The MCP drawing tools ran the creating command (path.create, shape.*)
and only then paint.setFill, paint.setStroke, stroke.set,
object.setProps and effect.apply, each as a command of its own.
check_draw_options checked dash, cap, join, opacity, name and effects up
front, but not the paints: a fill or stroke the engine can't read
(stroke: "currentColor"), or a swatch the document lacks, failed after
the object was made, so the call returned an error and left the path,
its undo step and its SVG node behind. Each styling command was also an
undo step of its own.
The tools now send the creating command and the styling commands as one
command.batch (crates/mcp/src/tools.rs, draw): the engine rolls the
whole batch back when a step fails, so a refused paint leaves no object,
no undo step and nothing in the SVG, and a good call is one undo step
named as the creating command names its own (Pen, Rectangle, …). The
styling steps act on the new object, which the creating command selects,
with item: null (its whole appearance), as the ids did before. The
reply gives the failed step's own error, without the batch's prefix.
docs/mcp.md says so.
Test: tests_drawopts::a_bad_paint_draws_nothing (invalid fill, stroke,
a missing swatch and a non-numeric strokeWidth on draw_path and
draw_shape: error, node count, undo history and SVG export unchanged;
a good call adds one undo step named Pen / Rectangle). It fails without
the change (draw_path with stroke "currentColor" drew a path) and passes
with it; the other draw option and command_batch tests still pass.
#2027 gave vectorcraft-text and vectorcraft-engine a `bundled-fonts`
feature for an embedder that supplies every document font at run
time, but vectorcraft-ui-egui (and vectorcraft-mcp) named the engine
with its default features, so an embedder that hosts the panels got
the fonts whatever it asked for. Each now has a `bundled-fonts`
feature of its own, on by default, that turns the engine's on, so a
downstream user keeps the fonts unless it asks otherwise. For a
member to leave the engine's default features out, the workspace has
to declare the engine without them, as it already declares
vectorcraft-text; the crates that used the engine's default now ask
for the fonts: the three apps and vectorcraft-testkit. Off,
vectorcraft-text's copy of the fonts (3.4 MB) is left out; theme.rs's
own copies of five of them, the panels' interface faces, stay either
way.
Nothing the workspace builds changes: for every workspace member and
for the whole workspace, with and without dev-dependencies, for all
targets, wasm32-unknown-unknown, aarch64-apple-darwin and
x86_64-pc-windows-msvc, `cargo tree -e normal[,dev] -f '{p} [{f}]'`
resolves the same features for every package as on main, apart from
the new feature names (the engine's `default` no longer listed;
ui-egui's and mcp's `bundled-fonts` and `default`).
A test of the panels can't see their configuration without the fonts
(their dev-dependencies turn them on), so `cargo xtask ci`'s
`no-bundled-fonts` step now also reads that tree for
vectorcraft-ui-egui and vectorcraft-mcp: with their default features
it must show the fonts, and without them, for any target (a wasm-only
dependency too), it must not. Documented in docs/development.md
(Fonts › Leaving out the bundled fonts).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The reference app adds a new fill with Ctrl+/ on Windows and Command+/
on macOS, and it adds a new stroke with Ctrl+Alt+/ on Windows and
Option+Command+/ on macOS. VectorCraft had Add New Fill and Add New
Stroke on the Appearance panel, with no default key for either.
appearance.addFill now has the default Cmd+/ and appearance.addStroke
has Cmd+Alt+/, which are Ctrl+/ and Ctrl+Alt+/ on Windows and Linux.
Like the panel's buttons and menu items, each key runs its command with
no params and adds the item to every selected object. Keyboard Shortcuts
lists both under Window › Appearance with their keys, and press_key
runs them for agents.
Docs: docs/mcp.md, docs/ui-parity.md, docs/target-app-parity.md.
`vectorcraft-cli run/convert --scale` and the MCP export tool's `scale`
are raster pixels per point, but they were passed to every export, and
DXF reads `scale` as drawing units per unit: a 2x PNG next to a DXF
doubled the drawing and left it unitless ($INSUNITS 0). They now go
only to raster formats. DXF's own scale stays reachable through the
format options.
The MCP server's reader queues at most 64 lines for the engine and
stops reading past that until the engine catches up. The reply to a
line that is not JSON, or that is longer than the line limit, skipped
that check: while a call had run for less than the busy limit (two
seconds by default), or while the engine was behind, every such line
added a reply to the queue. In the new test, 192 lines that are not
JSON leave 192 replies queued on main.
These replies now wait for room as requests do, and still go out at
once when the running call has passed the busy limit.
The EPS reader read PostScript by any name, but the Open and Place
dialogs hid .ps and .epsf files, and a .ps or .epsf file whose header
is only %! could not be opened or placed ("not a format VectorCraft
reads"). The reference app lists .eps, .epsf and .ps under Open and
Place.
- A read-only PostScript format (ps) joins the format table. A .ps
file opens and places through the EPS reader and reports the
ps format. document.export does not write .ps; file.print does.
- .epsf is an EPS extension.
- OPEN_EXTS and PLACE_EXTS list epsf and ps, so the dialogs, the CLI
help and the MCP open_file description list them.
UNASSOCIATED_EXTS lists them too: the macOS bundle declares neither.
- A .ps or .epsf file whose header is only %! now pastes as EPS when
copied in a file manager, as one with a %!PS header already did, and
clipboard.importEps reads PostScript whose header is only %!, which
it rejected before.
Docs: docs/mcp.md, docs/file-format-parity.md (.ps row),
docs/target-app-parity.md.
file.saveAsTemplate kept a name whose extension named another format
and wrote the template into it: {path: "card.png"} wrote .vctemplate
JSON to card.png, and card.pdf typed in the desktop save panel got it
too. file.saveAs {format: "ait"} answered that document.export writes
other formats, but document.export does not write .ait either.
file.saveAsTemplate now replaces another format's extension (card.png
is saved as card.vctemplate, or card.ait for format: "ait") and adds
the extension to a name whose extension names no format. The desktop
panel adds the template's extension to any name that does not end in
a template extension (card.pdf.vctemplate), as Save As adds its own.
Save and Save As still refuse ait, and the error now names
file.saveAsTemplate.
Docs: docs/mcp.md, docs/target-app-parity.md.
The MCP server read one line, ran it and answered it before reading the next,
so notifications/cancelled was never seen during a call and every request
(even ping) waited behind a running export.
- vectorcraft_doc::cancel: a thread-local cancel token and progress sink that
long work polls. Checkpoints: PDF pages, PDF flattening, geometry bake,
raster-effect flattening, drawing objects, raster artboards, and before any
file is written. A cancelled command returns EngineError::Cancelled and
changes nothing.
- Server::serve reads on its own thread: notifications/cancelled drops a
queued request or stops the running one (no reply, as the MCP spec says),
engine-free requests (ping, the list methods, prompts/get) are answered at
once, a request that would wait behind a call running for 2 s is refused
with -32003 naming that call, and _meta.progressToken gets
notifications/progress (page n of m). Queue and output are bounded.
- The remote backend stops waiting for the app when the call is cancelled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A Selection tool marquee selected every object it touched, a Group
Selection marquee selected the anchors inside it, and E pressed during
the drag did nothing. Now E switches either marquee to selecting only
the objects that lie entirely inside it, measured by their geometric
bounds, and E pressed again switches it back. Each marquee starts in
the default mode.
- The Selection tool selects a group only when the whole group lies
inside. Group Selection selects the objects inside groups, each as a
whole object. Shift still toggles. Direct Selection does not take E.
- The ruler guides the marquee selects with the art are limited the same
way: an artboard guide is selected when its whole length lies inside,
and a canvas guide is not selected.
- Tools receive letter keys as ToolKey::Letter. While a tool is busy, the
desktop app sends it each press of a letter it claims, without Cmd,
Ctrl or Alt and without key repeats, and headless press_key sends a
claimed letter the same way. With nothing dragged, press_key {key: "e"}
still chooses the tool bound to E.
Docs: docs/mcp.md, docs/target-app-parity.md.
The reference app resets the horizontal scale of type to 100% with
Cmd/Ctrl+Shift+X. VectorCraft had no command for it, and the chord
cut the selection: egui reports it as a Cut with Shift held.
type.resetHorizontalScale (Reset Horizontal Scale, default
Cmd+Shift+X) sets the horizontal scale to 100% in the text the Type
tool has selected, or else in all the text of every selected type
object, in one undo step. The value is the one the Character panel
shows, so a type object stretched to twice its width keeps 50% stored.
The vertical scale does not change. At a caret the command leaves the
text as it is. Type objects given in ids are reset in full, as
type.resetTracking does.
Docs: docs/mcp.md, docs/target-app-parity.md.
No command held Cmd/Ctrl+Alt+Shift+. and Cmd/Ctrl+Alt+Shift+,, so the
chords reached Increase and Decrease Font Size (Cmd/Ctrl+Shift+. and ,)
and changed the size by one Size/Leading increment.
Two commands, type.size.increaseByFive and type.size.decreaseByFive,
now hold the chords. They step the font size of the selected type, or
of the text the Type tool has selected, by five times the Size/Leading
increment (Preferences › Type). Agents reach them with press_key too.
Their labels have rows in every complete catalog.
Docs: docs/mcp.md, docs/target-app-parity.md.
The command catalogue is about 340,000 characters, more than many MCP
clients show, and `query` (the name agents reach for first) was refused as
an unknown argument. command_list and list_commands now take `query` as
another name for `filter`, `menu` to keep the commands under a menu path
("Object > Path"), and `brief` to leave out each command's params text,
which is most of the catalogue; the params of one command are a filtered
listing away.
A step whose id is draw_path or draw_shape runs that tool with its params,
checked as tools/call checks arguments; its result is the tool's, so later
steps can use "$N.id". The reply lists the ids the drawing steps made
under created. The unknown-argument check is shared with tools/call.
The drawing tools style the object they draw in the same call: dash, cap
and join go to the stroke.set that strokeWidth already sends, opacity and
name to object.setProps, and effects are applied in order as apply_effect
does. The options are checked (effect ids against effect.list) before
anything is drawn, so a bad one leaves no half-styled object behind.
The reference app resets tracking and kerning to 0 with Cmd/Ctrl+Alt+Q.
VectorCraft had no command for it.
type.resetTracking (Reset Tracking and Kerning, default Cmd+Alt+Q) sets
tracking to 0 and kerning to Auto, which the Character panel shows as 0,
in the text the Type tool has selected, or else in all the text of
every selected type object. At a caret it resets only the kerning of
the character before it, where Alt+Left/Right kerns. It is one undo
step. The range logic of type.step moves into a shared type_ranges.
When a call gives ids, type.step and type.resetTracking change all the
text of those type objects even while the Type tool edits text, as
paint.setFill does. Before, type.step changed the Type tool's selected
text instead.
Docs: docs/mcp.md, docs/target-app-parity.md.
document.export, document.serialize and document.exportSelection write
a PDF-compatible .ai (format: "ai"), and document.formats lists ai as
writable, but none of their params texts named it. The params text of
document.exportSelection named only png, jpg, webp, svg, svgz and pdf,
although the command writes all 20 writable formats. The description
of the MCP export tool and its row in docs/mcp.md named every format
of the tool's format enum except ai, and the range text and the row
said that only PDF writes several artboards.
The three params texts now list every format their command writes,
and document.export's range and useArtboards texts name .ai beside
PDF. The MCP export tool's description names ai, and its artboard and
range texts say that .ai, like PDF, writes one page per artboard. The
range text and the row also say that SVG, SVGZ and EPS write one file
per chosen artboard, and that rasters, DXF, EMF and WMF do so with
options.useArtboards: true.
Docs: docs/mcp.md (the export tool).
After the app restarted, the remote backend wrote the next request to
the old, closed connection and reported that the operation may have
completed, although the restarted app received nothing. Reused
connections are now checked without blocking before writing, and a
closed one is replaced first. If the app can't be reached, the error
says so and that nothing was sent. Failures after the request was
written still report uncertainty and are never replayed.
Loopback tests cover a restart between calls (the edit is applied
exactly once) and a refused reconnect. The lost-reply test from #1246
still passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BSB7fMQUccqFQekksDGWpo
The server dispatched requests and notifications without ever checking jsonrpc, so a missing or wrong version could invoke tools or flip notification state; Remote accepted any reply with ok/result/error without comparing ids, so a faulty endpoint could attribute a stray result to the current call. handle() now requires jsonrpc 2.0 before dispatch (other envelopes get -32600, or are ignored for notifications, which have no id for a reply) and Remote::call rejects a reply whose id doesn't echo the request's, dropping the connection so the next call starts fresh, with the usual uncertainty error. Three new tests cover wrong/missing versions (incl. batch entries and a state-changing notification) and the mismatch-reconnect; docs/mcp.md documents both. cargo test -p vectorcraft-mcp passes.
The server read stdio lines and remote replies without any byte bound, answered batches of any size, and command_batch kept every step result: one faulty peer or batch could allocate and work without bound. Request lines are now read in bounded chunks (16 MiB; longer lines get a -32600 reply, are skipped, and framing continues), protocol batches hold at most 64 messages, control-channel replies past 256 MiB drop the connection with the usual uncertainty error, command_batch takes at most 256 steps and keeps at most 8 MiB of step results (past that the batch stops with the reason, still closing its atomic undo group). Eight new tests cover the line/batch/step/result caps and the atomic-group close; docs/mcp.md and the command_batch tool text document them. cargo test -p vectorcraft-mcp passes.
text.create, text.createInPath and the MCP add_text tool now take a
`justify` (auto|left|center|right|justifyAll, as text.setStyle). Point
type already laid out centred or right-aligned text around its origin;
it just could not be made that way, so scripts measured the bounds and
moved the text. Changing it afterwards stays text.setStyle {justify}.
text.create, text.createInPath, text.setText and text.setStyle answer with
overflow: [id…] for the area type and type on a path whose text doesn't fit (left
out when all fits), and MCP add_text with overflow: true. add_text takes area
type's fit and fitMinPercent.
Drop a failed control connection and reconnect for the next request, never resend an operation that may already have completed.
A loopback regression applies a rectangle then drops its reply. It reproduces two edits before the fix and verifies one edit, an error, and successful inspection afterward.
Agents could set and clear the key object but not read it:
document.inspect, the control channel's ui.pointer reply and the
headless pointer_gesture reply left it out, and select.key returned
null. select.key with an id that was not selected, or with a value
that is not an id, such as "x", cleared the key without an error. A
compound-path member or an object inside a selected group could become
the key.
select.key now checks its id as select.set, select.add and
select.toggle do (#784). It returns an error for an id that is not a
selected top-level object and leaves the key as it was. Layers,
compound-path members and objects inside another selected object are
not top-level objects. select.key {id: null} used to clear the key and
is now an error; select.key {} still clears it. select.key returns
{key}, and document.inspect and both pointer replies include the key
object as "key".
In isolation mode a Selection-tool click can pick an object that
select.key does not accept as the key: a sublayer inside an isolated
layer, a member inside an isolated compound path, or an object in an
isolated group that is also selected. The tool no longer sends
select.key {id} for such an object. When the object is already the
key, the tool still sends select.key {}, which clears the key.
A recorded action whose select.key step names an object that is not a
selected top-level object when the action plays now fails and is
rolled back, as an action with a select.set step naming a deleted
object has failed since #784.
Docs: docs/mcp.md, docs/control-protocol.md, docs/target-app-parity.md.
An agent's build over MCP left one undo step per batch step (about 280
for one job). `command_batch` takes `atomic: true` (and an optional
`label` naming the step) to make the whole batch one undo step. It runs
the batch in an undo group, opened and closed by the new commands
`edit.undoGroup.begin {label?}` and `edit.undoGroup.end {cancel?}`, so
every backend (headless or the connected app) supports it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
DXF already outlines type unless preserve: editability, so say so next to
the outlineText flag; reword the new-art sentence and point to paint.default.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Omitted paint fields keep the current appearance, so agents can get a black outline when they only set a fill. Document the initial white fill and 1 pt black stroke and recommend explicit paint for outline-free art.
Clarify SVG/SVGZ outlineText versus PDF options.advanced.outlineText and PDF's live-text default, noting DXF also accepts the top-level flag. Descriptions and documentation only; drawing and export behavior are unchanged.
render_preview only rendered the first artboard, so an agent couldn't
look at pages 2 and on of a multi-artboard document without switching
to screenshot. It now takes an optional 0-based `artboard` (as
screenshot does) and refuses an index the document doesn't have.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
object.align with neither horizontal nor vertical returned success
and moved nothing. The command palette, and a key bound to it in
Keyboard Shortcuts, run it with no params. It now returns
"give horizontal or vertical", as object.distribute already did.
An action recorded while Align ran from the palette or a bound key
contains an object.align step with no params. Playing that action now
fails at that step, and command.batch rolls back the whole action.
Values outside a parameter's choices were read as another choice or
ignored. Align ignored an unknown horizontal or vertical and aligned
to the selection for an unknown to value. Distribute read an unknown
direction as center. Distribute Spacing and Shear read any axis other
than "vertical" as horizontal, and Reflect read any axis other than
"horizontal" or a number as vertical. A bounds or to value that was
not a string fell back to the default, and a spacing that was not a
number gave automatic spacing. The commands now return an error that
names such a value, before they move anything or record an undo
step. When a to, bounds, axis or spacing value is missing or null,
the command uses the parameter's default: Distribute Spacing without
an axis spaces the objects horizontally.
The run_command description and docs/mcp.md showed object.align with
an align param it does not have; they now use horizontal. The new
status-bar messages have rows in the de, es, fr, it, ja, ru and uk
catalogs.
Docs: docs/mcp.md, docs/target-app-parity.md.
add_text passed `color` to text.create only, and text.createInPath built its own character
style without a colour, so type on a path and path-based area type took the current fill.
text.createInPath now uses the same new-type style as text.create (size, font, style, color,
else the current fill), and add_text sends `color` on both routes.
The MCP test draws a blue rectangle (the current fill), turns it into on-path and area type
with color #ff0000, and reads red back through inspect_document.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts:
# ROADMAP.md
A graph's data can now hold blank cells: an empty CSV cell, a null in
`rows` (any cell that isn't a finite number), the rest of a short row,
or a row that has only a label.
- Columns, bars and stacked segments leave a blank cell out instead of
drawing a zero-height mark.
- Line and scatter graphs break their lines around it and draw no
marker there; the value axis ignores it.
- Area, pie and radar graphs draw it as 0, which is what it adds to
them.
The help pages don't describe blank cells, so this is our own choice:
a cell with no value has nothing to draw, which also matches how a
series with a missing year reads in a report.
In the file a blank cell keeps a 0 in `rows` and is listed in a new
`blanks` field ([category, series] pairs, written only when there are
any). A version that doesn't know blanks ignores the field and reads
the zeros, so files still open there, and files without blanks are
unchanged.
A number in straight quotes is a label, as the help page on graph
labels says (a year as a category name, for one). The CSV reader used
to strip the quotes first and read it as a value. Label columns are now
decided for the whole column: once any row starts with a label, every
row's first cell is its label, so a row with an empty label no longer
shifts its values one place left. The data window writes a header row,
labels that are empty, read as numbers or have spaces around them in
quotes, and blanks as empty cells, so they come back as they were.
graph.setData {} returns blanks in `rows` as null, so the rows it
returns can be edited and sent back. A label-only row keeps its place
with blank values; a row with neither a value nor a label is dropped,
and a first row that is a label followed by empty cells (`Year,,`) is
still read as a header with no series names.
The data grid is built in full for drawing and for the data window, so
it now stops at 10,000 categories of 256 series, which keeps a file
with one very long row from allocating without bound.
Behaviour (quoted labels): Add graph labels and data sets:
https://helpx.adobe.com/illustrator/desktop/automate-visualize-data/visualize-data/add-graph-labels-and-data-sets.html
vectorcraft-cli mcp and vectorcraft --control take the flags PhotoCraft
takes. While roots are in force, every path a command reads (open, place,
relink, SVG image links, library and profile loads, plug-ins, preset
imports, font searches) must lie inside the read root and every path it
writes (save, export, package, Export for Screens and for Web, print to
file, captures, library saves) inside the write root. Paths are made
absolute, links are followed (a missing file's nearest folder's), and the
result is compared by whole components, ignoring case on Windows; `..`
below a missing folder, dangling links and device names are refused. A
root left out grants none of its access; without the flags nothing
changes.
The checks live in one place, vectorcraft_engine::file_access, consulted
by the engine's file helpers and the app's file services. The headless
server confines its whole process; the app confines control requests,
the frames carrying input they inject and the background saves they
start, leaving the person at the keyboard and the app's own folders
alone. The screenshot tool now has the backend write its file.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A run (vectorcraft-cli run, the MCP command_batch tool) had no way to use
an id an earlier step returned. A parameter string that is exactly
$N.path or $last.path now takes that value from step N's result
(0-based), e.g. {"id":"$1.id"}; $$ starts a literal $, and a
reference to nothing fails the step. Objects given to a command are now
used as given: an ids element or id that isn't an object id (or an
unresolved reference) fails instead of the command acting on the
selection. text.setRangeStyle styles the selected type object when no
id is given, as text.setStyle does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A fourth Image Trace mode for flat-colour logos. The existing tracer follows hard pixel
edges, so an outline can only be as good as the pixel grid; here each pixel is read as a
blend of the logo's flat colours, the blend is interpolated smoothly (cubic B-spline), and
every outline is placed where two colours are equally present, to a fraction of a pixel.
Outlines are then cleaned the way a person would:
- corners the blur rounded are put back (line fits either side of a bend, intersected);
- near-straight runs and near-circular arcs become exact lines and arcs (Snap Curves To Lines);
- the rest is fitted with as few cubic Beziers as the tolerance allows (Schneider);
- touching colours overlap by 0.06 px so no hairline gap opens between independent fits;
- shapes are painted biggest first; transparent areas fully enclosed by artwork (letters
knocked out of a badge, the counters of an "o") are painted white so the logo reads the
same on any background, or become real holes with Ignore White (each hole belongs to the
smallest outline containing it);
- an outline whose fit strays more than 1.4 px from the raw contour is fitted plainly.
Colours are found automatically (and completed from pixels no blend of the known colours
explains, so a small accent is not lost) or named with `logoColors: ["#rrggbb", ...]` (white is
always its own class; colours closer than 12 count as one; at most ten). Gradients and photos
are refused with advice to use Color (a flat-colour score of the solid areas, 0.93). A
hard-edged source with no anti-aliasing is softened by under a pixel and fitted with looser
snap tolerances. Images over 1024 px on the long side are averaged down first (further, if
they are busy enough to make the unmixing expensive) and the result is mapped back exactly.
More real outlines than we will fit are refused, not silently truncated.
Also: a "Flat Logo" preset, the mode in the Image Trace panel (Method hidden: it always
stacks), the command docs and MCP prompt, es/fr/it/ru translations for the two new messages
and the mode name, and the ROADMAP.
Never-crash: no unwrap/expect/panic, every size and float is bounded and checked, with
tests for degenerate and extreme input (NaN/infinite parameters, 0x0 and 1x1 images, sparse
outlines, an anchor budget, a 200,000-entry palette) and a 250-run randomised no-panic
property test. The B-spline interpolation follows Unser et al. and Thevenaz et al., and the
curve fitter follows Schneider (Graphics Gems); both are written from the papers.
Measured (local experiment, not in CI): a flat-colour logo's vector original rendered down
to a soft 199x100 px PNG and traced. Mean edge error vs the original at 1600 px wide:
built-in colour presets 2.5 to 3.4 px, Flat Logo 0.78 px (0.48 px from a crisp source);
region overlap 0.88-0.91 vs 0.966. About 0.1 s for a 200 px logo.
Tests: 60 new in the trace crate (unmix, field, contours, geometry, Beziers, corners,
outline fitting, palette, end-to-end on synthetic shapes made in code) and 3 in the engine.
cargo xtask ci passes (fmt, clippy, test, assets, brands, layers, wasm): 4,290 tests.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
dotCooCoo