With the colour picker open on an image fill in Crop mode, a press on the layer closed the picker, and closing it committed the drag as a plain move: the whole layer moved instead of the picture inside it. The picker now treats a crop drag as its own control, as it does the open gradient's handles (#1852), so the press doesn't close it and the drag moves the crop.
A drag whose gesture something else committed under the held button went on calling the engine's preview, whose "no gesture in progress" error reached the canvas as a message. That error is now its own EngineError::NoGesture, which the canvas drops: the drag it belonged to is over, and it says nothing to the user.
Tests: crop_image_tests::crop_from_the_colour_picker_then_a_drag_on_the_image_moves_the_crop_not_the_layer (failed before: the crop didn't move) and a_drag_whose_gesture_was_committed_under_the_button_shows_no_message (failed before: "no gesture in progress").
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and finished; Sonnet 5.5 wrote at medium (cut off by an API error)
Shift+Cmd/Ctrl+V pasted into the selected layer's parent at its top-left,
but appended the copies, so they went to the top of the parent's layers
whatever was selected. In an auto layout frame that put them at the end of
the flow, away from the layer they were pasted over.
They now go just above the selected layer, in the copied order: in the
Layers panel the row above it, in an auto layout frame the item after it.
A section pasted over a layer in a frame still goes on the page, on top
of its layers, as before. Paste to replace already kept the replaced
layer's place.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
board.mindMap's default x and y came from the view's centre, which is on the page, but x and y are the root's place in its parent: under a section at (1000, 500) the map landed 1000 right and 500 down of the view's centre. The default now takes the centre through the parent's inverse world transform (a parent that can't be inverted gives a non-finite centre, which the coordinate check refuses). Explicit x and y stay parent-local. Tests: two in board_mindmap_tests (a section; a frame turned 30 degrees, with a given place carried through the frame).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove; Opus 5.5 briefed; Sonnet 5.5 wrote at medium
inspect.code {language: "tailwind", mobile} dropped every mobile max-width and put the desktop one behind lg:, so a fixed-size layer that is the same size on both frames (a 48 x 48 badge) stretched to the full width below 1024 px. A mobile max-width that equals the desktop one is now kept as a plain class, with no lg: copy. A differing max-width still goes fluid on mobile with the desktop cap behind lg:.
Tests: the issue's two frames (a 1440 and a 390 page, each with a 48 x 48 badge) keep max-w-[48px] unprefixed, and a page-wide banner still gets its lg: cap only. The slashed-names test gave the CTAs frame the same 342 px on both frames, which the new rule rightly keeps, so its desktop frame is now 600 px wide. docs/mcp.md notes the rule.
Model-Effort: Opus 5.5 drove; Fable 5.2 queued; Opus 5.5 briefed; Sonnet 5.5 wrote at medium
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
flutter_line skipped `spacing:` when the main-axis alignment string
contained "Space", but the strings are MainAxisAlignment.spaceBetween,
spaceAround and spaceEvenly (lower-case s), so it never matched and the
gap was written on top of the distribution. It now checks the MainAlign
value itself.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and wrote; Fable 5.2 spotted it
Part of #2271. The native targets took ONE colour from a layer (`first_fill`: the bottom visible solid, or a gradient's first stop) and dropped every other fill, so a frame with white, a gradient and 50% black came out as flat white.
Now SwiftUI, Compose and Flutter write every visible solid and gradient fill of a frame, group, rectangle or ellipse, in paint order, and a gradient as a gradient. A shared `visible_fills` gives each target the same list (document order, bottom first); each target only formats it.
- SwiftUI: chained `.background(...)`, top fill first, since each one goes behind the ones before it. A rectangle or ellipse fills with the bottom paint and lays a copy of the shape over it (`.overlay`) for each paint above, bottom first.
- Compose: chained `Modifier.background(...)`, bottom fill first, since each one draws over the ones before it. A linear gradient snaps to one of eight directions (a modifier doesn't know pixels); radial and angular ones sit at the box's centre.
- Flutter: a `BoxDecoration` paints one colour or gradient, so the Container keeps the top fill and each fill under it becomes a `DecoratedBox` around the widget, the bottom fill outermost. The drop shadow goes on the outermost box so it never lies over a lower fill. An instance's call is now made once, in `flutter_placed`.
A layer with one solid fill gives the same code as before, variable names included (only the paint at index 0, when solid, takes the `fill` variable's name). A lone gradient used to come out as its first stop's colour and is now a gradient.
Still not written: image and pattern fills in the native targets, ArkUI (one background and one gradient per component, so a stack needs extra nodes) and text with several fills. Figma's iOS and Android output was not measured: the order follows how each platform paints.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove; Fable 5.2 briefed; Sonnet 5.5 wrote at medium
inspect.code wrote one `background` per visible fill, so only the last survived. A non-text layer now gets one `background` with every visible fill as a layer, top fill first (a document's fills run bottom to top), joined with ", ". A solid above another fill is `linear-gradient(0deg, c, c)` since CSS allows a plain colour only as the last layer; the bottom solid stays a plain colour (and keeps its variable token); gradients, images and patterns keep their layers. A single fill is unchanged. Text keeps one `color` per fill. Tailwind emits a layered value as one `[background:...]` class. Native targets are unchanged.
Tests: several_fills_are_one_layered_background_top_first, one_solid_fill_stays_a_plain_colour, several_fills_are_one_layered_background_class.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove; Fable 5.2 queued; Opus 5.5 briefed; Sonnet 5.5 measured and wrote at medium
atan2 of two zeros reads as 180 degrees, so the centre gave 3; it now gives 60,
and the test asserts that exactly.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and fixed
A selected polygon or star shows a Count handle on its first point clockwise from the top (a polygon's vertex 1, a star's outer point 1, not inset). Dragging it sets the count to round(360 / the pointer's clockwise angle from the top point about the centre), measured in the shape's own frame and clamped to 3 to 60, written through node.set pointCount in one undo step. The readout and hover label read "Count N". Measured in Figma's web app on 2026-11-12; a polygon's maximum wasn't read, so it is clamped at 60 like the star's.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove; Fable 5.2 queued; Opus 5.5 briefed; Sonnet 5.5 measured and wrote at medium
The engine test now follows the issue: a list typed at 5, saved as .fig and opened, split by Backspace at the start of "b". The REST fixture is a top-level list, the junk sweep reaches the new loop through a listed text layer, and the document crate tests the new rule for nested, bullet, plain and over-deep levels. ParaList gets its doc comment back, and the revision row names the item written with 0.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove; Fable 5.2 queued; Opus 5.5 briefed; Sonnet 5.5 wrote at medium
Figma writes a list's start on every line of the list (measured in Figma's web app, 2026-11-12, from a saved .fig: a list started at 5 has listStartOffset 4 on all its lines, and the item split off by Backspace is written with 0). The import kept every line's value, so after a split the items after it restarted at the old start instead of 1.
The import now keeps a start only on the item where a numbered list starts counting, and gives every item that goes on counting 0, for .fig and REST listStartOffsets alike. The rule is one function in the document model, list_item_starts_counting, which the layout's list numbering now uses too. The export is unchanged and its re-import now equals what went in.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove; Fable 5.2 queued; Opus 5.5 briefed; Sonnet 5.5 wrote at medium
Measured in FigJam's web app on 2026-11-12: a sticky's own toolbar has a text-size menu (Small 16, Medium 24, Large 40, Extra large 64, Huge 96), a fresh sticky is Small = 16 (the 18 of 2026-11-09 did not reproduce), the note grows taller at Huge and shrinks back, a size chosen with two stickies selected sets both, and the sticky tool remembers the last size.
STICKY_TEXT_SIZE is 16 and STICKY_SIZES holds the table. board.setStyle {fontSize} sizes stickies' notes and sets the sticky tool's size (BoardTool.sticky_size); board.sticky takes the tool's size or a fontSize (1 to 1000). The contextual bar shows a size menu (the current name or number, a check on the current entry) when only stickies carry text; shapes keep S/M/L/XL. Docs: boards-parity rows, revision history and status line.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove; Opus 5.5 briefed; Sonnet 5.5 wrote at medium
bflatastic
Trevor BrownandClaude Sonnet 5.5
Denis GomesandClaude Opus 5.5
NetterSilverandClaude Opus 5.5