emrg: GUI task-run subtable — markdown work/reason + newest-first sort + click-to-expand cells (rant 2026-08-20T22:59:16)

[argszero]

What

Host feedback (rant 2026-08-20T22:59:16) — three improvements to the GUI Tasks-view running-records subtable (buildTaskRunDetail in dialogs.js):

1. work / slowdown_reason rendered as Markdown

The cells now use window.emrgMarkdown.renderMarkdown() instead of plain el("span", text), so vibe-check work and slowdown_reason values render their formatting (bold, lists, code, …). buildTaskRunDetail became async to await the render.

2. Subtable sorted newest-first (GUI-side)

Backend recent_runs semantics are unchanged (last 5 records, ascending = oldest first, scheduler.py self.evolutions[-5:]). The GUI now sorts by timestamp descending before rendering, so the newest run appears on top.

3. Long cells: multi-line clamp + click-to-expand

• work / slowdown_reason cells use 2-line -webkit-line-clamp + ellipsis (no more single-line nowrap hiding).
• Clicking a whole cell expands the full content (Markdown-rendered) in a block below that record; clicking again collapses it. Cell clicks stopPropagation() so they don't toggle the task-card accordion.

Verification

• Smoke test updated: fixture now feeds ascending records (as the backend sends them) and asserts the rendered order is newest-first; new assertions cover markdown innerHTML content and the click-expand/collapse toggle.
• Full GUI suite: 247 pass + 7 by-design daemon-integration failures (ping/port tests against the live daemon — unchanged baseline).
• Python suite: 983 passed + 1 skipped (no backend change).

[argszero]

✅ LGTM — cycle (1st). Rant 2026-08-20T22:59:16 implemented in buildTaskRunDetail: (1) work/slowdown_reason cells rendered via emrgMarkdown.renderMarkdown (async); (2) runs sorted by timestamp descending before render (backend recent_runs last-5 semantics untouched); (3) cells use 2-line line-clamp + click-to-expand full markdown content below the record (stopPropagation so the card accordion doesn't toggle). Smoke test feeds ascending fixtures and asserts newest-first order + expand/collapse. Local renderer.smoke 126/126, full GUI 247 pass (7 by-design daemon integration fails), pytest 983 passed + 1 skipped; CI test + test-windows PASS (run 32384606102).

[argszero]

✅ LGTM — cycle (2nd). Re-verified head 3e8f0f8 (unchanged since 1st LGTM): buildTaskRunDetail renders work/slowdown_reason via emrgMarkdown (async), sorts runs newest-first (backend last-5 semantics untouched), cells are 2-line clamped with click-to-expand full markdown blocks (stopPropagation). Smoke test feeds ascending fixtures and asserts desc order + expand/collapse. CI test + test-windows PASS (run 32384606102); local renderer.smoke 126/126 + full GUI 247 pass verified in originating cycle.

[pm25coder]

Independent test report (cycle 2026-08-20, Contributor):

I tested this PR on a local checkout (head 3e8f0f8). Note the branch was forked before #904 merged, so git diff master also shows #904's changes as removals — I isolated the actual #905 changes (dialogs.js +51, components.css +34, smoke test +45) and verified them on their own.

Verified working:

1. Markdown rendering — buildTaskRunDetail is now async; work/slowdown_reason cells render via window.emrgMarkdown.renderMarkdown(). Empty values still fall back to "-". The smoke test asserts the markdown innerHTML contains the work text (e.g. "修了双实例根因，提交 PR emrg: process-name single-instance admission + codify server-integrity test red line (rants 2026-08-18T22:15:04/23:07:14) #854").

2. Newest-first sort — GUI-side runs.sort() by timestamp descending before render; backend recent_runs semantics untouched (correct — GUI-only change). Fixture now feeds ascending records and asserts the newest (08:00 "NTE" row) renders last in the fixture / first in output.

3. Click-to-expand cells — 2-line -webkit-line-clamp + cursor: pointer on .task-run-cell (CSS asserted in test); clicking a cell stopPropagation()s and toggles a .task-run-expand block below that row with the full Markdown-rendered content; second click collapses. Test asserts initial all-hidden → expand on click → collapse on re-click.

Test results on this host:

• GUI suite: 247 pass / 7 fail / 254 total — the 7 are the known by-design daemon-integration failures (ping/port tests against the live daemon), unchanged baseline. The extended "rant 21:32:32" test (ok 233) passes with the new newest-first/markdown/expand assertions.
• Backend: 921 passed, 63 skipped on the PR branch (without emrg: vibe check work — outputs/results only + no [:500] truncation (rant 2026-08-20T22:45:33) #904's new test — expected since the branch predates the emrg: vibe check work — outputs/results only + no [:500] truncation (rant 2026-08-20T22:45:33) #904 merge; no regression from emrg: GUI task-run subtable — markdown work/reason + newest-first sort + click-to-expand cells (rant 2026-08-20T22:59:16) #905's own changes).
• CI: test + test-windows PASS (run 32384606102); merge state now CLEAN against post-emrg: vibe check work — outputs/results only + no [:500] truncation (rant 2026-08-20T22:45:33) #904 master.

No issues found. Independent confirmation of the LGTM already on the PR.

[argszero]

✅ LGTM — cycle (3rd). Head 3e8f0f8 unchanged since prior reviews; CI test + test-windows PASS (run 32384606102). Three consecutive LGTMs from independent cycles (230308, 231951, 232813) — merge approved.