The reporter's case: the pane sits in a frame over another layer. Exported on its own it keeps
the blurred layer beneath it and the frame's fill behind it, opaque, as on the canvas.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and wrote
An export of a layer draws only that layer, so a background blur or glass in it read an empty
backdrop. The export now renders the page's layers below the first exported layer, with the margin
the effects reach, as the backdrop (render_area_over), clipped to the export area; only the exported
layers are drawn on top. Layers without those effects, and every other export path, are unchanged.
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
Cover inherited Boolean bindings and independent same-component slot content. Strengthen text, HTML and stored-property assertions, split default and own-content cases, and document ownership-based binding resolution.
Measured in Figma: a layer's rotation field shows its rotation on the page, also inside a rotated frame or group, and Shift snaps that value. A test pins it; the doc comment on shared_rotation_degrees says so. Review item on #2096.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and wrote; Sonnet 5.5 measured in Figma
The eyedropper rendered its 1x1 sample with default render options, without the canvas's
magnified-image pixels, so on an image shown past 2x it took a bicubic blend of a 1 px line and
its neighbours: the reporter's #3D444D border beside a darker background came out darker. The
sample also began at the pointer instead of being centred on it, half a screen pixel off. It now
renders as the canvas does (magnified_image_pixels) and centres the pixel on the pointer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 found the cause from the reporter's values and wrote; Sonnet 5.5 investigated the flat cases at medium
Review follow-ups, tests only.
- The Pencil test and the `vector.simplify` test measure how far the result
strays from the samples (at most 1.5), not only its box.
- The Pencil test also draws with the Marker and the Highlighter, each stroke
in a fresh session, and takes the layer that stroke made.
- The `vector.simplify` test runs on an open stroke, a filled shape, a region
with its own fill, and a stroke beside a filled shape.
- The geom tests run at accuracies 0.1, 0.75 and 8.0, on quadratic and cubic
inputs as well as lines, and check both ways round how far the result and
the input are apart.
- A path that turns straight back is checked to be kurbo's fit of each side on
its own, with a point at the turn.
- Empty paths, lone moves and segments after a close are covered.
A junction, which any account can make, leads out of the root; the path
through it and back with `..` is the file in the root, to the check and to
Windows when it opens it.
Review of the fix for #1094:
- A path that can't be made absolute is refused instead of going on to be
resolved.
- A name on the way that can't be looked at for a reason other than nothing
being there (no permission, for one) is refused: it may be a link that
resolving would follow. Only "not found", "not a directory" and "invalid
filename" count as nothing there.
- Tests: the link part of the Windows test is its own test, so the spelling
and device checks no longer share a test with a part that is skipped where
links can't be made; the timing assertion is gone (the messages tell the
cases apart); link chains of 39, 40 and 41 links; a root on a share against
every network spelling, in both of the root's spellings; devices for read
and write; `..` after a link; paths that aren't UTF-8.
- Docs brought in line with the code.
Tests for the -129 leftward boundary, the vertical boundaries both ways, a drag with no move along the flow, and several moving layers using their combined box. Doc comments say whose edge is "leading" and what the caller passes; the parity row names the unchanged cases and the web-app measurement source.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 triaged and briefed; Sonnet 5.5 verified and wrote at medium
Tests only. Ctrl+wheel joins Cmd+wheel; panning (plain wheel, Cmd/Ctrl+wheel and Shift+wheel with Scroll wheel zoom on) is compared between invert on and off; the pinch runs with Scroll wheel zoom on and off and with factors above and below 1; the pref's off value is saved and reloaded.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 triaged and briefed; Sonnet 5.5 verified and wrote at medium
tool.annotation validates a named id or ids before changing the tool: a page or unknown layer is an error, and a named layer is always selected. Choosing the tool with tool.set (the toolbar) asks for the selected layer's note like Shift+T. annotation.new only sets the focus request for a selected, non-page layer, so it cannot wait and take the keyboard later. ToolId::works_in_dev_mode is the single list behind the pointer routing, the command allowlist and the toolbar. The press never picks the page. Tests for the invalid targets, the toolbar path, a drag, the re-focus, and a request for an unselected layer.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 triaged and briefed; Sonnet 5.5 verified and wrote at medium
Resizing a board table by its bottom edge, a corner or a height command left it at a fixed height, so its rows stayed at 40 with blank space below them (or were clipped when shrunk) and typing in a cell no longer grew the table. Measured in FigJam's web app (2026-11-09): a table has no lock-proportions control and never scales proportionally; a bottom edge grows only the last row, a corner drag shares the change equally over every row and column, text keeps its size, and a cell has a minimum size.
board_table::settle_resize runs first in the normalize pass (so a drag, node.set, transform.resize, MCP and live-session peers all agree): a table left at a fixed height gets the change in its rows' minimum height (the last row, or every row when both axes changed), is kept at least columns x 40 wide, and hugs its rows again. New cells get a minimum width of 40. FigJam's floor is about 52 and FigCraft's is 40; there are no per-column widths yet.
Docs: boards-parity Tables row, revision history and status line; the board_table module doc.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and briefed; Fable 5.1 (work seat) scoped; Sonnet 5.5 measured in FigJam and wrote at medium
The growth test scans the system fonts, as the other text-measuring engine tests do (#333), and
asserts the note grows taller unconditionally instead of only when text happened to measure.
docs/boards-parity.md no longer lists a sticky's text size among what the normalize pass derives.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 checked the findings and wrote
FigJam's sticky text is 18 (Small) and never changes as the note fills; the note keeps its width and grows taller (measured in the web app, 2026-11-09). A new sticky's text is STICKY_TEXT_SIZE (18) and the normalize pass no longer resets a sticky's text size by its length, so saved stickies keep the size they have. No size menu yet.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and briefed; Sonnet 5.5 measured in Figma and wrote at medium
On Windows, Roots::check resolved every path (canonicalize opens it) before
comparing it with the roots, so `\server\share\x` contacted the server, and
could send it the user's credentials, before it was refused. Now the path as
given and made absolute is classified by its spelling with craft_docpath
first: another computer that no root's share is on, a device path, or a DOS
device name anywhere in it is refused, and so is a link on the way that
leads to one (followed by its target's spelling, up to 40 links).
Fixes#1094
Adds the invert_zoom preference and the pref.invertZoom command, a checkbox in the Preferences window after Scroll wheel zoom. With it on, Cmd/Ctrl+wheel and a plain mouse wheel (with Scroll wheel zoom on) zoom the other way by the same factor; a pinch and panning are unchanged. Documented in docs/ui-parity.md.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and briefed; Sonnet 5.5 measured and wrote at medium
Adds ToolId::Annotation and tool.annotation {id?}. With a layer selected (or an id) it asks the UI for that layer's note at once; in the tool a press on a layer selects it and asks, a press on empty canvas does nothing, and the engine never writes the note. annotation.new shows the Inspect tab (not a toggle) and focuses the layer's new-note field once. The tool passes Developer Mode's gates and stays selected until V or Escape (inferred; Figma's placement, cursor and stay-in-tool behaviour are unmeasured). Docs, shortcut counts (137 documented, 136 bound) and tests updated.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove; Fable 5.1 briefed; Sonnet 5.5 wrote at medium
Within its own row or column, a dragged child's slot follows the box it would fill under the pointer (start bounds plus the unsnapped drag), by the edge facing the drag, not the pointer. Measured in Figma's web app, 2026-11-09. Dragging in from outside, wrapping layouts and grids keep their rules. Several moving layers use their combined box.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and briefed; Sonnet 5.5 measured and wrote at medium
docs/ui-parity.md conflicted on the wrap style and text direction rows: kept this branch's text with #2135's Developer Mode naming.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and wrote
Figma holds a US registration for DEV MODE (US 7574348). Display text only: the view.devMode label, Inspect tab, shortcut row, refusal messages, CLI help, docs and all locale catalogs now say Developer Mode. Command ids, preference keys and saved fields are unchanged; "Dev Mode" stays a search alias in the Actions palette. Adds a naming guard test.
Drops the "Figma-class" self-description from the MCP instructions and the "Accessibility at Figma" aside from pref.uiScaleUp help. Help text only; ids and preference keys unchanged. Adds a display-name guard over mcp/src and engine/src/cmd string literals.
kurbo's simplify_bezpath, which smooths Pencil samples and runs Simplify
path, fitted two common shapes with a curve that ran far off the path: a
stroke that goes out and back to the very point it left (kurbo keeps a
reversal as a smooth join) and nearly collinear samples (kurbo measures
its error across the path, not along it).
figcraft_geom::simplify_path cuts the path where it turns back by more
than a right angle, simplifies each run with kurbo, and keeps a run as it
was when its fit strays more than twice the accuracy from it. The Pencil
and vector.simplify use it.
One Escape selects nothing for a top-level frame, a top-level layer, two layers in one group,
layers in different parents, and a nested layer with a top-level one. With the focused frame
itself selected, one Escape leaves the focus view.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and wrote
The handler grouped equal offsets by a linear scan per point, quadratic when every point moves
by its own offset. Only a single move covering every point carries the regions' paint frames,
so it now checks in one pass whether all offsets are equal: if so the points move as one,
otherwise each moves by its own offset. The result is the same as before.
Tests: booleans, a valid x with a refused y, and the overflow case are refused with no undo
step; nothing selected is refused; a shared Y scrubs every point in one undo step and a Mixed
X doesn't scrub. selection_position's doc comment names its one user, vector.edit.bounds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and wrote
Found while testing crates/engine/src/cmd/file/open.rs:33 (open_recent, `.nth(i.max(0.0) as usize)`)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Setting each point's own coordinate moved the points one at a time, so when every point of a layer moved by the same offset (`{"x": {"add": 10}}` on all of them) `ops::move_vertices` no longer saw the whole network move and left the regions' paint frames behind. Points with the same offset now move together.
Verified: `moving_every_point_by_one_offset_carries_the_regions_paint_frame` failed before this and passes; vectoredit tests 92 passed, junk sweep 49 passed, clippy and fmt clean.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and wrote
Cause: with several points selected, the Point section's X and Y showed the selection's top-left and `vector.edit.position` moved the group so that its top-left landed on the typed value. Figma shows the value the selected points share (Mixed when they differ) and sets every selected point's own coordinate.
Fix: `vector.edit.position {x?, y?}` takes, per axis, a number (every selected point's own coordinate becomes it, in the parent's coordinates) or `{add}`, `{mul}`, `{div}` applied to each point's own coordinate; each point moves by its own delta with its handles, in one undo step; a zero divisor, a non-finite result, unknown keys and other junk are refused. A new `selected_parent_points` gives the panel the points in the parent's coordinates. The panel's X and Y are mixed-value fields: the shared value (compared within 1e-6) or Mixed, a typed value sets every point, and a relative edit typed into a Mixed field sends the object form. A scrub only starts from a shared value and sets it absolutely each frame inside one interaction, as the other point fields do.
Verified: engine tests for two points at different x, one point, add, mul and div, curve handles, one undo, the parent's coordinates and refused junk; panel tests for Mixed, typing -200 and a relative edit on a Mixed field, and a single point's value; the engine lib tests, the vector and field panel tests, clippy and fmt. Measured in Figma on 2026-11-08.
Not changed: the W and H row, other edited layers' points, and scrubbing semantics (the Mixed field cannot be scrubbed, as no start value exists).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove, measured and briefed; Sonnet 5.5 wrote at medium
Found while testing crates/doc/src/props.rs:1394 (ExportSize::parse, `num.parse::<f64>()`)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Found while testing crates/engine/src/cmd/edit/props.rs:248 (reparent, `index = num(p, "index").map(|i| i.max(0.0) as usize)`)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Cut with nothing to copy (only points that touch no segment) is Disabled and changes nothing, as Copy now is (Copy used to fall through to the whole layer); paste borrows the clipboard instead of cloning it; a partial copy or cut mirrors its SVG to the system clipboard; x without y (or y without x) is refused; the junk test asserts refusals change nothing and accepted results are finite; the edit.copy, edit.cut and edit.paste specs state the behaviour and the mode-specific paste response.
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
Found while testing crates/engine/src/cmd/file/export.rs:89 (export; same pattern at :332 in export_layers)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Measured in Figma: with a layer inside a group inside a frame selected (from the canvas, from the Layers panel, or after Tab), one Escape selects nothing; FigCraft selected the parent first. Escape still leaves text and vector editing, ends gestures and the pen first. Three tests updated to the measured rule (selection_behaviour, devfocus, proto_editor); the rest of the suites pass.
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 low
With nothing selected, one arrow press panned 20 px (200 with Shift); Figma pans about 64 px and five times that with Shift (measured at 100%, about +/-3 px). New engine test pins both steps; no existing test pinned the old values.
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 low
Measured in Figma: in vector edit mode Copy takes only the selected segments (a lone point, the segments touching it) and a paste outside makes a new vector in place; Cut removes them, splitting the path in the same layer and staying in edit mode; Paste while editing adds the clipboard's segments or a copied layer's outline to the vector's network in place, selecting them. FigCraft copied the whole layer, Cut left nothing to paste, and Paste while editing did nothing. 16 new engine and UI tests cover copy, cut, paste, undo and junk parameters; the whole engine suite passes bar the known font test.
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 high
component.createSet now refuses a selection where one layer is inside another, as component.combineVariants does, before changing the document (the nested layer was silently dropped from the set). Adds the component.createSet label to the 22 locale catalogs that list component.createMultiple.
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
The third entry of the menu several selected layers get: each layer becomes a component and they are wrapped in a component set in place, 20 px around, named as Figma names them (measured): "Property 1=<layer name>" in left-to-right order, prop=value names kept, and layers that all share a name give a set of that name with "Default" variants. Engine tests cover the three measured cases, geometry, selection, one undo and the refusal of instances; the UI test checks the three menu entries in order and that the third makes one set.
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
Measured in Figma: a layer at 7° (or -4.27°) Shift-rotated by its corner ends at 15°; FigCraft rounded the turn since the drag began, so a layer at 7° landed on 7 + k*15. The rotate gesture keeps the selection's shared rotation and Shift rounds base + turn; a selection with different rotations keeps rounding the turn. Test: a layer at 7° Shift-turned by 37° ends at 45°, by 3° at 15°, and a free turn is unchanged.
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
With two or more layers selected, the Design panel header's component button opens a menu with Create component and Create multiple components, as Figma's does; one layer keeps the plain button. Headless UI tests cover both cases (three frames: menu order, names kept, all selected, one undo; one frame: plain click), and an engine test pins createMultiple's names, positions and single undo step.
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
"You can apply a wrap style at three levels: Text layer, Paragraph, Text style". The layer level was done in #1871; this adds the other two, following the per-paragraph text direction (#1830) and the list style in text styles (#1876).
ParaList::wrap (None follows the layer) with para_wrap_style, set_wrap_style (the layer level, clears overrides), set_para_wrap_styles and common_wrap_style; restyle_breaks takes a style per paragraph; ParagraphStyle::wrap_style; text.wrapStyle and node.set wrapStyle with start/end; inspect lineWrapStyles; CSS text-wrap only when the paragraphs agree; the Wrap style control follows the paragraph; a read-only row in the style editor. The .fig mapping is left for a later change.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove; Fable 5.1 briefed; Sonnet 5.5 wrote at medium
The test now compares the stored setting with a constructed expected value (every multi-word field, print and quality at a non-default value) after export.add and after the node.get -> node.set round trip, and feeds the snake_case keys to node.set directly, so a field dropped by a write cannot pass and the PrintSettings aliases are exercised.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and briefed; Sonnet 5.5 wrote at medium
node.get reports an export setting's two-word fields in snake_case, which node.set refused, so a setting could not be read and written back unchanged. ExportSetting now accepts those names as serde aliases; the camelCase names stay the canonical form. Follow-up the maintainer suggested on #2008.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove and briefed; Sonnet 5.5 wrote at medium
The colour profile paragraph in docs/mcp.md said SVG and PDF export
Display P3 files converted to sRGB, without the RGB qualifier the export
tool text now carries. It now says SVG and RGB PDF convert to sRGB, and
PDF with `cmyk` converts to sRGB the same way and then to DeviceCMYK.
The comment above the export match in file.rs says the same.
