canvas

The canvas worker stores diagrams as editable source. A canvas is a named record — mermaid text, or an excalidraw scene JSON for a freeform whiteboard — kept under a stable 8-character id. The worker never renders anything; the console does the drawing, so a canvas::* call appears in chat as the live diagram and the canvas page lists and edits everything stored. The source is the artifact, which is what keeps a diagram editable instead of frozen into pixels.

Ids are stable across updates on purpose: revise a diagram as the conversation evolves and every earlier reference to it keeps working. Records persist in the state worker, so nothing is lost across a restart.

Mermaid should not be written from memory. canvas::syntax returns the families the renderer actually supports with a working example each, and canvas::validate parses source without storing it — together they mean a broken diagram never lands in the store or renders as an error card.

When to Use

• A diagram would say it better than prose — an architecture, a sequence of calls, a state machine, an ER model: create a mermaid canvas and let the console draw it.
• About to write mermaid source: call canvas::syntax first, narrowed to the family being written.
• Generated source in hand: canvas::validate before canvas::create or canvas::update.
• A diagram needs revising: canvas::update on the existing id, not a new canvas — the id is what earlier references point at.
• A spatial sketch a person will edit by hand on a whiteboard: create with format freeform and an excalidraw scene JSON as the source.
• Drawing while a person watches: create a freeform canvas, then canvas::element::add one shape or small group per call — the open console board shows each step as it lands.
• Finding what is already drawn: canvas::list, newest first, with an optional format filter; canvas::element::list for the shapes on one freeform board.

Boundaries

• Nothing here renders. Headless callers get source back, never SVG or pixels; drawing happens in the console.
• Not a file store and not a document editor: the source is one diagram, capped by the configured max_source_bytes. Files belong to the shell worker.
• canvas::validate checks that source parses, not that the diagram is any good — a valid diagram can still be the wrong diagram.
• A canvas's format is fixed at creation; canvas::update changes name and source, not format.
• canvas::delete is idempotent (deleted=false for an unknown id); the other id-taking functions error on unknown ids.

Functions

• canvas::create — store a new canvas from mermaid text or an excalidraw scene JSON; mints the stable id and derives the mermaid family.
• canvas::get — read one canvas by id, editable source included.
• canvas::list — every stored canvas, newest first, optionally filtered by format; capped by the configured max_list.
• canvas::update — revise a canvas's name and/or source by id; the id never changes and the mermaid family is re-derived.
• canvas::delete — remove one canvas by id; unknown ids report deleted=false rather than erroring.
• canvas::syntax — the mermaid syntax reference: every supported diagram family with a summary and a working example, optionally one family.
• canvas::validate — parse source without storing it; reports validity, the derived family, and per-issue messages with line numbers where known.
• canvas::element::add — append shapes to a freeform canvas one drawing step at a time; the open console whiteboard shows each call as it lands.
• canvas::element::update — merge properties into one element of a freeform canvas by element id: move, recolor, or relabel a shape.
• canvas::element::delete — remove elements of a freeform canvas by element id; unknown ids are ignored and the response reports the count.
• canvas::element::list — the freeform board map: id, type, position, size, and text per element, read before updating or connecting shapes.

The element family works only on format: freeform canvases (excalidraw scene source). Mermaid canvases are edited as text through canvas::update.