- add 操作手册/培训手册 option and label mapping in knowledge processing
- suggest OPERATING_MANUAL on review form when filename contains
手册/指南/教材 and authority is unset/EMPLOYEE_SUBMISSION (prefill only)
- train/owner-profiles.vue: table + create/edit dialog + enable/disable
with confirm; project options from org snapshot with manual fallback;
concerns limited to 500 chars with content-discipline notice
- route registered under 训练运营 with operationsConsoleRoles
- default persona falls back to owner-calm; picker below persona cards
(compact single row), default own community via defaultProfileId,
storage persistence incl. explicit none, hidden on empty/failure
- profileId passed to realtime session creation when selected
- remove owner-impatient persona; owner-calm renamed to 业主
- aihr_realtime_owner_profile table (tenant+project unique, enabled,
version) with formal migration
- admin CRUD under moderator role gate; project code validated against
org snapshot and project name taken from snapshot authoritatively
- employee endpoint returns display-safe fields plus defaultProfileId
from the employee's own community; profile body never exposed
- session creation renders {community}/{profile} server-side; client
text can never enter instructions; invalid profileId rejected,
default resolution fails closed
- duplicate create returns 409
- citation chips: 正式经验 (CASE_EXP) / 同事经验·有奖征答 (COMMUNITY)
- citation navigation: CASE_EXP opens case detail via ?caseId=,
COMMUNITY goes straight to question detail via community-question
detailRef, falling back to the question list
- disable human-help action when the answer was already marked helpful
(server would 409)
- server-side employee position resolution (org snapshot) with hard
position gate for EXPERIENCE/CASE sources; formal sources unaffected;
unresolvable position fails closed to general-scope only
- merge published experiences (aihr_case_record, CASE_EXP) and moderated
best answers (COMMUNITY, no submitter identity) into the citation pool
as REFERENCE tier; auxiliary sources never alter formal answers and
fail without breaking the main chain; grading re-grades ROUTE into
GUIDANCE when only auxiliary evidence exists
- case recall endpoints enforce the same position gate
- moderator endpoint to promote a best answer into a CANDIDATE case
record (idempotent via source_ref unique key, never auto-published)
- migration aihr_20260822_case_source_ref_and_best_answer_index_mysql8.sql
- COMMUNITY citations carry community-question:{id} in detailRef for
direct navigation
- CHANGELOG: 2026-08-05 entry with deploy evidence, verification
levels and open items
- runbook: execution log entry + two new migrations in the sequence
- AGENTS.md: macOS GNU stat shim requirement for release scripts
- bump @dcloudio/* to 3.0.0-alpha-5020320260803001 to match the
HBuilderX-Alpha 5.23.2026080313 debug base (fixes HTML5+ Runtime
5.14/5.23 mismatch dialog on device)
- add .env.development with VITE_SMS_VERIFICATION_ENABLED=true so the
SMS code field shows in run mode; on-device debug talks to prod
backend where verification and the pilot fixed code are enabled
- update MOBILE_APP_PACKAGING.md toolchain line
- narrow amount-semantics clarification to adjudicative intents only
- persist pending clarification on conversation and resolve follow-up
replies as ANSWERED/GAVE_UP/NEW_TOPIC (max one round per field)
- add AihrEvidenceGradingService: FULL/PARTIAL/GUIDANCE/ROUTE arbitration,
policy-claim post-validation, reference answers labeled as non-formal
- unblock retrieval source gating; tag citations with evidence tier,
keep PUBLISHED/HUMAN_VERIFIED/effective-date authorization gates
- replace operator-facing no-evidence copy with actionable route exits
(human help, direct finance channel, web research)
- allow fragment-less down feedback for own NO_EVIDENCE queries so
ROUTE answers can escalate to human help (fragment_id nullable)
- migrations: aihr_20260804_pending_clarification_mysql8.sql,
aihr_20260804_feedback_fragment_nullable_mysql8.sql
73712994 disabled voice-first on native by gating the TTS preload behind
!isNativeAppRuntime(), which left App users with the degraded text bubble
plus a 播报 button. The voice-line, 转文字 toggle and 收起全文 markup were
still present but unreachable.
Re-enable the preload on every platform and open the transcript when the
message is created, so the answer is readable while synthesis runs.
Collapse the transcript when playback starts rather than when synthesis
finishes: on-device synthesis settles in 2-6s, so collapsing on ready
pulled the answer away mid-sentence. transcriptPinned records a manual
toggle so the automatic collapse never overrides the reader.
Bound the preparing bubble at 20s (the upstream DashScope ceiling) so a
stalled synthesis degrades to the plain text bubble instead of showing an
endless placeholder duration.
Replace the tautological assertion that pinned the previous behaviour by
matching source strings with assertions on the behaviour itself.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Covers the App-Plus batch in this series: renderjs realtime practice, request
timeout abort, App-Plus date/query API replacements, and on-demand native
voice. Native permissions and plugins are unchanged, so this ships as a normal
cloud rebuild.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Document the `/api/aihr/org/sync-changes` change feed shipped in 30ef6a8f:
`employee.id` is the stable subject key, the dry-run-then-apply order, and the
fail-closed rules that full `/sync` keeps for masked phone numbers and
duplicate memberships. Adds the endpoint row and a curl example alongside the
existing full-sync entry.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The planner falls back to CLARIFY whenever no keyword matches and the model
plan is unavailable or malformed. Mid-conversation that produced a dead end:
a follow-up such as "如果我要报销，一般需要查哪几个文件？" was answered with a
clarification request even though the previous turn had established context.
When a non-media request carries a conversationId and a contextVersion above
zero (i.e. at least one turn has completed), route CLARIFY to KNOWLEDGE_QA
with a grounded knowledge search.
Tradeoff: this also suppresses the deliberate CLARIFY for explicitly vague
phrasing ("你看着办", "帮我处理一下") from the second turn onward, which will
now run a knowledge search instead of asking back. Blank input is still
rejected earlier by requireQuestion.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
On native App the eager TTS request delayed the answer bubble: the message was
created in `preparing` state, so the text stayed behind a voice line that was
still waiting on a round trip.
Skip the preload when `isNativeAppRuntime()`, marking the voice `unavailable`
so the text and citations render immediately. The existing `unavailable`
branch already exposes a play button that calls `toggleAnswerSpeech`, so voice
stays reachable on demand instead of being dropped. H5 keeps preloading.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
App-Plus has no WebRTC or getUserMedia in the logic layer, so realtime
practice was H5-only and the native entry had to advertise a record-then-
transcribe downgrade.
Move the media/WebRTC engine into `realtime-browser-engine.js` and drive it
from `RealtimePracticeAppBridge.vue`, whose renderjs script runs inside the
system WebView where those APIs do exist. The logic layer keeps the
authenticated calls: SDP exchange and `search_knowledge` tool invocation stay
server-proxied, so the access token is never handed to the view layer. Both
sides talk over the renderjs bridge only in session payloads the server
already assembled.
Verified against a real App build (`app-renderjs.js` holds RTCPeerConnection
and getUserMedia; `app-service.js` holds the SDP endpoint, renderjs side has
zero hits). Microphone permission, live transcription, remote playback,
foreground/background switching and disconnect fallback still require Android
and iOS device verification.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`URLSearchParams` is absent from the App-Plus JS core, and
`toLocaleDateString('sv-SE', { timeZone: 'Asia/Shanghai' })` depends on full
ICU data that the App runtime does not ship, so the work-result pages built
either a broken query string or the device-local date.
Build query strings with `encodeURIComponent`, and centralise the business day
in `currentShanghaiDate()`, which shifts the epoch by a fixed UTC+8 offset
before slicing the ISO date. Asia/Shanghai has no DST, so the fixed offset is
exact rather than an approximation.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Some uni-app platforms ignore the `timeout` option, leaving the returned
promise pending forever and hanging the calling page with no error path.
Drive the deadline locally: keep a single-shot `finish` guard so success,
failure and timeout can only settle once, and on timeout call `task.abort()`
before rejecting so the in-flight request is released instead of leaking.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`@dcloudio/uni-app-plus` was never declared, so `uni build -p app` silently
fell back to an H5 shell: exit code 0, `DONE Build complete.`, but the output
was only `index.html` (still carrying the `/h5/assets/...` base path) with no
`app-service.js`, `app-renderjs.js` or `manifest.json`. Because
`docs/MOBILE_APP_PACKAGING.md` lists that command as a local verification
step, App platform code was reported as built without ever reaching the App
compiler.
Declare the compiler at the same version as the other `@dcloudio` packages,
guard the version alignment in `tests/app-packaging.test.mjs`, and document
that success is judged by the emitted artifacts, not by the DONE line.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- rename brand docs to 帮道 prefix; reframe 20260721 plan as historical snapshot
- compress AGENTS.md rule detail into API_INTEGRATION.md, keep boundaries only
- scrub test phone numbers from design-qa, DEV_SETUP, org-sync and rebuild docs
- document summary-card evidence binding contract and 20260725 config migrations
- pin relative dates in README and audit baseline
admin
