The unused linter reports eight symbols in prune.go that the archive-to-file change left without a caller; the strategy type and its hint mapping, which are still used, stay.
The setting promised it above the input box while the chip draws inside the composer's top edge, as the design board shows. Both lines of the setting now say top.
With two currencies the third chart took a row of its own and left the right half of the page empty; the spend charts now stack in one column beside the usage chart. The daily table's header is 11px medium in the interface face, not 13px bold mono, and its rows take 9px padding.
Overview: four figures (requests with the main/sub split, cost, tokens, cache hit), each source as a bar with its cost, and a footer with the work time and a link to the context details; the context window is no longer drawn twice. Context details: the compact action, the MCP count and the manage link share one footer row instead of a server list, and a row with a breakdown shows its figure and a chevron, not a bordered pill.
The veil starts below the title bar, so with the enlarge mode gone its text showed sharp above the blurred app. A strip beside the veil, not inside it because a blur only reaches what is within its own element, now blurs that band and lets clicks through to the title bar's buttons.
The sidebar's settings button and the common settings page wore a circle with eight rays, which is the sun the appearance page and the light-or-dark control already are, and at 16px the two could not be told apart. Both now use one toothed gear; the appearance page keeps its sun.
1360 by 900 at most, over the dimmed app; a smaller window still gets it filled. On a 2000px window the content was either 700px columns or a block with 180px of empty margin each side. Follows the revised board Set-Wide.
The sheet already fills the window, so enlarging only took it over the title bar and added three window buttons beside its own two. The control, the stored preference, the window buttons it summoned and the drag-region rules that went with it are removed; the board draws no buttons in that corner besides closing.
A block that applies on this machine at once says nothing; one that needs a restart, belongs to the workspace and so on gets a tag beside its name, and the page title carries one line saying what the usual is. Descriptions stay on one line with the whole text on hover, rows take 6px padding, and a row with no description is a single line. Follows the board Set-Full.
The right column of a settings row was sized to the widest line of the note under its control, so at 572px a column took 501 and the label shrank to one character. It is now capped at 320px and the note wraps. The page title takes the block's width and centering.
Settings: the card column is centered, 760px for single-column pages and 1160px for two-column ones, with 常用 added to the two-column pages; below 860px it stays one column. Changes: past eight files a new or untracked file shows a tag in its own header instead of a folded row under it, the top line splits modified from untracked, and the header weights, the 14px checkbox and the note's weight follow the board. Follows the new boards Set-Wide and WB-Changes-Many.
The model panel pins its title, filter and footer and scrolls just the list inside its own column, with the effort column fixed beside it and both capped at 440px. The agent preset menu is 340px with 12px descriptions and the workspace dialog 340px. TOKENS.md now states the picker panel rules the board Spec-Pickers draws instead of the older 12px-corner, 11px-description line. Menu items can be marked foot to stay under the scrolling rows.
600px instead of 700, capped at 440px tall; model rows are one 34px line with the provider note and badges beside the name; effort rows lose the strength bars and the third line; the column fits its content and stays in view, and its footer is the single line the board draws.
300px panel, a bold title with the scope on the right, two-line rows with a 12px description, the current row in a grey wash with a blue tick. The unimplemented Ask row and its two strings are gone; it comes back when the mode exists.
No border or box: a 26px line in the transcript's own type, the running label shimmers as text (the same 1.6s light as thinking) and nothing else moves while it is folded, open it is an indented list on a hair rule, a failure is a small pill, and a finished group folds to one faint line of count and time. Follows the Think-Exec light variant.
360px panel, a small two-segment switch with the creation note under it, a source-branch label, 32px search and branch rows in the mono face with a pill for the current or occupied branch, and the one-line note at the foot. The per-row chevrons, dividers and the switch's icons and ticks are gone; the folder name and path stay as quiet lines because the board does not draw the current-folder page.
The note is a borderless warn wash inside the input box with a text link, not a bordered strip with a filled button. The pill's pull-up margin assumed nothing above it in the box; with a note present it now takes its own line.
The board sets body text at 500; the default weight ladder is 500 / 600 / 700, so 500 is --w-reg. Rules written as --w-med drew every line one step heavier than the board, and on Windows 600 falls on the bold face. Side panel answers also take the board's 1.7 line height.
The mark for a model came only from the preset a provider was installed from, so an entry pointing at api.deepseek.com by hand drew a letter tile. The address is now read through the curated preset serving it, as vision models already are. Setup service cards draw the mark their vendor declares.
Compaction moves the folded span into <session>.outputs/fold-<hash>.md and leaves a pointer the model reads back with read_file. The summarizer path, its retries, bounds and tests are removed.
Answering an ask with no selection ends the turn normally instead of reporting a failure.
Find bar, overview and context cards, compact dialog, image preview, root-missing banner, toasts, side panel and prompt refine are aligned to the Chat-Tools board; the agent preset picker follows the Agent-Picker board.
A fixed ten seconds reported a hang on a loaded Windows runner where the first-run test had already spent forty seconds. testenv.Budget exists for waits that only catch a hang.
A removal that lost a race with the walker was dropped, so the directory stayed behind and, because names are never reused, a later walk could not tell it from an unreadable one and failed after the five-second re-stat window. The churn now retries the removal until it succeeds, which is what the never-reused-name invariant assumes.
The running-copy check matched processes whose path it could not read by name, and the name list included cua-driver.exe, which other computer-use apps ship too. An elevated or orphaned driver from another app therefore read as Studio still running, and the retry loop could never succeed. Only the app's own executable and its host are matched by name now.
yhh