vd_add_asset
ChatGPTRegister an image under a name, and optionally place it. The source is one of: {"url": "https://…"} (public https only); {"data": "<base64>", "type": "image/png"} for bytes you actually have. A picture the person gave you or has on their machine is neither — you cannot pass on an image you can see, and no tool reads file paths. For that, do not call this tool: write the image with a src nothing supplies yet and a name saying what goes there (image nav.logo … src=logo name="Your logo"). The canvas shows it as a slot the person drops the file into; carry on, and tell them it is there. Never write an asset line yourself: place the image with image <id> src=<name>.
vd_add_icon
ChatGPTPlace Lucide icons as vector nodes, by name — a name from vd_search_icon, or one you know (old names like "home" or "alert-circle" still work). Up to 20 in one call, one undo step; all land or none do. x, y and size take a number or an expression such as "(@cta.left + 16)". The stroke scales with the size as Lucide’s does; color defaults to the document’s ink token, or the text colour.
vd_assets
ChatGPTEvery image registered in a document: its name (what image … src= refers to), type, pixel size, bytes, attribution, and which nodes use it — then the slots still waiting for the person to drop a picture into.
vd_create
ChatGPTStart a new .vd document. Leave "dsl" out for an empty canvas with no page — insert the frame you need (vd_patch: insert frame page x=0 y=0 w=1280 h=720) — or pass a skeleton: title, tokens, fonts and the frames. Then add each region with its own vd_patch rather than writing the whole design here: the person sees each call land, and one large call is a long wait on an empty canvas. Load the vdesign skill first if you do not know the DSL (vd_skill). Nothing is created if the DSL would not parse. You cannot save it. Against a running app the new document lives in a window until the person using it saves — vd_list shows which documents that applies to.
vd_list
ChatGPTThe documents available here, with the handle every other tool needs. Call this first: a handle you guessed is how a session starts by editing the wrong document, and there is no other way to learn one.
vd_open
ChatGPTA link that opens a document vd_list marks "not open" in vDesign, on the web or the desktop app. Give it to the person: such a document was made while vDesign was not open, and is kept for a week until they open it. Its handle keeps working afterwards.
vd_patch
ChatGPTApply mutation statements (set / insert / delete / move / motion) to a .vd document. The dialect is the skill’s reference/edits.md (vd_skill path "reference/edits.md"). Nothing is written if any statement fails to parse.
vd_problems
ChatGPTEverything wrong with a .vd document: what would not parse, and what would not draw. Worth calling after any edit you are not certain about.
vd_read
ChatGPTThe current state of a .vd document. Defaults to an outline — every node with its id, type and box — which is what most edits need. Ask for a node id to see one subtree in full, or mode "full" for the whole design. Never returns image bytes.
vd_remove_asset
ChatGPTUnregister an image. Refused while nodes still draw it, unless force is true, which deletes those image nodes too.
vd_search_icon
ChatGPTFind icons in the Lucide set (1,900 line icons) by name or by what they show: "settings", "notification", "arrow right". One line per icon, its name then its tags. Place one with vd_add_icon — never write an icon’s path data yourself. Lucide has no brand logos (GitHub, X, YouTube): leave a logo as an image slot.
vd_show
ChatGPTShow the person a document in the chat, as a preview of its pages. Call it once, when a piece of work is done, never after each vd_create or vd_patch: every call draws another preview. Pictures show as placeholders.
vd_skill
ChatGPTLoads the vdesign skill. Author and edit vDesign documents — posters, slides, hero sections, card grids, any laid-out visual document — in the .vd DSL. Use when asked to create, change, or read a vDesign document, a .vd file, or when a task involves designing a layout that should stay editable and relational rather than being flattened to fixed coordinates. Returns SKILL.md unless asked for another file; ask for one only when you need it, which is what keeps this cheap.