/disable, pin, run now, edit, open session, delete, delivery visibility, and execution metadata.
- Improve sidebar structure so pinned quick triggers are contained and the navigation remains scannable.

Tasks:
- [x] Apply data-dense dashboard guidance to Task Center layout.
- [x] Replace task-card grid with a responsive list/table layout.
- [x] Keep all execution-time controls untouched.
- [x] Restructure sidebar navigation and pinned triggers into a scrollable, organized area.
- [x] Update focused tests if accessible labels or copy change.
- [x] Run typecheck, focused tests, full tests, and E2E/build checks as needed.

Exit Criteria:
- Task Center uses a compact list layout and shows more tasks per screen.
- All existing task actions remain available.
- Sidebar groups are clearly separated and pinned tasks do not overwhelm the menu.

Result:
- Replaced the task card grid with a compact list/table layout that keeps task name/content, enabled state, schedule metadata, delivery info, pin, run, edit, open-session, and delete actions in each row.
- Moved scheduled-task metrics into compact chips in the panel header.
- Kept `ScheduledTaskDialog` execution-time controls unchanged.
- Put sidebar navigation into a scrollable structured area and bounded pinned quick triggers so chat/settings remain reachable.
- Verification passed:
  - `pnpm run typecheck`
  - `pnpm vitest run tests/unit/tasks-page.test.tsx tests/unit/task-center-store.test.ts tests/unit/today-page.test.tsx`
  - `pnpm test` (97 files / 640 tests; existing MaxListeners warnings remain)
  - `pnpm run build:vite` (existing Vite chunk warnings remain)
  - `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`

## 2026-05-13 Immediate History Session Visibility

Status: complete

Goal:
- Make newly started sessions appear in the sidebar history immediately instead of waiting for `sessions.list`.
- Cover both desktop new chats and Task Center / background automation sessions.

Tasks:
- [x] Add a shared local session visibility patch for optimistic history rows.
- [x] Mark new desktop sessions with `updatedAt` so they sort into the visible history area without turning into permanent activity.
- [x] Mark outgoing messages as active sessions immediately.
- [x] Mark cron trigger and Gateway started/completed sessions visible immediately.
- [x] Add focused regression tests.
- [x] Run typecheck, focused tests, full tests, build, and smoke E2E.

Result:
- New blank sessions get a local `updatedAt` immediately, so the history list can show them near the top while `cleanupEmptySession` can still remove unused blanks.
- Sending a first message now upserts the session row, label, and activity timestamp in one store update.
- Cron trigger calls insert a visible `agent:main:cron:<id>` session with the task name before the backend refresh returns.
- Gateway started/completed notifications insert background sessions optimistically, so scheduled automation runs appear as soon as the frontend hears about them.
- Verification passed:
  - `pnpm run typecheck`
  - `pnpm vitest run tests/unit/chat-session-actions.test.ts tests/unit/gateway-events.test.ts tests/unit/cron-store-fetch-dedupe.test.ts tests/unit/tasks-page.test.tsx tests/unit/task-center-store.test.ts`
  - `pnpm test` (98 files / 645 tests; existing MaxListeners warnings remain)
  - `pnpm run build:vite` (existing Vite chunk warnings remain)
  - `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`

## 2026-05-13 Task Execution Conversation Follow-up

Status: complete

Goal:
- Treat each Task Center task as one stable follow-up conversation.
- Jump to that conversation immediately after user-triggered execution.

Tasks:
- [x] Use a stable `agent:main:cron:<jobId>` session key for Task Center execution records and open-session actions.
- [x] Switch to the fixed task session and navigate to Chat when clicking Task Center "Run now".
- [x] Switch to the fixed task session and navigate to Chat when using sidebar pinned task triggers.
- [x] Add a regression test for Task Center "Run now" switching to the fixed task conversation.
- [x] Run typecheck, focused tests, full tests, build, and smoke E2E.

Result:
- Task Center run-now and sidebar pinned triggers now immediately open the task's fixed conversation.
- Each execution of the same task continues to land in the same `agent:main:cron:<jobId>` session, keeping follow-up context together.
- Verification passed:
  - `pnpm run typecheck`
  - `pnpm vitest run tests/unit/tasks-page.test.tsx tests/unit/task-center-store.test.ts tests/unit/cron-store-fetch-dedupe.test.ts tests/unit/chat-session-actions.test.ts`
  - `pnpm test` (98 files / 646 tests; existing MaxListeners warnings remain)
  - `pnpm run build:vite` (existing Vite chunk warnings remain)
  - `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`

## 2026-05-13 Task Conversation Naming

Status: complete

Goal:
- Show Task Center conversations directly with the task name.
- Prevent task conversations from being renamed to the first execution prompt.

Tasks:
- [x] Force task session label/displayName to the task name when running, opening, or sidebar-triggering a task.
- [x] Keep cron session labels protected during history hydration and message sends.
- [x] Let Gateway automation events backfill the task name from local cron jobs when available.
- [x] Add focused regression coverage.
- [x] Run typecheck, focused tests, full tests, build, and smoke E2E.

Result:
- Task sessions now display the task name directly instead of `agent:main:cron:<id>` or task prompt text.
- History refresh no longer overwrites cron task names with the first user message.
- Verification passed:
  - `pnpm run typecheck`
  - `pnpm vitest run tests/unit/tasks-page.test.tsx tests/unit/chat-history-actions.test.ts tests/unit/cron-store-fetch-dedupe.test.ts tests/unit/gateway-events.test.ts tests/unit/chat-session-actions.test.ts`
  - `pnpm test` (98 files / 647 tests; existing MaxListeners warnings remain)
  - `pnpm run build:vite` (existing Vite chunk warnings remain)
  - `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`

## 2026-05-13 Task History Merge Reliability

Status: complete

Goal:
- Fix task conversations sometimes showing no prior history or only the current prompt.
- Preserve the stable task conversation while merging available Gateway transcript and cron run-log fallback history.

Tasks:
- [x] Diagnose task conversation history loading.
- [x] Merge cron fallback history even when `chat.history` returns partial messages.
- [x] Deduplicate near-identical fallback/Gateway messages.
- [x] Make failed cron run fallback messages visible instead of filtered as system messages.
- [x] Add regression tests for partial Gateway history and failed run fallback visibility.
- [x] Run typecheck, focused tests, full tests, build, and smoke E2E.

Result:
- Cron task sessions now combine Gateway `chat.history` with `/api/cron/session-history` instead of using the fallback only when Gateway history is empty.
- If Gateway only returns the current execution prompt, previous cron run summaries still appear in the same task conversation.
- Failed task run fallback messages now render as assistant messages with `isError`, so failures are visible in the task conversation.
- Verification passed:
  - `pnpm run typecheck`
  - `pnpm vitest run tests/unit/chat-history-actions.test.ts tests/unit/cron-routes.test.ts tests/unit/tasks-page.test.tsx tests/unit/gateway-events.test.ts tests/unit/cron-store-fetch-dedupe.test.ts`
  - `pnpm test` (98 files / 649 tests; existing MaxListeners warnings remain)
  - `pnpm run build:vite` (existing Vite chunk warnings remain)
  - `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`

## 2026-05-12 Customer Clean Install Stabilization

Status: complete

Goal:
- Fix the customer clean-install path where the app restores an old login/session, skips initialization, starts Gateway, then cannot chat because the managed OpenClaw runtime is incomplete.

Tasks:
- [x] Inspect customer logs and screenshots.
- [x] Confirm packaged runtime is missing `docs/reference/templates/AGENTS.md`.
- [x] Confirm managed runtime cannot resolve OpenClaw self-imports such as `openclaw/plugin-sdk/provider-model-shared`.
- [x] Preserve required OpenClaw runtime templates in bundled and packaged resources.
- [x] Add a self-reference package/symlink for managed runtime module resolution.
- [x] Make setup status validate actual runtime files instead of trusting only `setupComplete`.
- [x] Prevent Gateway auto-start unless setup is verified complete and user is authenticated.
- [x] Restore NianxxPlay compiled `generated-results` route while still excluding user generated files.
- [x] Verify with bundle/build checks and targeted tests.

Current Findings:
- Customer log shows `Missing workspace template: AGENTS.md` during chat.
- Customer log shows `ERR_MODULE_NOT_FOUND: Cannot find package 'openclaw' imported from .../dist/extensions/codex/prompt-overlay.js`.
- Customer log shows `providerKeys=0` and diagnostics screenshot shows missing local model credentials; this remains a likely next blocker after runtime repair.
- NianxxPlay also reported missing `.next/server/app/generated-results/[...path]/route.js` after excluding generated results; the bundle filter now keeps compiled routes while continuing to exclude `public/generated-results`.

## Enterprise Space Service Integration

Status: in_progress

Goal:
- Connect the desktop client to the service-side enterprise space and application/skill distribution APIs.
- Keep the current single-account-single-service product assumption, while making backend endpoints configurable enough to match CloudClaw once repository access is available.

Tasks:
- [x] Check whether `https://git.nianxx.cn/brother7/CloudClaw` can be read from this workspace.
- [x] Record the current desktop API assumptions for enterprise space, config, and app manifest.

## 2026-05-13 Task Center Iteration

Status: complete

Goal:
- Promote tasks into a primary product module so ordinary business users can reuse capability-pack workflows manually or on a schedule.

Completed:
- Added `/tasks` and redirected `/cron` to `/tasks?tab=scheduled`.
- Reworked the customer sidebar into 快速使用 / 任务 / 对话 business groups.
- Added task center tabs: 快捷任务、定时任务、执行记录.
- Moved visible quick-task configuration out of the capability-pack settings tab.
- Added unified quick-task prompt composition shared by chat and task execution.
- Added local scheduled-task bindings and task run record persistence with separate storage keys:
  - `yinian:scheduled-task-bindings`
  - `yinian:task-run-records`
- Added `tasks` i18n namespace for Chinese and English.
- Updated delivery smoke to cover task center and `/cron` compatibility.
- Added unit coverage for prompt construction and task-center store behavior.

Verification:
- `pnpm run typecheck`
- `pnpm test`
- `pnpm run build:vite`
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`
- [x] Add configurable endpoint support for enterprise space/config/app manifest APIs.
- [x] Normalize common backend field names for app/skill manifests.
- [x] Add fallback behavior when the enterprise space service has not implemented config or manifest endpoints yet.
- [x] Add focused tests for enterprise space and manifest normalization.
- [x] Run `typecheck` and focused unit tests.

Known Blocker:
- Anonymous git access requires authentication, but source was readable through the user's logged-in Chrome session.
- Current CloudClaw backend does not yet expose enterprise app/skill distribution APIs; desktop falls back to empty registry until those routes are added.

## Product Simplification Pass

Status: complete

Goal:
- Remove customer-facing fields and metrics that are not supported by the real service today.
- Keep the product lighter and more credible for non-technical enterprise users.

Tasks:
- [x] Audit login, sidebar, Today, Settings, mock control-plane, and tests for unsupported business claims.
- [x] Remove account profile fields such as role/job/location when they are not returned by the server contract.
- [x] Remove or replace unsupported operational metrics such as health score/运营健康度.
- [ ] Keep only service-backed or local-real features:
  - 登录
  - 当前服务名称
  - 应用列表/同步状态
  - 知识库文件
  - 对话与历史
  - 定时任务入口
- [x] Run focused and full verification.

Result:
- Today now shows service/application readiness instead of unsupported operational health, fake pending business metrics, or recent run traces.
- Login no longer claims fixed time savings, fixed v1 app counts, or frontend key ownership.
- Sidebar and Settings no longer show service city/role/profile metadata.
- Shared YINIAN service type no longer models role/city/permissions/OTA as product-level fields.
- Mock fixtures and contract examples now use generic B-end service/application language.

## UI Customization Phase 1: Brand Foundation

Status: complete

Tasks:
- [x] Persist project-level design system for `智念助手`.
- [x] Create a new app icon concept and SVG sources.
- [x] Generate desktop icon outputs for macOS, Windows, and Linux.
- [x] Rename customer-facing product surfaces to `智念助手`.
- [x] Update packaging metadata, desktop shortcut names, tray labels, and window title.
- [x] Keep low-level ClawX/OpenClaw technical compatibility names where they still protect inherited behavior.

Result:
- Added `design-system/智念助手/MASTER.md`.
- Updated `resources/icons/`, `src/assets/logo.svg`, `electron-builder.yml`, `package.json`, Login, Sidebar, TitleBar, tray, and startup metadata.
- Verification passed.

Next UI Step:
- Continue with UI Customization Phase 2: extend YINIAN primitives into a fuller design kit and refactor Login/Today/Skills shell details against the new brand system.

## UI Customization Phase 2: Core Surface Refinement

Status: complete

Tasks:
- [x] Extend `src/components/yinian/ui.tsx` with reusable primitives for mature product surfaces.
- [x] Refine Login into a more branded enterprise desktop entry.
- [x] Refactor Today onto section panels, metric cards, info rows, and list items.
- [x] Refactor Skills Manager card internals onto shared info rows and panels.
- [x] Refine sidebar service context as a workspace context surface.
- [x] Verify typecheck, focused unit tests, and Vite build.

Result:
- Core YINIAN pages now share a stronger design vocabulary without changing backend/session/skill semantics.

Next UI Step:
- Continue with UI Customization Phase 3: visual QA and polish pass across sidebar density, Chat/Settings inherited ClawX pages, and customer/admin navigation boundaries.

## UI Customization Phase 3: Inherited Surface Brand Boundary

Status: complete

Tasks:
- [x] Group Sidebar navigation into B-end operations, AI work, and extension sections.
- [x] Keep existing route test ids and navigation behavior stable.
- [x] Replace major Chinese customer-visible Chat/Settings/Setup ClawX strings with 智念助手.
- [x] Swap Setup welcome logo alt/asset to 智念助手.
- [x] Verify typecheck, focused unit tests, Vite build, and navigation E2E.

Result:
- Customer-facing shell now has a clearer information architecture and less inherited ClawX branding.

Next UI Step:
- Continue with UI Customization Phase 4: full visual QA screenshots and deeper inherited page polish for Chat, Settings, Channels, Agents, Models, and Cron.

## UI Customization Phase 4: Inherited Page Visual Polish

Status: complete

Tasks:
- [x] Remove oversized serif title treatment from inherited Chat, Settings, Models, Agents, Channels, and Cron surfaces.
- [x] Normalize inherited page headings toward the 智念助手 system-font dashboard style.
- [x] Remove remaining major Chinese customer-visible ClawX copy from Agents, Cron, Channels, Chat, Settings, and Setup.
- [x] Keep technical GitHub/upstream references available where still useful.
- [x] Verify typecheck, focused unit tests, Vite build, and inherited navigation/provider E2E.

Result:
- Inherited pages now feel less like the original ClawX shell and closer to the 智念助手 product system while preserving behavior.

Next UI Step:
- Continue with UI Customization Phase 5: screenshot-based visual QA across authenticated YINIAN flow and legacy E2E flow, then fix concrete layout issues found at desktop widths.

## UI Customization Phase 5: Screenshot Visual QA

Status: complete

Tasks:
- [x] Add a production YINIAN visual smoke test that bypasses legacy setup only at the renderer level.
- [x] Capture desktop screenshots for Login, Today, and Skills at 1440x900.
- [x] Fix the fresh-user production routing issue where legacy setup could override the 智念登录入口.
- [x] Verify typecheck, Vite build, and the new visual smoke test.

Result:
- Added `tests/e2e/yinian-visual-smoke.spec.ts`.
- Baseline screenshots are written to `test-results/yinian-visual/01-login.png`, `02-today.png`, and `03-skills.png`.
- Fresh production sessions now land on `/login` before authentication and `/today` after login.

Next UI Step:
- Continue with per-surface detail polish for Settings, Channels, Agents, Models, and Cron, using screenshot comparison as the acceptance gate.

## Product Positioning Pass: B-end Workspace Model

Status: complete

Tasks:
- [x] Change the primary navigation group to `快速使用`.
- [x] Replace customer-visible hotel-specific copy with workspace/business-object language.
- [x] Update mock tenant names and demo skill labels for broader B-end demos.
- [x] Add forward-compatible workspace aliases without breaking existing storage/API field names.
- [x] Refresh visual screenshots and verification after the repositioning pass.

Result:
- 智念助手 now reads as a configurable B-end AI Agent desktop product, where different customers can install or receive tailored Skills for their own business object.
- Existing `hotelId`/`hotel_id` internals remain intentionally stable until a dedicated server-contract migration is planned.

Next Product Step:
- Define the generic tenant model in the PRD/server contract (`workspace` / `businessObject`) and plan a controlled API rename only after server compatibility is clear.

## Shell Simplification Pass: Account-Bound Service

Status: complete

Tasks:
- [x] Remove the old top workspace bar from the authenticated shell.
- [x] Move current service identity into the left sidebar.
- [x] Remove workspace selection from the customer-facing shell.
- [x] Move sync and logout controls into the sidebar footer.
- [x] Refresh screenshot QA after the shell change.

Result:
- The product now assumes one account corresponds to one B-end service in the UI, while lower-level compatibility fields remain untouched.

Next UI Step:
- Continue simplifying inherited admin surfaces so the left sidebar remains the single persistent navigation and account-control surface.

## Sidebar IA Pass: Service-Managed Configuration

Status: complete

Tasks:
- [x] Hide developer/configuration surfaces from the production sidebar.
- [x] Keep Models, Agents, Channels, and Cron available only through E2E compatibility navigation.
- [x] Reorganize the customer-facing sidebar around two primary areas: `快速使用` and `对话`.
- [x] Move account actions and service-managed capability notices into Settings.
- [x] Add Settings to the visual smoke baseline.

Result:
- The production shell now presents Today, Skills, New Chat, Chat, and Settings as the only persistent customer navigation.
- Model/provider, Agent orchestration, channel, and scheduled-task configuration are documented in Settings as service-side managed capabilities.
- Logout and service config sync live in Settings instead of competing with the primary sidebar navigation.

Next UI Step:
- Continue refining the Chat landing layout so the `对话` area feels as deliberate as `快速使用`.

## Sidebar UX Pass: Dashboard Card, Quick Use, Chat History

Status: complete

Tasks:
- [x] Move Dashboard entry from the quick-use menu to the top service card.
- [x] Rebuild `快速使用` as three direct actions: Skills, 定时任务, 知识库.
- [x] Add a Knowledge page and route for user-uploaded files.
- [x] Rebuild the `对话` area as New Chat plus hoverable History.
- [x] Make History click open the latest session and hover reveal the history list.
- [x] Refresh visual smoke screenshots and compatibility navigation E2E.

Result:
- The sidebar now reads as a compact launch surface: service card for Dashboard, quick actions for tools, and a quiet chat module with history on demand.
- Knowledge Base v0 is implemented as a local upload/list/search surface, ready for later server indexing.

Next UI Step:
- Upgrade Knowledge Base from local UI state to persisted service-scoped registry when the server API is ready.

## CloudClaw Direct Plugin Install Pass

Status: complete

Tasks:
- [x] Read CloudClaw repository through the user's logged-in browser session.
- [x] Confirm the available integration point is `plugins/openclaw-cloud-sync`.
- [x] Build a 智念内置 `cloud-sync` OpenClaw plugin mirror.
- [x] Auto-install the mirror into `~/.openclaw/extensions/cloud-sync` before Gateway launch.
- [x] Auto-write `plugins.entries.cloud-sync.config.serverUrl` into `openclaw.json`.
- [x] Include the static plugin mirror in dev build and packaged afterPack flows.
- [x] Patch plugin metadata and command registration shape for the current OpenClaw runtime.
- [x] Verify Gateway startup and `openclaw plugins list` show `cloud-sync` as loaded.

Result:
- CloudClaw's existing OpenClaw cloud-sync plugin is now directly installed and enabled by 智念助手.
- This does not yet mean enterprise application/skill distribution is available from CloudClaw; that server API still does not exist in the inspected repo.

Next Integration Step:
- Decide whether 智念助手 should expose Cloud Sync as a customer-facing setting/action, or keep it hidden as infrastructure until the CloudClaw server URL and account contract are finalized.

## Version 3.5: Application Center Large-App Integration

Status: planned

Goal:
- Move 应用中心 from a lightweight launcher into the first real large-application host inside 智念助手.
- Keep 能力包 focused on agent/skill capability expansion, while 应用中心 hosts fixed product applications that may be native React modules or embedded web applications.
- Use one large application as the 3.5 proving ground for routing, launch/close behavior, state persistence, permissions, service context, and desktop-grade UX.

Current Baseline:
- 应用中心 exists at `/app-center` with a compact app-launcher layout.
- Built-in app registry exists in `src/stores/app-center.ts`.
- App item typing exists in `src/types/app-center.ts` and shared YINIAN app-center types.
- Small built-in demo apps have been removed from the customer-facing app center.
- 应用中心 now focuses on the single NianxxPlay large-app entry.
- 能力包 has been renamed from application/skill language and now owns 快捷任务配置.
- 快捷任务 can inject `使用{能力包名称} skill` into the chat send path, is single-select, and is single-use per send.
- Dev port has moved from `5173` to `5188` to avoid local project conflicts.

Target Large App Candidate:
- Local project path: `/Users/inmanx/Documents/NianxxPlay`.
- Product name: `NianxxPlay智念视频助手`.
- Current positioning: Seedance 2.0 AI video generation workspace for ordinary users and local merchants.
- Current modes:
  - 宣传片制作
  - 创意复刻
- Technical shape:
  - Next.js 15 app router application.
  - Own routes for home, studio, projects, project details, billing, and API routes.
  - Own local MVP data layer at `.data/app-state.json`.
  - Own asset upload, OSS, Seedance task creation/query, prompt assembly, project history, and credit logic.
- Verified status:
  - `npm test`: passed, 3 files / 11 tests.
  - `npm run build`: passed, with existing `ali-oss` dependency-chain dynamic require warning.
- Integration implication:
  - NianxxPlay is too large to treat like a small native utility route.
  - It is a good 3.5 candidate for an embedded web large-app host, with a local service process during internal testing and a packaged/served app strategy later.

Scope For 3.5:
- Define the large-app host contract:
  - app identity, source, type, route/url, icon, description, permissions, and service-context needs
  - native app vs embedded web app launch behavior
  - close/back behavior and app-level route ownership
  - app state restore boundary
- Upgrade 应用中心 UI from static launcher-only semantics to an app hub:
  - launch large apps
  - show fixed built-in apps clearly
  - keep tag filtering compact
  - avoid introducing “enable/disable” semantics for applications
- Integrate the first large application:
  - choose app name and business workflow
  - decide native React implementation vs embedded web
  - define data/API dependencies and local fallback behavior
  - add route shell, loading/error/empty states, and close behavior
- Add verification:
  - app-center unit tests for registry/filter/launch decisions
  - large-app smoke tests
  - `pnpm run typecheck`
  - `pnpm run build:vite`
  - targeted visual check at desktop size

Out Of Scope Until Explicitly Promoted:
- App marketplace/downloading from server.
- Multi-window app management.
- Server-driven per-user app permissions unless the backend contract is ready.
- Real embedded web sandbox hardening beyond the first controlled integration.

Risks:
- Current 应用中心 uses a static local registry; large-app delivery may need a stronger manifest and app lifecycle model.
- Current webview-type apps open externally; true embedded web applications need a safer host strategy.
- Large apps may need their own navigation and data domain, so they should not be crammed into the small-card launcher model.
- Customer-facing naming must keep 应用中心 and 能力包 distinct: 应用中心 opens product apps; 能力包 expands agent capabilities.

Next Recommended Action:
- Use NianxxPlay as the concrete 3.5 target.
- First implement an embedded-web app host inside 智念助手:
  - app launcher entry
  - app detail/host route
  - loading, error, reload, close/back controls
  - local development URL support
  - safe external-link behavior
- Then connect the NianxxPlay local dev app as the first hosted large application before deciding the production packaging strategy.

## Version 3.5: Bundled NianxxPlay Runtime

Status: planned

Goal:
- Ship NianxxPlay as a desktop-managed companion application.
- 智念助手 should install, update, launch, stop, and health-check NianxxPlay together with the desktop app.
- Customers should not need Node.js, npm, pnpm, Next.js, or terminal knowledge.

Preferred Architecture:
- Build NianxxPlay as a Next.js standalone server bundle.
- Copy the standalone bundle into `build/apps/nianxx-play/`.
- Add the bundle to `electron-builder.yml` as an `extraResources` entry, similar to bundled OpenClaw.
- At first-run or app update, copy the bundled app into an app-managed user runtime directory:
  - macOS/Linux: under app userData or `~/.zhinian/apps/nianxx-play`
  - avoid writing mutable state into `process.resourcesPath`
- Launch the app from Electron main process as a managed child runtime.
- Host it in 智念助手 through an embedded web app route, using a local loopback URL.

Required NianxxPlay Changes:
- Add `output: 'standalone'` to `next.config.ts`.
- Move mutable local data away from `process.cwd()`:
  - `.data/app-state.json`
  - local upload fallback storage
- Support env-driven runtime paths:
  - `NIANXXPLAY_DATA_DIR`
  - `NIANXXPLAY_UPLOAD_DIR`
  - `NIANXXPLAY_PUBLIC_BASE_URL`
- Do not bundle `.env.local`.
- Move real API keys to desktop-managed secure configuration or future server-issued config.
- Make the app work behind a dynamic local port instead of assuming `localhost:3000`.

Required Desktop Changes:
- Add an `electron/apps/nianxx-play` runtime manager:
  - resolve bundled and managed app paths
  - install/update managed app from bundled resources
  - find a free local port
  - launch and stop the standalone server
  - probe health/readiness
  - expose app status over IPC
- Add an application host route:
  - `/app-center/nianxx-play`
  - loading, error, reload, close/back controls
  - embedded webview/iframe target URL from app runtime manager
- Add setup/update integration:
  - initialization step for bundled applications
  - runtime diagnostics in Settings
  - shutdown cleanup on app quit
- Add packaging scripts:
  - `bundle-nianxx-play.mjs`
  - `package`/`build` pipeline integration
  - release artifact size check

Main Risks:
- `public/seedance-starter-assets` is currently large, roughly hundreds of MB, and will significantly increase installer size.
- The current NianxxPlay project root is about 1GB mostly due to `node_modules`, `.next/cache`, and media assets; production bundling must exclude dev cache and raw node_modules.
- Next.js standalone output still needs static assets and selected runtime dependencies.
- API keys in `.env.local` are sensitive and must never be shipped as a plain file.
- If NianxxPlay runs as an HTTP server, port management and crash recovery must be owned by the desktop main process.
- Local-only uploaded files cannot be used by Seedance unless they are reachable via public OSS; the local fallback remains preview-only.

Acceptance Criteria:
- Fresh customer install can open NianxxPlay from 应用中心 without installing Node/npm.
- App restart restores NianxxPlay project history from the app-managed data directory.
- Desktop update can replace NianxxPlay code without deleting user projects/assets.
- App quit stops the managed NianxxPlay process.
- `pnpm run typecheck`, `pnpm run build:vite`, NianxxPlay `npm test`, and NianxxPlay `npm run build` pass.

Service-Managed Boundary:
- NianxxPlay account, permission, secret, quota, and billing decisions belong to the server.
- The desktop app must not expose real service credentials in UI or ship them in plain files.
- Desktop responsibilities are limited to:
  - local installation/update of the application bundle
  - process lifecycle
  - health status
  - embedded display inside 应用中心
  - passing only server-issued, scoped runtime context when needed.

## Version 3.5: NianxxPlay Bundling Implementation Pass

Status: implemented in first pass

Goal:
- Move NianxxPlay from a development-only embedded URL to a packageable desktop companion app.
- Keep the existing 应用中心 UI and product flow intact.
- Make packaged builds launch NianxxPlay without requiring customers to install Node.js, npm, pnpm, or Next.js.

Implementation Steps:
- Configure NianxxPlay for Next.js standalone output.
- Add a desktop packaging script that:
  - builds NianxxPlay
  - copies only standalone runtime files, static assets, public assets, and content manifests
  - excludes `.env*`, `.data`, dev caches, and raw development dependencies
  - writes a bundle manifest for diagnostics
- Include the prepared bundle through `electron-builder.yml` extra resources.
- Update the Electron-side NianxxPlay service manager to:
  - resolve source-project runtime in development
  - resolve bundled standalone runtime in packaged builds
  - start standalone bundles through Electron's own Node runtime via `ELECTRON_RUN_AS_NODE`
  - inject a user-writable data directory and upload/result directories
  - stop the managed process on app quit
- Update NianxxPlay local persistence to respect desktop-managed runtime paths.

Verification:
- NianxxPlay `npm run build`
- desktop `pnpm run prepare:nianxx-play`
- desktop `pnpm run typecheck`
- desktop `pnpm run build:vite`
- confirm no `.env*` files are copied into bundled resources

## Mac Pilot: Optional Native Clipboard Gatekeeper Fix

Status: complete

Goal:
- Stop the repeated macOS security prompt for `clipboard.darwin-arm64.node` on customer machines.

Implementation:
- Remove optional `@mariozechner/clipboard*` packages at OpenClaw bundle generation time.
- Bump the Yinian OpenClaw runtime patch marker so existing managed runtimes reinstall from the cleaned bundle.
- Clean known user OpenClaw dependency caches before Gateway launch, covering prior pilot installs.
- Add a focused unit test for runtime cleanup behavior.

Verification:
- `pnpm run typecheck`
- `pnpm vitest run tests/unit/optional-native-cleanup.test.ts`
- `pnpm run build:vite`
- `pnpm exec zx scripts/bundle-openclaw.mjs`
- `find build/openclaw -name 'clipboard.darwin-arm64.node' -o -path '*@mariozechner/clipboard*'` returns no output.

## 2026-05-06 Complete Project Self Check

Status: complete

Goal:
- Re-run the delivery health checks after the latest mac pilot fixes.
- Stabilize stale test expectations that no longer matched the current product language and homepage/ability-pack UI.
- Verify the pilot packaging resource chain without producing a new customer installer.

Verification:
- `pnpm vitest run tests/unit/language-detection.test.ts tests/unit/chat-session-actions.test.ts tests/unit/task-visualization.test.ts tests/unit/today-page.test.tsx tests/unit/yinian-skills-page.test.tsx`: passed, 29 tests.
- `pnpm test`: passed, 93 files / 620 tests.
- `pnpm run typecheck`: passed.
- `pnpm run package:pilot`: passed.
- Resource scans:
  - no `tavily` in `resources/skills/preinstalled-manifest.json` or `build/preinstalled-skills`.
  - no `clipboard.darwin-arm64.node` or `@mariozechner/clipboard*` under `build/openclaw` or `release`.
  - NianxxPlay bundle includes `server.js`, `node_modules/next/package.json`, and `.env.runtime`.
  - NianxxPlay bundle scan found no bundled upload directory, generated-results directory, or `.next/cache`.

Notes:
- Removed `tavily-search` from the preinstalled skills manifest and regenerated `build/preinstalled-skills`.
- `package:pilot` still emits known Vite chunk/dynamic-import warnings and OpenClaw patch-skip warnings.
- NianxxPlay bundle remains large, currently about 419 MB in `build/apps/nianxx-play`.

## 2026-05-09 Office Skills Runtime Hardening

Status: complete

Goal:
- Consolidate the new MiniMax office/document skills so customer installs do not fall back to "please install LibreOffice/global npm/pip" guidance.
- Ensure startup/initialization can prepare the common runtime layer for `docx`, `pdf`, `pptx`, and `xlsx`.
- Keep the public skill names stable as `docx`, `pdf`, `pptx`, and `xlsx`.

Implementation Steps:
- Confirm the MiniMax replacement bundle is generated with stable frontmatter names and Zhinian runtime policies.
- Add a reusable office-skill dependency check/repair path for:
  - bundled/project Node modules used by `pptx`
  - managed Python 3.12 plus common packages used by `pdf`/`xlsx`
  - .NET availability for `docx`
- Trigger the repair path during Gateway warmup without blocking normal startup more than necessary.
- Expose the result through diagnostics/admin tooling so support can see what is missing.
- Validate with typecheck and targeted runtime scans.

Verification:
- `pnpm run bundle:preinstalled-skills`: passed.
- Managed Python 3.12 import check: `pypdf`, `reportlab`, `pandas`, `openpyxl`, `matplotlib`, and `bs4` all import successfully.
- `openclaw config validate`: passed after removing invalid provider `timeoutSeconds` and `models.pricing` writes.
- `openclaw skills list --json`: `docx`, `pdf`, `pptx`, `xlsx`, and `design` are openclaw-managed and enabled.
- `pnpm run typecheck`: passed.
- `pnpm vitest run tests/unit/model-diagnostics.test.ts tests/unit/diagnostics-routes.test.ts`: passed, 4 tests.
- `pnpm test`: passed, 93 files / 622 tests.
- `pnpm run build:vite`: passed with existing Vite chunk/dynamic-import warnings.

Notes:
- `.NET` is still not installed on this machine. The runtime diagnostic treats this as a visible warning for advanced `docx` OpenXML flows, not as a blocker for the basic office runtime.
- OpenClaw CLI still warns that the config was last written by newer OpenClaw `2026.4.29` while the shell `openclaw` command is `2026.4.15`; the desktop-managed runtime remains the intended source for packaged users.

## 2026-05-12 Customer Clean Install Stabilization

Status: complete

Goal:
- Fix the customer clean-install failure where the app skipped initialization, entered the workspace, and then could not chat on machines without a prior OpenClaw install.

Completed:
- Preserve OpenClaw `docs/reference/templates` in bundled and packaged resources.
- Create a managed runtime self-reference at `~/.openclaw/runtime/openclaw/node_modules/openclaw`.
- Treat `setupComplete` as a cached hint only; main-process setup verification now resets stale state when runtime/config/model auth/office markers are missing.
- Gate Gateway startup behind verified setup plus authenticated Yinian session.
- Keep NianxxPlay compiled generated-results route code while excluding user-generated output directories.
- Bundle six preinstalled skills for the pilot package: `docx`, `pdf`, `pptx`, `xlsx`, `design`, `image-search`.
- Include internal pilot model auth and NianxxPlay runtime env for closed testing only.

Verification:
- `pnpm run typecheck`: passed.
- `pnpm test`: passed, 94 files / 628 tests.
- `pnpm run package:pilot`: passed.
- `pnpm run package:mac:pilot:arm64`: passed.
- Packaged resource scan passed for OpenClaw templates, NianxxPlay route code, preinstalled skills, and absence of the optional clipboard native module.
- `codesign --verify --deep --strict`: passed.
- `spctl --assess`: rejected as `Unnotarized Developer ID`; notarization remains the next distribution blocker.

Deliverable:
- `/Users/inmanx/Desktop/智念助手-0.1.0-mac-arm64-20260512.dmg`

## 2026-05-12 Design And HTML Slides Skill Split

Status: complete

Goal:
- Keep the design capability in the Zhinian app first, then promote globally after app testing.
- Split design direction from HTML deck execution.

Completed:
- Upgraded `design` into a general design director/design-system skill for product design, UI/HTML direction, PPT narrative, brand direction, redesigns, and design reviews.
- Added `html-slides` as a new preinstalled execution skill focused on browser-based HTML decks and PPTX-to-HTML conversion.
- Bundled `beautiful-html-templates` HTML/JSON/runtime resources without screenshots.
- Bundled lightweight `frontend-slides` references for viewport fitting, animation patterns, and PPTX extraction.
- Bumped preinstalled versions:
  - `design`: `2026-05-12-design-system-v2`
  - `html-slides`: `2026-05-12-html-slides-v1`
- Synced the local OpenClaw test install under `~/.openclaw/skills`.

Verification:
- `python3 resources/skills/local/design/scripts/search.py "B2B SaaS presentation dashboard" --design-system`: passed.
- `pnpm run bundle:preinstalled-skills`: passed.
- Preinstalled resource scan found no screenshots and no `.app` directories under `html-slides`.
- `python3 -m py_compile` passed for both bundled and local `extract-pptx.py`.
- `node node_modules/openclaw/openclaw.mjs skills list --json`: `design` and `html-slides` are eligible and enabled.
- Temporary 8-slide HTML smoke deck passed 1280x720, 1440x900, and 390x844 overflow checks.

## 2026-05-13 Playwright Chromium Runtime Hardening

Status: in_progress

Goal:
- Prevent customer tasks from stalling on task-time `npx playwright install chromium` downloads.
- Reduce collisions between app-driven browser setup and user/manual Playwright installs.
- Make Chromium/browser preview readiness visible in diagnostics without blocking PPT/HTML deck delivery.

Implementation Steps:
- Set an app-scoped Playwright browser cache path in Gateway child process environment.
- Add Office/Slides runtime diagnostics for Playwright module and Chromium binary availability.
- Update `html-slides` and preinstalled skill bundling policies to forbid task-time Chromium downloads.
- Run targeted typecheck/tests and update the generated preinstalled skill bundle if needed.

Verification:
- `pnpm run typecheck`: passed.
- `pnpm run bundle:preinstalled-skills`: passed after bumping `pdf` and `html-slides` skill versions.
- `pnpm vitest run tests/unit/playwright-runtime.test.ts`: passed, 2 tests.
- Real Playwright probe with app-scoped browser cache: Chromium executable exists after symlink mirroring.
- Generated preinstalled `pdf` policy now forbids task-time browser installers, and `html-slides` policy explicitly forbids `npx playwright install chromium`.

## 2026-05-13 Cron Run Final-State Reconciliation

Status: complete

Goal:
- Prevent Task Center from showing a recovered cron run as failed when the artifact was actually produced.

Implementation Steps:
- Inspect local cron run logs, desktop output, and matching session trajectory files.
- Reconcile failed cron summaries with `trace.artifacts.data.finalStatus`.
- Update cron session fallback messages to prefer recovered final assistant summaries.
- Cover the behavior with cron route tests and run project verification.

Verification:
- `pnpm run typecheck`: passed.
- `pnpm vitest run tests/unit/chat-history-actions.test.ts tests/unit/cron-routes.test.ts tests/unit/tasks-page.test.tsx tests/unit/cron-store-fetch-dedupe.test.ts`: passed, 32 tests.
- `pnpm test`: passed, 98 files / 652 tests; existing MaxListeners warnings remain.

## 2026-05-13 Cron Conversation Live Refresh

Status: complete

Goal:
- Ensure a Task Center run shows this run's query and response/running hint in its fixed task conversation without requiring manual refresh.

Implementation Steps:
- Make cron fallback history produce visible user query and assistant running/empty messages.
- Optimistically append current run query and running placeholder when a task is manually triggered.
- After manual trigger completion, refresh the active fixed task conversation and clear running state.
- Fold run-specific cron event session keys into the fixed task conversation key.
- Force current cron conversation history refresh on terminal cron events.

Verification:
- `pnpm run typecheck`: passed.
- `pnpm vitest run tests/unit/gateway-events.test.ts tests/unit/cron-store-fetch-dedupe.test.ts tests/unit/cron-routes.test.ts tests/unit/tasks-page.test.tsx`: passed, 25 tests.
- `pnpm vitest run tests/unit/chat-history-actions.test.ts tests/unit/chat-session-actions.test.ts tests/unit/chat-runtime-event-handlers.test.ts tests/unit/gateway-events.test.ts tests/unit/cron-store-fetch-dedupe.test.ts tests/unit/cron-routes.test.ts tests/unit/tasks-page.test.tsx`: passed, 61 tests.
- `pnpm test`: passed, 98 files / 655 tests; existing MaxListeners warnings remain.

## 2026-05-13 Sidebar Navigation Restructure

Status: complete

Goal:
- Make the left sidebar more structured and space-efficient:
  - conversation entry at the top, directly below Home/Today.
  - Task Center followed by up to five pinned quick tasks.
  - App Center and Knowledge moved to the bottom, above Settings.

Implementation Steps:
- Move App Center and Knowledge from the primary quick-use area into a footer utility group.
- Add a dedicated conversation container for New Chat and History below the service/home card.
- Group Task Center and pinned quick task triggers together in the main navigation area.
- Limit sidebar pinned task rendering to five items.
- Add a unit test that locks the expected sidebar order and the five-item pinned task cap.

Verification:
- `pnpm run typecheck`: passed.
- `pnpm vitest run tests/unit/sidebar-layout.test.tsx`: passed, 1 test.
- `pnpm vitest run tests/unit/sidebar-layout.test.tsx tests/unit/tasks-page.test.tsx tests/unit/task-center-store.test.ts`: passed, 7 tests.
- `pnpm run build:vite`: passed with existing Vite chunk/dynamic-import warnings.
- `pnpm test`: passed, 99 files / 656 tests; existing MaxListeners warnings remain.
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`: passed, 1 test.

## 2026-05-13 App-Scoped Local Preference Recovery

Status: complete

Goal:
- Restore quick capabilities and channel account remarks after they appeared to disappear when the dev renderer origin changed.
- Prevent future dev-port or renderer-origin changes from hiding local business preferences.

Implementation Steps:
- Restart the dev app on the original data-bearing origin `http://localhost:5173`.
- Add an app-level Electron local-preferences store and Host API route.
- Sync quick capabilities from origin-local storage into app-level preferences, and hydrate from app-level preferences when the current origin has no data.
- Sync desktop user name, workspace display name, and channel account remarks through the same app-level preference store.
- Add unit tests for both hydration and migration directions.

Verification:
- Confirmed the app-level preference file contains the user's quick capabilities `写word文章` and `做网页`.
- Confirmed account remarks are present, including `徐明微信`, `小石的微信`, `徐明飞书`, and `宗琦A2A`.
- `pnpm run typecheck`: passed.
- `pnpm vitest run tests/unit/quick-tasks-store.test.ts tests/unit/yinian-local-prefs.test.ts`: passed, 4 tests.
- `pnpm vitest run tests/unit/yinian-skills-page.test.tsx tests/unit/tasks-page.test.tsx tests/unit/sidebar-layout.test.tsx tests/unit/quick-tasks-store.test.ts tests/unit/yinian-local-prefs.test.ts`: passed, 16 tests.
- `pnpm run build:vite`: passed with existing Vite chunk/dynamic-import warnings.
- `pnpm test`: passed, 101 files / 660 tests; existing MaxListeners warnings remain.
- `pnpm exec playwright test tests/e2e/yinian-delivery-smoke.spec.ts`: passed, 1 test.