admin/prop-ai-hr
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
601 Commits 1 Branch 0 Tags
11250811a6796dee8bcca154498293b5c416bb22
Commit Graph
601 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
admin 11250811a6 fix: show work capture context before actions 2026-07-21 05:01:43 +08:00
admin 4536d607e3 feat: add project-scoped work assistant results 2026-07-21 04:44:15 +08:00
admin 53b15e506c fix(mobile): unify empty/loading states on question board and points pages 2026-07-21 01:45:14 +08:00
admin 6d2b5ec8b9 fix(mobile): polish candidate pages — clean text buttons, StateBlock empties, standard subpage header 2026-07-21 01:30:39 +08:00
admin 2198a92bae feat(mobile): unify page states with shared StateBlock component 2026-07-21 01:02:39 +08:00
admin ee28465c30 feat(mobile): redesign supervisor exam list, results and edit pages 2026-07-20 17:20:40 +08:00
admin da4c7f3847 feat(mobile): redesign supervisor team and assign flow with member detail and result pages 2026-07-20 16:47:26 +08:00
admin eea1ca1c6f feat(mobile): redesign profile pages per two-role design spec 
Employee profile moves the growth summary to a compact five-metric
strip with a right-aligned detail pill, masks the phone number, adds
red-filled entry icons, keeps the three key evidence scores always
visible, and switches history rows to icon + inline score with a
view-all link. Supervisor profile masks the account phone as well.
 2026-07-20 15:42:13 +08:00
admin 2dfee9f7ee fix(mobile): stack supervisor learner-mode banner title and stop button squeeze 
The inline panel-title ran into the description text and the bordered
return button was compressed into an ugly two-line wrap; give the copy
column block layout and keep the action button on one line.
 2026-07-20 14:10:19 +08:00
admin 5184672b61 fix(mobile): default h5 dev proxy to local backend on 8080 
Bare npm run dev (including hosted preview sessions) has no portless
route, so the previous default target https://wygj-api.localhost was
unreachable and every API call failed with HTTP 500. scripts/dev.sh now
passes VITE_PROXY_TARGET explicitly, keeping the portless flow intact.
 2026-07-20 13:55:50 +08:00
admin 6f2159db73 feat(mobile): redesign supervisor review flow with completion page 
Rework the pending-review list with priority badges, filters and score
ordering, rebuild the review detail around a large-score header, score
bars, dialogue bubbles and a guided review form, and add a completion
page that summarizes the review outcome and the dispatched follow-up
assignment. All three pages resolve trainee display names through the
team snapshot so organization names replace raw app account labels.
 2026-07-20 13:16:37 +08:00
admin b6a7af9e51 fix(backend): return canonical extPartyId in practice review detail 
The reviewDetail query referenced a.trainee_name without declaring the
table alias, which 500s whenever org-sync display fields are enabled,
and the mobile payload never exposed extPartyId, so the supervisor
review pages could not resolve organization member names.
 2026-07-20 13:16:18 +08:00
admin 0e46591d5f feat(mobile): redesign supervisor home as workbench dashboard 2026-07-20 11:48:01 +08:00
admin 11ef5b9721 test: align practice assertions with current behavior 2026-07-20 11:36:55 +08:00
admin dd7058a783 test: adapt exam idempotency test to model service constructor 2026-07-20 11:33:34 +08:00
admin 175a047952 fix: prevent NPE on practice finish without measured response latency 2026-07-20 11:27:13 +08:00
admin dd62533dd1 fix(mobile): prevent select-all button label wrapping in exam editor 2026-07-20 10:28:27 +08:00
admin 7716f0040d feat(mobile): AI question generation panel in supervisor exam editor 
Hints input + preset chips + count picker; generated questions land in
the existing editor blocks for the supervisor to review before publish.
 2026-07-20 10:16:40 +08:00
admin 4e9ccfa8b5 feat(backend): employee name fallback labels and AI exam question generation 
- practice team members / exam targets: when person_name is empty, show
  phone-masked label (员工·尾号XXXX) instead of a bare 员工
- POST /api/aihr/mobile/exams/supervisor/ai-generate: LLM-generated exam
  questions grounded on SOP knowledge fragments, strict JSON sanitize,
  scores rebalanced to 100; clear errors when model unavailable
 2026-07-20 10:16:40 +08:00
admin 875a1cfd2a fix(mobile): supervisor dashboard color semantics and polish 
- workbench metrics: red only when value > 0 (was always red, even zeros)
- stable alert renders as calm green strip, not a clickable pink alarm card
- quick-entry header: drop duplicated member count
- header subtitle shows today's date
- team: low-risk member cards use neutral grey left bar (was red for all)
- team: 'normal' risk count is green (was red)
- team: difficulty sort chips flex-centered
 2026-07-20 09:39:28 +08:00
admin 260b7f851e chore(mobile): reuse dev-api proxy config for vite preview 2026-07-20 09:18:39 +08:00
admin 95759d7f26 feat(backend): switch mentor TTS voice to cloned shifu-warm voice 
Already deployed to production jar on 2026-07-20; committing to keep
repo in sync with the live service.
 2026-07-20 09:03:58 +08:00
admin 578592b21d fix(mobile): flex-center pill buttons and fix learning task card layout 
- learning: task-list/snippet-list use column direction (row squeeze caused
  one-char-per-line vertical text in today-task cards)
- learning/profile: fold-toggle flex-centered
- login: code-button flex-centered
- practice: transcript-toggle / bubble-action / style-chip flex-centered
- questions: filter-chip flex-centered
- sop: modal-close flex-centered with 32px hit area
- web-ai: bubble-action flex-centered
 2026-07-20 09:03:58 +08:00
admin 0a5fc9a463 feat(assistant): unify confirmed memory capture 2026-07-20 02:30:06 +08:00
admin 46a6156d8c fix(mobile): highlight new-chat button and flex-center composer labels 
- AskSourceHeader: solid brand-red new-chat button, flex-centered label
- replace padding+line-height centering with flex on chips, hold-talk,
  send, bubble-action, transcript-toggle, system-text, scope-note
- carries the in-flight AskSourceHeader extraction on sop/web-ai pages
  (parallel session work; its test updates intentionally left uncommitted)
 2026-07-19 23:21:24 +08:00
admin 61bc7191d1 feat(web-ai): support qwen image search safely 2026-07-19 22:58:20 +08:00
admin a20ca50b87 feat(mobile): redesign profile page with identity card and deduped entries 2026-07-19 22:39:57 +08:00
admin c17d83fc59 feat(web-ai): render markdown answers safely 2026-07-19 22:02:49 +08:00
admin 7bd2c540fa feat(web-ai): support qwen web search agent 2026-07-19 21:32:39 +08:00
admin 6853facbe0 fix(web-ai): store encryption key in database 2026-07-19 20:20:57 +08:00
admin c47bf748cb fix(mobile): replace emoji glyphs with uni-icons and flex centering app-wide 
- ChatComposer panel contract: callers now pass uni-icons type names
  (image/sound/list/help/checkmark) instead of emoji
- voice playback buttons: sound icon + pulse animation, no more 🔊/◼ chars
- practice recent list: rare symbols ⌕⌁◉ (tofu risk) -> wallet/gear/chatbubble
- sop resource icon ▶/文 -> 视/文 character badges
- back arrows: 22 char-based ← -> uni-icons back across 21 pages
- line-height centering -> flex in 8 spots (back buttons, media-remove,
  turn-voice capsule, option-marker, text-button, play-question)
 2026-07-19 20:16:05 +08:00
admin 925b48e436 feat(work-report): add voice-first reporting flow 2026-07-19 12:17:11 +08:00
admin 24da833fa6 feat(personal-assistant): add confirmed project memory 2026-07-19 08:19:24 +08:00
admin 151cc26caf fix(mobile): precise centering in ask page switch and composer 
Replace line-height centering with flex + fixed heights, and swap
emoji/unicode glyphs (＋ × ⌨ 🎤) for uni-icons vector icons so
alignment and rendering are consistent across platforms.
 2026-07-19 03:27:46 +08:00
admin 767e40ca7b docs(work-report): add implementation agent prompt 2026-07-19 03:21:16 +08:00
admin 6074258ea6 docs(work-report): define voice-first reporting flow 2026-07-19 03:17:16 +08:00
admin 9210b7dde6 feat(speech): switch mentor voice to anna and slow warm speed to 0.88 
diana auditioned too bright for the mentor persona; anna measures
steadier (median F0 220Hz, lowest variance) in production auditions.
 2026-07-19 03:01:48 +08:00
admin 7c67a38508 docs(personal-assistant): link voice-first design reference 2026-07-19 02:54:30 +08:00
admin 5760a5b70a feat(speech): switch mentor TTS voice to female diana 
Code-only commit; production deployment handled separately.
 2026-07-19 02:38:35 +08:00
admin 4c2b5deb00 feat(mobile): redesign today home and switch mentor persona to female 
- today: immersive community hero + hold-to-talk voice card, numbered
  schedule rows, common-functions grid (learning/report/question-board)
- login: female mentor bust; sop: female mentor chat avatars
- speech: raise browser-fallback mentor pitch to match female voice
 2026-07-19 02:38:18 +08:00
admin a8460bbaed docs(personal-assistant): define confirmed memory and simpler UX 2026-07-19 02:32:26 +08:00
admin 1028c55f42 docs: record ask memory production release 2026-07-18 21:33:44 +08:00
admin 064816a784 feat(knowledge): add short-term ask memory 2026-07-18 21:22:49 +08:00
admin 2eda757483 docs: define ask-master conversation memory plan 2026-07-18 20:47:26 +08:00
admin 9a13ba7873 docs: record production TTS profile verification 2026-07-18 18:34:57 +08:00
admin 68d5600534 feat(speech): add role-aware emotional TTS profiles 2026-07-18 18:27:14 +08:00
adminandClaude Fable 5 c0538ad8c3 feat(mobile): learning page fold round 2 — deepen APP-like compression 
用户复查后继续压缩(折叠不删减):
- 岗位 SOP:要点默认只显示 2 条,与答案共用「展开全文(含N条要点)」
  开关(原要点全量常显,该面板 1122px 是首因)
- 训练前预习卡:默认只显示「要点」,红线/可说话术收进
  「展开红线与可说话术」(练前场景里预习卡与练页预习视图重复度高)
- 错题本默认 3→2 条(展开全部不变)
- 学习后动作:三个竖排大按钮改为一行三列

高度变化:5452(初始)→3890(第一轮)→3039px(本轮,累计 -44%);
剩余为 8 个功能面板骨架高度,继续压需砍面板,violates 不影响功能原则。
面板分布:错题本 701→544、岗位SOP 1122→738、预习卡 596→361、
学习后动作 234→116。

验证:本地折叠/展开断言全过 + 生产复核 height<3300、双折叠就位。
发布备份 h5-before-$TS。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-18 13:43:57 +08:00
adminandClaude Fable 5 0a7eb38e05 feat: work-report module V1 — four report types with review flow 
客户 2026-07-18 语音需求落地:「报」分四种——优秀案例/操作视频/
完整SOP/有用知识,员工提交、主管或运营审核。

后端(新包 org.dromara.aihr.report):
- aihr_work_report 表(自建,utf8mb4_general_ci 显式声明,含提交人/
  审核人/状态/附件 OSS 编号,提交人与状态双索引)
- POST /api/aihr/work-report/reports 员工提交(类型/标题/说明校验)
- POST /attachment 附件上传(≤100MB,OSS 落盘,ossId 字符串安全)
- GET /reports/mine 我的上报;GET /reports 审核队列(canModerate 门禁,
  复用社区 Policy);POST /{id}/review APPROVE/REJECT
  (仅 PENDING 可审,重复审核 400)
- 审核通过后的入库动作留给运营在案例库/知识库人工完成(不自动入库,
  遵守'入知识库必须人工确认'原则)

员工端:
- 新页 pages/user/report(替换'建设中'占位):四类型卡片选择、
  标题+说明(说明支持录音转写)、附件选择上传(视频类必传校验)、
  我的上报列表(待审核/已通过/已驳回+审核意见)
- 「我」页常用功能入口文案更新

生产验证(端到端):
- API:员工提交→我的上报可见→员工查审核队列403→主管队列可见
  →审核通过→重复审核400 全通过
- UI:我页入口→上报页→选类型→填表→提交→列表显示'待审核',
  历史记录显示'已通过'+审核意见
- 后端发布(备份 ruoyi-admin.jar.bak-20260718123032,25s恢复),
  H5 已发布(备份 h5-before-20260718123032)

二期待办:主管端待审列表页(现主管可用接口审核)、通过后一键转
案例库/知识库、上报采纳积分奖励(等客户定规则)。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-18 12:38:14 +08:00
adminandClaude Fable 5 d5ea4b5451 feat(mobile): restore two old-H5 interaction loops in chat form 
旧 H5 vs uniapp 逐功能盘点(客户要求'html 版本逻辑保持')后确认的
两个交互缺口,以会话形态补齐:

1. 学练闭环:总结卡底部新增「去练一遍」(旧版'开始跟读'),
   关卡后直达练页预习视图,学完话术立刻练。
2. 建议采用:老师傅提示气泡新增「用这句话回复」(旧版'建议回复'
   的真实数据版),ChatComposer 暴露 setDraft(text)——填入输入框并
   切键盘模式,员工可修改后发送;对打字困难的老年员工是关键路径。

盘点其余结论:旧版功能全部已迁移或被更优实现替代;
团队短板热力图为写死演示数据,真实版记 P2(需后端维度聚合);
演示角色切换/假数据脚手架确认不迁。

验证:本地+生产 E2E——总结卡→去练一遍→预习视图;
对练→老师傅提示→用这句话回复→输入框回填→发送进入下一轮。
发布备份 h5-before-$TS。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-18 11:29:05 +08:00
adminandClaude Fable 5 13ec3bf20c fix(case): align project scope query with 2026-07-16 org snapshot contract 
客户反馈核验中发现:案例素材上传对所有 APP 用户报
「当前账号没有可用项目身份」——projectScopesForAppUser 查询快照表
已不存在的 project_ext_org_id 列(07-16 组织新契约更名为 project_code),
DataAccessException 被兜底吞掉后 scope 恒为空。全代码库仅此一处引用
旧列名(aihr_case_record 自身列不受影响,读写走同一解析器语义一致)。

修复后生产全链路跑通:主管上传语音(TTS合成测试音频)→ ASR 转写
→ organize 结构化(背景/处理/结果)→ curate 入库 → 员工 records 可见。
后端已发布(备份 ruoyi-admin.jar.bak-20260718095809,重启25s恢复)。

同轮生产验证(有奖征题闭环,零代码改动):员工发问→运营审核→
后台发悬赏征题(rewardPoints=50+deadline)→员工回答→审核→采纳最佳
→积分0→50到账(明细正确)、重复采纳不重复发奖、员工调管理接口403、
员工端问题榜/积分页 UI 呈现全部正确。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-18 10:01:53 +08:00