[PeterGuy326]

Preflight

• Searched open and closed PRs for related changes.
• No credentials, personal data, private paths, or private screenshots.
• Single squashed commit 16cee40; parent is main @ 15ad3ab (current head), ancestry clean.

Summary

Sept 27 client acceptance pass, delivered as one squashed commit on top of current main. Five areas of the desktop client move together:

1. Live markdown editing — documents are edited in place through Vditor IR (docs/LiveMarkdownEditor.tsx) instead of a read-only render plus a separate composer path, with a full shortcut set and an in-app cheatsheet.
2. Turn timeout semantics — the turn timeout becomes an inactivity window instead of a hard wall-clock cap, so a turn that is still producing output is no longer killed mid-stream.
3. Navigation — entries are grouped by context, the employee header stays stable while switching modules, and editor scrolling is bounded to its own container instead of the page.
4. Docs chrome — the document toolbar is consolidated and the collaboration chrome simplified.
5. Memory page — the source sidebar is replaced by top tabs; the file list and frontmatter rendering are compacted.

Change

Renderer (apps/desktop/renderer)

• docs/LiveMarkdownEditor.tsx (+378) / .css (+169): new Vditor IR editor surface, shortcuts, cheatsheet.
• docs/DocsPanel.tsx (+842), DocsModule.tsx, DocViewer.tsx: toolbar consolidation, live editing wired into the docs flow.
• approvals/ApprovalDetailDrawer.tsx (−446) → approvals/ApprovalDetail.tsx (+378): drawer replaced by a dedicated detail surface; ApprovalQueue.tsx updated to match.
• turns/TurnPanel.tsx / TurnThread.tsx, groups/GroupsPanel.tsx, memory/MemoryModule.tsx: navigation grouping, stable header, bounded scrolling, memory tabs.
• markdown/Markdown.tsx: readable-render adjustments for the new editor.

Main process (apps/desktop/src)

• docs-ipc.cjs (+131), workspace-ipc.cjs (+23), main.js, preload.js: IPC for live document editing and workspace reload.

Server (apps/server)

• routes/docs.ts (+452): document read/write surface backing the live editor.
• engine/driver-cli.ts, turns/store.ts: inactivity-window turn timeout.
• context-sources.ts, server.ts: wiring.

Shared / i18n

• packages/shared/src/docs.ts (+53), api.ts, errors.ts; packages/ui/src/locales/{en,zh}.ts (+107/+109).

New dependency: vditor@4.0.0 (pinned) in apps/desktop/package.json.

Evidence

Item	Result
Diff	71 files, +8598 / −2045
Base	main @ 15ad3ab, parent of head 16cee40
New/expanded specs	docs-lifecycle.test.tsx (+889), live-markdown-editor.test.tsx (+469), App.test.tsx (+765), approval-enhancements.test.tsx (+387), docs-ipc.test.cjs (+88), server/test/docs.test.ts (+422), reports.test.ts (+75), turn-driver.test.ts (+28)
New smoke scripts	smoke-live-markdown.cjs, smoke-approval-workspace.cjs, smoke-graph-scroll.cjs, smoke-navigation.cjs, smoke-loading-motion.cjs
CI on this PR	pending — will be reported before merge

Honest caveat

This commit is an acceptance bundle produced on a machine that cannot push to GitHub directly; it was transported as a signed-off bundle and restored here, so the authorship path is bundle → fetch → push rather than a normal local branch. Local build/test verification is being run in parallel; results will be posted as a comment before merge, and merge is gated on green CI plus review.

Verification

• Pending: renderer suite + server suite + production build (results to be appended in a comment).
• CI: ubuntu / macOS / CodeQL / Dependency Review on this PR.

Acceptance

• Document editing works in place, shortcuts and cheatsheet reachable from the editor.
• A long turn that keeps emitting output is not terminated by the timeout.
• Navigation groups entries by context; the employee header does not remount on module switch; editor scroll is contained.
• Memory page shows top tabs instead of a source sidebar; file list and frontmatter are compact.

[PeterGuy326]

Technical review — layout regression blocks merge

CI on this PR is not green, and the failures are all in one place: layout.

Job	Result
Node 24 / ubuntu-latest	✅ pass
CodeQL / Dependency review / Scorecard	✅ pass
Unsigned installers (macOS arm64, Windows x64)	✅ pass
Node 24 / macos-14	❌ fail (Verify workbench → check-layout-parity.mjs)
Layout smoke macOS arm64 (#127 AC-004)	❌ fail
Layout smoke Windows x64 (#127 AC-004)	❌ fail
Layout parity / macOS vs Windows	⏭ skipping

Concrete errors from the macOS Verify workbench log:

layout parity FAILED:
 - macos: column bottom delta 40px exceeds 2px
layout parity FAILED:
 - left column width differs: mac 560px vs win 500px (max 4px)
Error: .../owb-parity-JJ7V4x/mac.json: layout is null — the two-column org module did not render
Error: .../owb-parity-V82q9y/mac.json: layout has no usable viewport measurement; a height difference cannot be attributed without it
Error: .../owb-parity-PKzH9R/mac.json: layout measurement never settled within its budget; a mid-layout sample cannot be compared across platforms

These line up directly with the layout-touching parts of this commit: context-grouped navigation, the now-stable employee header, bounded editor scrolling, and the memory page's sidebar-to-tabs change. The most serious signal is the two-column org module did not render — on macOS the org module is not producing a measurable two-column layout at all, which is a functional regression, not a cosmetic 2px drift.

The green ubuntu job does not contradict this: layout parity and the staging layout smoke are cross-platform checks and are the ones that run the packaged app, so they are the authority here.

Not merging. Requested follow-up before this can move:

1. Re-run the layout smoke locally and diff the org module's two-column measurements against main @ 15ad3ab; fix whichever of the four layout changes (nav grouping / header / scroll bounds / memory tabs) introduced the 40px bottom delta and the 560 vs 500px left-column split.
2. Make sure the org module reaches a settled layout within the measurement budget on macOS — layout measurement never settled suggests a late or animation-gated layout pass.
3. Re-push and confirm all three layout jobs are green, then this is mergeable.

Local note (informational only): the renderer suite on my machine reports 636 passed / 5 failed, but that environment is missing the pinned design-system package (@fullstack-ai-infra/ui/styles.css cannot resolve), so local numbers are not evidence here — CI is.

[PeterGuy326]

Update — rebased onto current main and the layout regression is fixed

Two things happened since the review comment above.

1. Rebase. main moved to b3704c45 when #515 merged (it touches App.tsx), which left this PR CONFLICTING / DIRTY. Rebased onto b3704c4; the only conflict was the activeModule union plus the new contactsOpen / groupListHost / conversationHeaderHost / groupsOpened state in App.tsx, resolved by keeping both — the new modules and state from this branch, and #515's widened ConfigurationCategory for settingsInitialCategory. Diff is unchanged at 71 files, +8598 / −2045.

2. Root cause of the layout failures — found and fixed.

The packaged layout smoke (#127 AC-004) does not navigate: it launches the real renderer and measures the two-column org workspace from its first paint. This branch moves the entry module to the new conversation module, so that first paint no longer contains .owb-org-module__left / .owb-org-module__pane--right > .owb-turn-panel, and the measurement came back null:

Error: .../mac.json: layout is null — the two-column org module did not render

That single cause produced all three red jobs (macOS Verify workbench, Layout smoke macOS arm64, Layout smoke Windows x64).

Fix (3114747): main loads the renderer with ?owbLayoutModule=org in layout mode only, and App.tsx honours that query when seeding activeModule. A normal launch carries no query and still opens the conversation module, so the acceptance change to the entry point is preserved — only the measurement path is told which module to open.

One correction to my earlier comment: the column bottom delta 40px and left column width differs: mac 560px vs win 500px lines are not a second regression. They are the expected stderr of scripts/test/layout-parity.test.mjs (the cases at lines 36 and 40 assert that those failures are detected). The real failure was only the null layout.

CI re-running on head 3114747.

[PeterGuy326]

Correction — the layout failures are structural, and my previous fix was a misdiagnosis

Head 3114747 is withdrawn (branch back to 04be66d). It did not fix anything, and the reasoning behind it was wrong. Here is what the evidence actually says.

What #127 AC-004 measures. apps/desktop/src/packaged-smoke.cjs measures exactly two elements and nothing else:

document.querySelector(".owb-org-module__left")
document.querySelector(".owb-org-module__pane--right > .owb-turn-panel")

Those elements no longer exist on this branch. On main, App.tsx renders the left column itself:

<OrgWorkspaceSplit left={<div className="owb-org-module__left"> … PositionCard … } />

On this branch App.tsx contains no OrgWorkspaceSplit and no .owb-org-module__left at all. The org module now renders either

• <section className="owb-organization-directory"> (structure directory), or
• RelationshipGraph (overview),

and the conversation side has moved into its own module (.owb-module-pane.owb-employee-conversation). The measured two-column object has been taken apart by this refactor, so the measurement has nothing to sample and returns null:

Error: .../mac.json: layout is null — the two-column org module did not render

The entry-module change (conversation instead of org) makes it worse but is not the cause: even with the org module opened, the two measured selectors are gone. My ?owbLayoutModule=org patch assumed the columns still existed underneath, which is why it changed nothing.

Also correcting my earlier comment: the column bottom delta 40px and mac 560px vs win 500px lines are expected stderr from scripts/test/layout-parity.test.mjs (lines 36 and 40 assert those failures are detected), not a second regression.

This is now an AC-004 contract question, not a bug I can patch unilaterally. Three ways forward, and they are a product/acceptance decision rather than a review call:

1. Keep the two-column org workspace — restore OrgWorkspaceSplit + .owb-org-module__left for the org module. Contract unchanged, CI green, but it gives back part of the Sept 27 restructure.
2. Re-point AC-004 at the new layout — decide which two containers now constitute "the two columns" (for example the conversation module's list vs thread) and update the selectors plus the parity thresholds together.
3. Retire the two-column measurement for this shape and measure whatever the new first-class layout is — which means editing fix(ui): align position-record and turn panels + Chinese IME + Windows parity evidence #127's acceptance criteria deliberately.

I have left the branch at the rebase (04be66d, 71 files, +8598 / −2045) with no extra patch, so whichever option you pick starts from a clean base. Tell me which one and I will implement it and drive CI.

[Bindy-lbb]

Review outcome: I cannot approve this head yet because the current required checks are 10/11, not 11/11. Node 24 / macos-14 failed after the layout checks completed successfully.

The failing assertion is renderer/test/project-board.test.tsx → clamps schedule bar calculations…: it expects the bar for Partially invalid date task (whose only valid date is hard-coded as 2026-09-22). On the macOS run dated 2026-09-28, the 28-day schedule window begins on 2026-09-28, so that task is correctly rendered as outside the window and no bar exists. The test needs a date relative to the test window (or must navigate the window) before CI can be green.

On the requested AC-004 point: the final selector pair is sidebar + shell main, not the nested turn panel. That matches the restructured first-paint conversation layout; bottom alignment remains enforced and comparing the cross-platform 48px height offset is appropriate. Minor follow-up: update check-layout-parity.mjs comments/error wording that still call the main region a “turn panel.”

[Bindy-lbb]

已按当前 head 5fcc024 复核。AC-004 的实际测量目标（协作侧栏 + shell main）与高度差跨平台稳定性规则都正确，布局 smoke 也通过。\n\n但本轮 verify 的两个 Node 24 job 仍失败，不能 approve： 的排期用例把部分无效日期任务的唯一有效日期固定为 。运行日为 2026-09-28 时，它已落在 28 天窗口之外，故没有可断言的 schedule bar（CI 报 ）。请将该日期锚定到测试窗口/固定系统时间，或在用例中导航窗口；重跑绿灯后我会立即复核批准。

[Bindy-lbb]

更正上一条 review 的文字渲染：阻塞项位于 renderer/test/project-board.test.tsx。排期测试把部分无效日期任务的唯一有效日期写死为 2026-09-22；在 2026-09-28 运行时它位于 28 天窗口之外，CI 因此找不到 Partially invalid date task 的 schedule bar。将日期锚定到测试窗口或固定系统时间（或显式导航窗口）后，两个 Node 24 job 应可恢复通过。

[PeterGuy326]

Fixed the date-dependent failure Bindy flagged in CHANGES_REQUESTED: renderer/test/project-board.test.tsx hard-coded 2026-09-22 / 2026-09-28 in the schedule-window tests, and the board renders a 28-day window anchored at the current week's Monday — once the run date passed 2026-09-28 those fixtures fell outside the window and their bars disappeared.

Changes (commit 56be576):

• Added an isoFromToday(offset) helper; window-sensitive fixtures in clamps schedule bar calculations… and safely groups malformed or invalid dates… now derive dates from today (inverted task = today+1/today, partial task due = today+5, valid task = today+1→today+4), so the assertions stay calendar-independent without pinning system time.
• The span task (2025→2027) and form-input tests are window-independent and unchanged.

Verified locally: vitest run renderer/test/project-board.test.tsx → 20/20 passed. Head is 5fcc024 + this one test-only commit; no source changes.

[Bindy-lbb]

Verified the date-dependent schedule fixtures now derive from today, and all 11 checks pass on 56be576.