* claim: #1486
* Compare picture: overlap and edge offsets of the model against a reference picture (#1486)
inspect.compare_picture lays the model's silhouette over the part found in a
PNG or JPEG and slides it to the best overlap: overlap (0..1), the model's
edges against the picture's in mm at nine stations, and an overlay PNG. The
MCP tool compare_picture runs it and hands the overlay back as an image block.
inspect.silhouette grows a reusable silhouette::outline for the model side.
* chore: re-trigger checks (defaults line already present in PR body)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* i18n: carry the Compare Picture label in Arabic, Hebrew and Vietnamese too
The complete-catalog guard (every_complete_catalog_carries_every_english_message)
failed because the pending file covered only en/ja/zh; ar, he and vi are
complete catalogs and need the key as well.
* Compare picture: one uniform scale, capped fit, sturdier part finding
Review on #6416: fit no longer stretches the axes apart (width is the default,
both scales uniformly by width and reports aspect_mismatch); the fitted model
takes the same MAX_GRID cap as the picture; the border flood seeds only from
border pixels near the border's main colour, reads alpha, retries at a
stricter step, and says which case failed; the slide searches the whole
placement range instead of a tenth around the centre; the model's holes get
the picture's fill rule; crop becomes crop_px / crop_fraction with explicit
units. Pins #1486's bracket (overlap 0.865, bottom -10 mm at the upright) and
adds a regression test per problem.
* Compare picture: alpha only for a see-through border, part_fills_crop, outline-only wording
Review re-test on #6416: one translucent backdrop pixel no longer switches the read to
alpha — alpha decides only when most of the border is see-through, and any ink is then
part, so a part drawn at alpha 100 survives; part_fills_crop: true takes the crop box as
the part itself, for one view cropped tight off a board where the part takes most of the
border and the flood would keep the backdrop strip instead; a third backdrop pass compares
each pixel to the border's main colour, so a white part on (245, 245, 245) JPEG at quality
75 separates; the command, tool and docs descriptions now say the compare is outline only:
holes are not compared. A banded light backdrop stays a follow-up. A test per case.
* MCP compare: format the picture tests
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Dahak Builds <dahakbuilds@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
* MCP: batch and execute take expect {volume, bodies, bbox} and flag a mismatch (refs #3380)
An optional `expect` argument on the `batch` and `execute` tools: the volume
(with a relative `tol`, default 0.002), body count and bounding box (0.05 mm)
the agent computed before building. Once the call has gone through the server
measures the design as the `measure` tool does and the reply gains
`expectation: {ok, measured, expected, mismatch}`. A mismatch flags the tool
result `isError` with the mismatch text first; the design is kept as built.
A malformed `expect` is refused before anything runs. Without `expect`
nothing changes, and a batch step carrying one is a step with a stray key.
In 29 logged agent sessions every agent computed these numbers beforehand
and then spent a `measure` call plus a comparison by hand after every stage;
one missed a 0.25 mm skin until it compared. Volume and body count are
kernel-independent checks.
Refs #3380
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* MCP expect: filter then map for the bbox coordinates that are off (clippy)
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* MCP: expect's coordinate text helper no longer shares a name with the JSON reader
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: bherbruck <tenbee91@gmail.com>
The `initialize` instructions were one paragraph of command shapes. Replace
them with version-neutral guidance: read `skill_reference solidcraft-modeling`
first, plan the numbers, build in a few `batch` calls, the primitive routes
and the fillet-shell-inside-holes order, what follows a parameter, `measure`
after each stage with a coalesced parameter check, and `timeline.edit` /
`parameters.change` / `edit.checkpoint` to fix instead of rebuilding. The
Assistant's system note gets the same pointer; docs/mcp.md says so.
`instructions_name_only_real_tools` now checks the command ids as well as the
tools, and `instructions_recipe_builds_a_parametric_shelled_box` builds a part
the way the instructions describe and checks its volume, axes and parameters.
Refs #3380
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
* MCP: measure summary and body_topology faces: cylinders for small verification replies (refs #3380)
`measure {"summary": true}` replies one short object per body (name,
volume_mm3, bbox, faces, closed) with body_count and total.volume_mm3,
as a tool and as a batch step. `body_topology {"faces": "cylinders"}`
replies only the cylindrical faces, the kernel's split halves merged
by radius and axis line into {radius, origin, axis, extent, names},
and no edges. Both defaults are unchanged. The merge uses only what
model.faces and model.edges already reply.
Measured on a 13-feature enclosure, body_topology is 74 KB and measure
1.1 KB per body; agents filter them with scripts, and seven of 29
benchmark runs asked for exactly a cylinders-only listing.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* MCP: keep body_topology's cylinder merge next to check_parameters' without a name clash
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: bherbruck <tenbee91@gmail.com>
A new tool that checkpoints the design, sets every value in one
parameters.change, reports measure, errors and cylinder axes, and
restores the checkpoint, so the loop set_parameter / measure /
body_topology / set back costs one call and one rebuild.
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
* claim: #9184
* Control channel: a request's own param_check level, applied and confirmed by the app (#9184)
`engine.execute`, `engine.batch` and `engine.script` take an optional
`param_check: "off"|"warn"|"strict"` in their params. The app runs the
request at that level through the same `run_checked` path and batch
preflight a headless session uses, puts its own level back, and every
reply says the level it ran at as `param_check` beside `ok`. A command
that switches designs keeps the app's level; any other word is refused.
Readings and exports answered on a copy off the window thread keep the
level too.
`control.capabilities`, answered at once on the control-server thread,
lists what the channel does beyond the methods every version has:
`{"param_check": ["off", "warn", "strict"]}`. An older app answers
"unknown method".
The MCP bridge asks it once per connection. When the app lists the
level, every call and batch goes with it and the app checks against its
own commands; otherwise the bridge checks before sending, as #9073 does.
At strict, a reply that doesn't say the app checked the call is an error,
and the next calls are checked by the bridge. Also: `add_param_warnings`
takes the reply by `&mut`, and `with_param_check` sits beside
`Remote::new`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Control channel: a design-switching command is checked at the request's level
Review fixes for #9184. `component.edit_in_place`, `component.finish_edit`,
`view.next_tab` and `view.previous_tab` ran at the app's own level and
echoed `off`, so a misspelt call went through at strict. They now have
their parameters checked at the request's level before they run (and
still run at the app's own level, as each design keeps its own), and the
reply echoes the requested level. The MCP bridge at strict now requires
the echo to match its level, not just to be there. `param_check` is read
only by `engine.execute`, `engine.batch` and `engine.script`; other
methods ignore it. The off-thread test now shows the design change
computing off the window thread before it is refused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #9072
* MCP: --param-check checks calls to the running app (#9072)
`mcp --connect` (and `mcp` once it bridges to an app) took
`--param-check` and forwarded every call unchecked, while the app's own
session is always off. `Remote` now checks each `engine.execute` and
`engine.batch` against this server's command contracts before sending:
`strict` refuses a call, or a batch with such a step, without sending
anything, and `warn` adds `param_warnings` to the app's reply (each
step's result in a batch). A command with no contract here goes to the
app unchecked. `Auto` carries the headless session's level over when it
bridges to a late app. `batch_preflight::contracts` is the contract half
of the existing preflight, for a runner in another process.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #4071
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Define a named parameter inline from any value input (#4071)
Any value input, typed or sent as a command's expression parameter, takes
`name = expression` (or `#name = expression`): the design gets the user
parameter `name` with that expression, and the input stores `name`, in the
command's one undo step.
- doc: the lexer skips a definition's `name =`, so until the design stores
it the text evaluates as its expression and commands read and check it
like any value. `Document::define_inline` then turns every stored
definition (feature inputs and placements, dimension parameters, sketch
pattern and text values, joint and coupling values, sheet metal and
plastic rules, canvas planes) into a user parameter, the unit following
how the input reads a bare number. A taken, reserved or empty name, two
different values for one name, and a definition in a parameter's own
expression or a configuration cell are refused; all or nothing.
- engine: `Session::execute` runs it after any command whose parameters
hold definition-shaped text; the reply has `defined` (the first name,
as `sketch.dimension` already gives) and `definitions`. A refusal fails
the command, which leaves the design as it was.
- `parameters.change` with `new = expression` renames the parameter it
changes (a dimension's or feature input's own, references following)
and sets it; when the value is just `new`, it sets `new`. `changes`
refuses definitions. `expr.evaluate` reports `defines`.
- UI: lengths keep the name when the design unit is applied; sketch text
sends a definition as typed and previews only its value; the box corner
and offset Distance 2 use the value; fields whose command keeps no
expression (sketch scale, move, chamfer) refuse a definition with the
way to name it. The dimension editor says which parameter it named.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Links: update linked designs that changed while the assembly's tab was away
A design coming back to its tab compares the modification time and size of its linked
design files with what they were when it went to the background, and runs Update Links
on the worker thread for those that changed: saved in another tab, another window or by
a checkout. Preference update_changed_links (Design, Linked designs) turns it off.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Docs: the assembly guide says linked designs update when their tab returns
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Links: Vietnamese strings, now that vi is a complete catalog
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Links: remember what each design last read, so an update that fails is tried again
A design's linked files are now recorded as the design last read them: when Update Links reads a
link without an error, or when the link is first seen. Coming back to the tab reads again what
differs, one link per run, so a file that cannot be read holds up no other and is tried again
the next time, and a part changed while the assembly was shown is taken up on its next return.
Closing several tabs together reads nothing on the way, and the web build, which keeps no file
times, looks at nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Links: no read during a save-changes prompt, records for adopted designs and missing files
A tab shown to answer "save changes?" reads no link before its save, and Cancel there or on a
batch lets the tab left in front read what it skipped. A design opened in place of the blank one
records its links at once, a link whose file is missing is recorded as missing so it is read when
the file comes back, and a read queued for a design that is no longer shown by the time it starts
is dropped. More changed links than can wait at once are read on the next return.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Assemblies: edit a linked component in place, with the assembly around it
Activating a linked component opens the design it comes from in a tab of its own, with the
rest of the assembly drawn faded around it where the component sits, so the part's own history,
browser and commands are at hand. Finish Editing in the browser saves the part, goes back to the
assembly and closes the part's tab when the edit opened it; coming back to the assembly updates
the changed link (or Finish Editing runs Update Links when that is turned off). Update Links in
the part also reads the assembly's links again, off the window thread, and redraws the assembly
around the part; a result for an assembly changed meanwhile is dropped with a message.
The part is an ordinary tab, so saving, autosave, recovery, closing and quitting treat it as any
other design. The model scene's ghost triangles are now drawn (only the preview's were).
Preference edit_linked_in_place turns it off. Builds on the update of changed links when an
assembly's tab returns.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: a double-click opens it, Activate only activates
A double-click on a linked component edits it in place, in the browser as on a face in the
view (around the placement clicked), and the component's menu gets Edit in Place. Activate and
the browser's radio button make it the active component as they always did, so a linked
component can still be activated without opening its design.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Double-click on a component's face: edit a linked one in place, activate any other
A double-click on a face in the view activates the component the body belongs to, at that
placement, as a double-click on its row does; a linked one is edited in place. Bodies of the
design itself or of the active component are left alone, and so is a double-click with a
modifier. One preference, double_click_components, covers both gestures.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: fixes from review
An edit whose part never opens is forgotten instead of starting on a later open of the same
file, and the web build starts the edit after its synchronous open. Finish Editing goes on only
for a save into the linked file and in the part's own tab, waits for a "changed on disk"
answer, and says so when the part was saved elsewhere. Showing the assembly no longer drops a
refresh, which now records the links it read so coming back reads nothing twice. The faded
assembly is worked out once per picture, a bad mesh index skips its whole triangle, and the
last two status texts come from the catalogs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: Finish Editing on the MODEL tab and the mouse's back button
Finish Editing is a command now, component.finish_edit, with its schemas: the app window
answers it, anywhere else nothing changes, as with view.fit. While a linked design is edited in
place the MODEL tab ends with a FINISH EDITING panel, pinned at the right as FINISH SKETCH is,
with a new left-arrow icon; the browser keeps only the line naming the assembly. The mouse's
back button, or a keyboard's Back key, finishes too unless a dialog, tool or sketch is open.
The assembly guide describes editing in place.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: fixes from the second review
Saving the assembly waits while it reads its links for the part
Finish Editing waits for dialogs, tools, sketches and prompts open in the part, also when run as a command
The assembly reads the edited part explicitly when it does not check its links on its own
The FINISH EDITING panel cannot be customized
A tab opened by an edit stays that edit's to close when the part is entered again
The mouse's back button finishes editing; the keyboard's only while no text field has focus
Refreshing the assembly keeps its selection and active component
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: fixes from the third review
component.finish_edit run as a command opens nothing and answers with an error when it cannot finish; engine.batch and engine.script refuse it
A double-click or Edit in Place while a dialog, tool or sketch is open in the assembly keeps it open
What changes in the assembly while it reads its links without changing its document stays: picks, activation, components dragged but not captured
Showing the assembly while it reads its links starts no second read
A new or opened design in the part's tab ends the edit
Finish Editing keeps the part's tab open while its agent chat runs
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: Update Links run as a command in the part refreshes the assembly too
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: the ribbon button reads Finish Edit, as Finish Sketch does
A bare "Finish" took the catalogs' surface-finish translation (German "Oberfläche").
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: fixes from the fourth review
Finish Edit waits while the part's agent is in the middle of its turn, instead of keeping the tab of any chat open
A part saved while its assembly refreshes is read again when the edit finishes: the refresh records the files as it found them
A double-click on a linked component while something is open in the assembly says so instead of activating it
Commands are recognised in any spelling the engine takes, so engine.batch and engine.script cannot slip Finish Edit through
Update Links run as a command refreshes the assembly from the control channel only, not for the app's own reads
Moves of components the refresh removed are not carried over, and a refusal is one message
The command description and the guide say that the command does not save the part
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: the command is Finish Edit, as Finish Sketch is
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: a command to start it, and fixes from the full review
Added component.edit_in_place {component, path?}: the double-click and Edit in Place in the menu run it, and as a command it asks nothing, so a network location not allowed yet or a design open elsewhere makes it fail instead
Finish Edit asked by the part's own agent finishes once that agent's turn ends; the button still waits for the turn
The parameters and design properties panels, with text typed and not applied yet, hold up entering and finishing an edit
A refresh is not taken up under an undo, a queued command or a feature still being computed in the assembly
Refusals say what holds them up (a refresh, a file being read or written) instead of nothing
The browser strip names the assembly that is refreshing, not any assembly
The schema description says the command does not save the part
Korean strings for every new message
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: fixes from the second full review
The MCP batch tool refuses component.edit_in_place and component.finish_edit as steps instead of sending them one by one
Edit in Place asked while the assembly's agent is in its turn opens the part once the turn ends; a double-click then is refused rather than stopping the agent
Both commands answer queued: true when they wait for the agent's turn, and stopping the agent forgets what it queued
An undo, a queued command or a delete check running in the design holds up entering and finishing an edit
A refresh redraws every edit of the assembly, the parts in other tabs too
In an assembly never saved, a double-click activates the linked component as before
The control and MCP docs list the new batch and script refusals, and the commands list every way they fail
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: fixes from the third full review
The MCP batch tool refuses the two commands however a step names them, an execute step too
A queued Edit in Place is checked fully when asked for, a second one is refused, and both queued requests wait for what runs in the design instead of being dropped
Disconnecting the agent forgets what it queued, as stopping it does
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Edit in place: help text for the Finish Edit button
main now requires help for every toolbar command (#7134), and Finish Edit
is one. Written in every language the branch's catalog carries.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #6032
* View angles: an optional readout under the orientation gizmo and a command to set them (#6032)
View > View angles (also Preferences > Graphics, or view.angle_readout) shows the
view's pitch, yaw and roll in degrees and Ortho/Persp under the axis view or cube,
updated live; off by default. Clicking the readout opens a small card to type
exact angles, which runs view.apply.
The angles are the camera's own (Camera::view_angles / with_view_angles), about
the up axis: yaw is the heading of the line of sight (Front 0, Left 90, Right -90,
Back 180), pitch the eye's elevation (Top 90, Bottom -90), and a positive roll
turns the model clockwise. view.apply documents this convention and now refuses a
pitch past +/-pi/2.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7181
* MCP save: a folder the agent opened or saved a design in stays in scope (#7181)
The agent's default folders held the open design's folder only while that
design stayed open there, so after a new design, an opened STEP file or a
save-as elsewhere a save beside the earlier design was refused, while the
same save through the control port worked.
The MCP server now keeps, for the rest of its run, the folder of every
design file a call opened or saved (open, save, and doc.open, file.save,
file.save_as through execute or batch, path given in full) and, headless,
of every design open in its session. Such a call only succeeded because the
folder was already in scope, so naming a path adds nothing. The folders get
the open design folder's terms (inside home or temp, not hidden, hidden
entries unreachable), at most 16. Bridged, they go to the app as
design_folders, which it judges the same way.
The save tool description and docs/mcp.md say how to add a folder with
--allow-path.
* MCP scope: a design's folder is the folder its file resolves to (#7181)
A link in another folder to a design inside the agent's folders passes the
scope check, which follows it, but its own folder was what got remembered
(and, for the open design, what counted while it was open). Take the folder
of the file as the check resolves it.
* Control test: a hidden file the export format accepts (#7181)
---------
Co-authored-by: MateuszSychXpl <151755914+MateuszSychXpl@users.noreply.github.com>
* claim: new designs save as .sc
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* New designs save as .sc; .solidcraft files still open and keep their name on Save
DESIGN_EXT is now "sc"; DESIGN_EXTS lists sc, solidcraft and solvecraft and
every open path (dialogs, drag-drop, recent files, CLI, MCP, links, the web
picker) accepts all three. Save of an opened .solidcraft file writes that
file; Save As, autosave, recovery and crash reports use .sc. Windows, macOS
and Linux register .sc alongside the older extensions.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Update UI tests for the .sc default extension
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The app id moves from ai.storyteller.solidcraft to io.tangencylabs.solidcraft: macOS bundle id
(the CLI signs as io.tangencylabs.solidcraft.cli), Linux desktop/metainfo/MIME file names, icon
name, Wayland app_id / X11 class, Flatpak, Snap common-id, nfpm/FreeBSD/Nix/AppImage staging,
the web app's icon paths and the docs.
Existing installs keep working: the Flatpak repository marks both old ids
(ai.storyteller.solidcraft and ai.storyteller.solvecraft) end-of-life with a rebase straight to
the new id, re-marking the solvecraft ref that pointed at the 0.8 id; the manifest grants
read-only access to both old ~/.var/app folders and the first launch copies settings and
autosaves from the newest one; AppStream <replaces>/<provides> and X-Flatpak-RenamedFrom list
both; the AppImage removes menu entries it wrote under either old id. The data and settings
folders (solidcraft / SolidCraft) are unchanged. The MSI UpgradeCode and ProgIDs are unchanged.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #5853
* Batch: refuse a step that would fail at its turn before the first step runs (#5853)
engine.batch and the MCP batch tool checked only a batch's shape up
front. A misspelled command id in a late step was found at its turn,
after every earlier step had run, and then all of it was rolled back: a
160-step batch with a typo as its last step ran for a minute and
changed nothing.
solidcraft_engine::batch_preflight::check finds, before anything runs,
the first step the batch would certainly refuse at its turn: a step
without a command, an unknown command id (the error names the closest
ids, as at the step's turn) or, with strict parameter checking, a call
that breaks its command's typed contract. Anything that depends on the
design is left to the step's turn.
- engine.batch (control channel) checks against the app's registry and
its session's level; view.pick, which the app answers itself, passes.
- The MCP batch tool checks when the commands run in its own process
(headless, with --param-check from the backend). A running app may be
another version with other commands and checks the batch itself; its
reply's not_run is passed on.
- The refusal is the failing step's reply with completed 0, undone 0,
not_run listing every step and a note.
Tests that used an unknown command only to make a late step fail now
use a step that fails at its turn (a bad extrude distance), so they
still test rollback, timings and file.new as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FN5eEMrqWWXqi65FcK8GXq
---------
Co-authored-by: sascha2101 <221609842+sascha2101@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #6646
* Control channel: document.revision, a design revision and a document id (#6646)
A program that follows the open design asks document.revision and exports
only when the answer moves: document_id names the open design (kept through
edits, saves and renames, another for each design started or opened, never
shared between two tabs or two runs) and revision moves whenever the design,
what is hidden or an uncaptured component position changes. It stays put on
reads, selections and view changes, unlike the session's view revision.
The revision is read lazily: the session remembers the design it last
reported (held weakly, so Arc::make_mut can never change it in place and its
address can't be reused) and takes a new number from one counter for the
whole run when the design is another one. The control method answers at once,
also while a slow command computes; engine.execute reaches the same command
for scripts and MCP.
* document.revision: hash the run's id prefix with the process's random keys
A clock set before 1970 gave every run the prefix 0, so a restarted app could
answer a new design with the id and revision of the old one. The docs no
longer promise an answer at once: a screenshot waiting for an opened design to
be meshed still holds the requests behind it.
* document.revision: its label in the complete catalogs (ja, zh)
* claim: #6985
* Control channel: control.subscribe and design.changed events (#6985)
A connection that sends control.subscribe gets the active design's id and
revision in the reply (as document.revision answers), then a
{"event":"design.changed","params":{document_id, revision}} line whenever
either changes, so a live link need not poll.
The window publishes the design at the end of each frame's logic and
drawing while a connection is subscribed (one atomic load otherwise). Each
subscribed connection waits on that from a thread of its own and writes
the latest after a 100 ms gap, so a burst of changes is one line. Replies
and events share one lock per connection, each written whole (#6390).
* control.subscribe: the reply before any event, a second one refused
The events thread started before the subscribe reply was written, so a
change right after the window's answer could reach the client before the
reply that named the older revision. The subscription now starts holding
the connection's line lock and writes the reply under it.
A second control.subscribe on a subscribed connection is refused on the
connection itself: its answer could arrive behind a newer event. When no
thread or socket is left for the events, the reply says it did not
subscribe instead of ok.
* control.subscribe: a second one does not lose a finished late reply
The already-subscribed answer came before the busy check, so a request that timed out and had finished since lost its reply: control.pending then said the app ended it without one. It now comes after the busy check and drops a finished late reply as every other request does.
* control_server: declare the subscribe guard at the top of the loop
Clear of the lines #6643 changes (its token check moves to where the declaration was).
* control_server: the subscription's items beside serve
Away from the lines next to send, where #6643 adds token_of.
* control test: New on a blank design reuses its tab, so redo the box first
---------
Co-authored-by: MateuszSychXpl <151755914+MateuszSychXpl@users.noreply.github.com>
* Docs: on Windows, give claude mcp add the CLI's full path (#5895)
The Windows installer puts solidcraft-cli.exe in C:\Program Files\SolidCraft but not on PATH,
so the README's and docs/mcp.md's `claude mcp add solidcraft -- solidcraft-cli mcp` registers a
server that can't start there. Say so next to the command and give the full-path form.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Docs: word the Windows full-path note so it holds once the installer sets PATH (#5895)
#5898 makes the MSI add its folder to PATH. Say "if a terminal doesn't find solidcraft-cli
(older Windows installers don't put it on PATH)" instead of stating that the installer never
does, so the note stays true whichever change lands first.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Docs: the Windows CLI path is the default, not the only place (#5895)
The MSI installs to ProgramFiles6432Folder and lets people choose the folder
(WixUI_InstallDir), so C:\Program Files\SolidCraft\ is only the x64 installer's default; the
x86 MSI on 64-bit Windows lands in C:\Program Files (x86)\SolidCraft\. Say "by default" in
the README and docs/mcp.md (review on #5896).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Every new command comes with input and output schemas
A test fails for a command without a contract or an output type, unless it is a planned!
placeholder or listed in schema/unconverted.txt, which holds the 400 commands that predate
the rule and only shrinks: the test also fails when a listed command has its schemas, when
the list names a command that doesn't exist, or when it isn't sorted.
`checked!` now takes only the command's id and checks the contract that id publishes
(`run_checked_as`), so the contract enforced is always the one described. Another test runs
every command that has a contract in strict with an unknown key, so a contract is never
published without `checked!`, under the command's own id, enforcing it.
A new book page, Command schemas, explains how to write a contract and a reply type, the
tests to add, the shared field types and how to add one, with a checklist; the failure
messages point to it. "Adding a command", CONTRIBUTING.md and AGENTS.md now include the
schemas, and the MCP tool description and docs/mcp.md no longer list the converted commands.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* List the commands added on main since: view.layout, view.history_layout, view.normal_to
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Command schemas guide: what participants and targets do, checking the evaluation, warning-free replies
- `participants` changes the result for some features and nothing for others; where it and
another key override each other, or where it has no effect, say so in that key's
description or in the summary.
- The example primitive's `targets` default to every body in the feature's component, not the
bodies it meets; they're ignored when `participants` lists bodies.
- Check what each value does where the feature is evaluated, not only how the handler reads it.
- A reply test runs every form in Warn and checks for no `param_warnings` (Strict for a reply
that isn't an object).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* List the 43 commands added on main since: frame, component, render and others
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* List the 11 commands added on main since: browser folding, standard views, parameter sweep, path flange
browser.collapse_all, browser.expand_all, inspect.parameter_sweep, sheet.path_flange and the
seven view.* standard views (back, bottom, front, isometric, left, right, top) landed on main
without schemas; the gate flags them, as intended. 457 commands listed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Schema rule warns instead of failing for now
A command without input and output schemas prints a warning listing it
(test output and a CI annotation). SCHEMA_RULE_STRICT in coverage_tests.rs
(or SOLIDCRAFT_SCHEMA_RULE_STRICT=1) makes it fail again. Docs say "should,
and will become required".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
* claim: #3983
* Agent skill for reproducible bug reports (#3983)
skills/solidcraft-reporting has an agent gather the version, steps and
exact error, reproduce the problem in a fresh design (a command script or
one batch with rollback off), shrink it while it still fails, check the
numbers with measure and a screenshot, keep private designs out, search
for duplicates, and write the report in one shape: Reproduced on, Steps
with a minimal script, Expected, Actual, Attachments.
Listed in skills/README.md and docs/mcp.md; the bug template's steps
field points agents to it. The skill-name check now covers its files.
* Reporting skill: GitHub takes a design only as a .zip
At launch the app adds the solidcraft entry to each installed AI client off the window thread,
names newly connected clients once, and refreshes a stale path. Settings > Assistant has
"Make SolidCraft available to AI apps on this computer" (on by default); off removes the entries it
added. Entries the user changed are left alone; automation runs never touch the files.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
solidcraft-cli mcp setup (alias install) and Help > Connect AI assistants edit each installed
client's own configuration in place, with this program's absolute path: OpenCode, Claude Desktop,
Claude Code, Cursor, Codex, Gemini CLI, Copilot CLI, Windsurf, Zed and VS Code. Only the solidcraft
entries change; a stale path is refreshed and an entry the user changed is left alone.
Plain mcp keeps looking for the app while its headless session has no work, and says on every
tool result that it is headless when the app was not found.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The AppImage writes a ~/.local/bin/solidcraft-cli launcher from its desktop-integration step
(same opt-outs, never over a file it did not write). The macOS app links the CLI into
/opt/homebrew/bin or /usr/local/bin at launch when one is writable, and Help > Install
solidcraft-cli command does it with an administrator prompt otherwise. Snap and Flatpak are documented.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #5822
* Batch: consecutive parameter edits can rebuild the design once (#5822)
An agent's batch of parameter edits, one parameter per step, evaluated the
design once per edit: an edit early in the timeline rebuilds nearly all of
it, and 21 such edits ran for minutes.
engine.batch and the MCP batch tool take coalesce_parameters (default
false). With it, each run of consecutive plain parameter edits
(parameters.change with only a name and an expression, or with changes) is
set as one parameters.change {changes}: one rebuild for the run. Any other
step ends a run, so later steps see the design up to date. Every step
still gets its own result (coalesced: the run's length); the run is one
undo step. A run whose combined edit is refused (a new parameter, which
changes can't make; a bad expression) runs step by step, so a refused edit
is still named by its own step. The app's off-thread evaluation computes
the run the same way. The grouping is solidcraft_engine::batch_params,
shared by both runners.
21 parameter edits on a plate with 21 extruded bosses (in-process MCP,
ci profile): 40.1 s step by step, 3.5 s coalesced, the same volume.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Batch coalescing: say a run skips the states between its edits; test it against step by step (#5822)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #4099
* Measure: a face or edge picked by its point alone (#4099)
inspect.measure (and the MCP measure tool) refused a face or edge item
without an index ("missing field `index`"). Face and edge items are now
read as body + point with an optional index: the point finds the face
(the same lookup face picking uses) or the nearest edge within the pick
tolerance, before the selection is built. A missing body or a point on
no face or edge is a clear error. Explicit indices still work.
Rebased onto main as one commit (was 120d26f7, df488f9a, db9d1073 plus
merges of main).
- A request the app reports as still running (#1618) is waited for on the
same connection through `control.pending`; past the wait the error says it
was not cancelled and will still apply (or only write its file), later calls
are not sent behind it, and the new read-only `pending` tool returns its
result once it lands.
- A failed batch says what really happened: `rolled_back` only when the
design is back where the batch started (as #3288 / #3242 on main: the app's
`rolled_back` is believed; for an older app, every completed step undone;
only a rollback asked for is reported as one), plus `applied`, `not_run` and
`failed.still_running`.
- A batch step may name a tool (set_parameter, list_commands,
describe_command, execute, measure, inspect_design, undo, redo, export, save,
open, new_design, crash_reports); screenshot, body_topology, batch and
pending are refused before anything is sent.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* docs(mcp): Windows setup for Claude Desktop (paths, two servers, Store config file)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(mcp): say the Store build can keep the config elsewhere, not that the other path fails
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The app opens its localhost control channel on port 27878 (or the next free
one) unless Preferences > Assistant turns it off; --control and
SOLIDCRAFT_CONTROL_PORT still win. A current.port discovery file lets
solidcraft-cli mcp find the running app with no arguments.
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
* claim: #4966
* Agent chat: serve the bundled skills' reference through an MCP tool (#4966)
The bundled skills start with "read reference.md", but SolidCraft's own
chat turns the agent's file tools off, so its agent could not follow them.
The MCP server gains a read-only `skill_reference` tool that returns a
skill's documents from copies built into the binary. It takes a skill
name and a file name from a fixed list, never a path, and refuses an
unknown skill or file with the available ones. The skills point to it,
and docs/mcp.md and skills/README.md describe it.
* claim: #3654
* Control channel: control.running lists what the app is running, across connections (#3654)
control.pending answers only for the connection that asks, so a second client,
the MCP bridge after a reconnect or a script saw {"pending": null} while the app
was busy. control.running answers for the whole app, the same on every
connection and at once (the connection's own thread answers it, also while it
is busy with a timed-out request): every request the app has been sent and not
answered yet, oldest first, with its connection, id, method, command id (or a
batch's step count), seconds since it arrived and whether it still waits its
turn. A timed-out request stays listed until the app answers it, also after its
connection closed. Parameters are never shown.
The window marks a request started when it takes it (ControlRequest.started),
which tells queued from running. The MCP server gets a read-only `running` tool.
---------
Co-authored-by: MateuszSychXpl <151755914+MateuszSychXpl@users.noreply.github.com>
* joint.inspect: what each joint is attached to, for agents
A read-only `joint.inspect {joint?}` gives each joint's definition: for each
side the occurrence, its component and path, the snap as saved, and what
each persistent face/edge name refers to (the feature that made it, and the
sketch and curve a swept side came from), whether the origin resolves and
why not; plus kind, offset, angle, flip, values with limits, expressions,
the motion links it is in, and its solve errors and conflicts. Its id ends
in `.inspect`, so agent gates treat it as reading only.
MCP `inspect_design` includes the definitions when the design has joints
(`joints: false` leaves them out). joint.list and joint.inspect share the
motion-link JSON.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* i18n: catalog entries for Inspect Joint (en, ja, zh)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Spike: typed parameter contracts and JSON Schemas for solid.box and solid.extrude
The schema RFC's spike (plan/schema-spike.md). solid.box and solid.extrude get
typed contracts with generated JSON Schemas (2020-12), checked in front of the
unchanged handlers by a per-session level: off (the default, which changes
nothing), warn (the call runs and gains param_warnings) or strict (the call is
refused and the design left alone). Also engine.describe and the MCP tool
describe_command, `solidcraft-cli mcp --param-check off|warn|strict`, a
FeatureResult type for feature replies, and an env-var audit recorder
(SOLIDCRAFT_PARAM_AUDIT) used to check the contracts against every test.
Local spike branch; not for upstream as is (see the results report).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Parameter contracts: put the audit recorder behind a feature; drop two ignored test keys
The audit recorder is now on only in a build with the engine's `param-audit` feature
and `SOLIDCRAFT_PARAM_AUDIT` set. Its code is still compiled, linted and tested without
the feature. `scripts/param_audit.py` summarises a test run's audit logs, and the schema
module's docs give the steps for converting a further command.
Two tests passed keys their command ignores, which the audit found: `position` on
`solid.box` (`hole_face_tests.rs`) and `symmetric` on `solid.extrude` (`face_tests.rs`).
Both are removed, so the tests build what they always did. Comments no longer call
the contracts a spike.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
Measure now reads the picked item's exact shape (cylindrical faces, circular and arc edges,
sketch circles, arcs and lines, profiles, planes, points):
- A circle, hole or arc shows Ø, R and its length (circumference, or arc length and angle).
- A slot (a sketch profile, or a half-cylinder end of a slot in a solid) shows its width,
centre-to-centre length and overall length.
- Two items where one is round measure from the centre or axis by default (axis to axis,
axis to plane, edge or point), with the closest and farthest distances listed too; the
panel switches between centre, min and max, and the canvas line follows.
- inspect.measure and MCP `measure` (new `items` argument) add diameter, radius,
center_distance, min and max; existing fields are unchanged.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Fixes#1167. While a sketch is edited, a body that a rolled-off, unsuppressed Remove deletes is not shown (unless the sketch is built on it); the history stays rolled back, only the view changes. Drawing, picking, screenshots, thumbnails and the browser's Bodies folder agree; document.inspect reports removed_later. Checked on main + this branch: cargo check --workspace, the engine sketch_edit_removed/occurrences tests and the UI viewport/browser tests.
solidcraft-cli snapshot takes --camera JSON and --view with a saved view name (through view.apply, as the MCP screenshot does), plus --no-grid and --no-axes; the MCP screenshot and ui.render take grid/axes, the headless render skips a hidden origin axis, and the app's ui.render follows the Grid and Origin switches. Closes#1245.
file.insert_step and doc.open on a STEP file take `up` (turn a Y-up file so its up is +Z) and `names` (names for the imported bodies). The MCP tool and the app's control requests pass them through; an import without them works as before. Tests in the kernel's STEP reader, the engine, MCP and the file-load UI; docs/step-import.md and docs/mcp.md describe both.
A solidcraft-mechanisms agent skill with a worked example that a test replays through the MCP server (the actuator follows the hinge with the stated stroke and the drive stops at the rod's limit); the skills index and docs/mcp.md list it.
The bridged MCP server starts when no app is listening, says so in its instructions and in each tool error, and connects on the first call after the app starts. A batch that could not be sent says the app is not running. docs/mcp.md describes it.
A design note for SolidCraft in the browser: tiered storage, a SharedWorker job broker, GPU dispatch and in-browser agent control through the same engine commands, with the generic pieces in craft-libs. Linked from the MCP and web getting-started docs.
Lands the Render workspace (R3), bounded preview-image jobs with cancellation and atomic PNG export (R4a), and opaque PBR looks as one change.
- Render workspace: an overlay-free product preview on the installed GPU device. Returning to Design restores its camera. render.scene and render.settings.apply are added.
- Image jobs: render.preflight/start/status/cancel/export run bounded CPU preview jobs. Export captures the job, overwrite choice and document, and checks the caller's file scope before writing.
- Opaque PBR: looks can carry metal fraction, roughness and dielectric IOR. The product viewport and preview PNG use them; geometry and mass are unchanged. The v1 catalog export refuses PBR rather than dropping it.
- Render setups and PBR looks are saved under design format 21. Labels are in en/de/ja/ko/zh/ru.
Supersedes #706 and #800. Refs #620, #673, #756, #1004.
Contributed by @0xlillith.
Co-authored-by: lillith <35337584+0xlillith@users.noreply.github.com>
A design carries metadata (format 21): project, part number, revision, author, organization, description and license, plus other tools' data under reverse-DNS keys, kept as written. Edited in the Design Properties window (Document Settings), set by document.metadata.set, reported by inspect_design, and written into STEP, IGES and 3MF where the format has a place for them. Translated, with tests.
PathScope::check reports a missing folder inside the scope as NotFound with the nearest existing folder, instead of calling it outside the allowed folders. Paths outside the scope and dangling links are still refused. Fixes#1077.
Colin Faulkingham