Uncanvased slices → Event Model board (Excalidraw)

This is the backward pass: authored slice YAML → one Event Modeling board as an .excalidraw file. A bundled deterministic generator does the layout — run the script; never hand-author Excalidraw JSON.

Quick start

python3 .claude/skills/slice-to-diagram/scripts/generate_diagram.py slices/ -o board.excalidraw

paths may be a directory (scanned for *.yaml) and/or individual slice files; all selected slices land on ONE combined left-to-right timeline.

Flag	Effect
--compact	names only, no field lists (big-picture or workshop view)
--no-specs	leave out the Given / When / Then area

Workflow

1. Decide the selection: the whole slices/ folder, or the files the user named.
2. Run the generator. It prints the output path, counts (columns, duplicated events, story hints), the event streams, and the slice order with ids.
3. Report the output path and slice order. The file opens at excalidraw.com or in the Excalidraw VS Code extension. Do not echo the JSON.
4. Surface every warning verbatim:
   • ! off-board reference drawn as ghost: a name no selected slice produces;
   • ! read model … sits right of its reader: a sequencing error;
   • ~ story: …: a reader placed before an event that changes it.
5. If there are story hints, offer to fix the timeline: each hint names the slice to move later (raise its sequence:), which removes the duplicate. If no event stream is set, mention that context: splits the events lane. Both are modelling decisions, so ask before editing slice files (the img-to-slice gap review covers them as out-of-story and no-stream).

The board

It reads as a storyboard, left to right in time:

Lane	Holds
One per actor (Customer, Branch Agent, System Agent, …)	only UI wireframes (with example data) and ⚙️ processors
Commands / Read models	blue commands and green read models, each with its fields ((tag) marks DCB tags)
Events · <stream>	one orange lane per context:; events show their fields
One per external system (Stripe, …)	an automation's external_call, a translation's external event
Specifications	each slice's Given / When / Then scenarios, under its column

Rules of the board

• It tells one story. Slices follow sequence:; a reader sits after the events it reads.
• Arrows never point left. If a view reads an event that happens later, it gets its own hatched (duplicate) copy of it, and the script prints a story hint. Duplicates are a smell to fix by reordering, not decoration.
• Loops are not story errors. An event recorded by the view's own consumer (the ⚙️ it triggers, or the button on the screen it feeds) is written on the view as cleared by … / updated by … instead of an arrow.
• Events are the only shared contract. Commands and read models are drawn once, in their own slice.
• A screen is one frame. A read model's screen and the button for the next command are one wireframe when the slices are neighbours (badge 12·13).
• Every element carries its slice id as a corner badge; duplicates carry the producer's id. img-to-slice uses the badges to match an edited board back to its files.
• Specs are first-class. Scenarios are drawn under their slice; failures appear as red stickies with the message as a white note.

Slices are written by the img-to-slice skill; its REFERENCE.md defines the YAML.

Details

See REFERENCE.md for the per-pattern mapping, where example data comes from, the duplicate and loop rules, the palette and geometry, and how to extend the script.