Get completion and validation in your editor#

Panel documents are YAML, and Scenet publishes a JSON Schema for them. Any editor with a YAML language server can therefore offer completion, hover documentation and inline validation — with no Scenet-specific plugin at all.

The schema is generated from the compiler’s own pydantic models, so what your editor suggests cannot drift from what actually compiles. A test fails if the shipped copy goes stale.

Any editor, via a schema comment#

Put this on the first line of a panel document:

# yaml-language-server: $schema=https://azias.github.io/scenet/schemas/panel.schema.json
panel:
  size: [1000, 800]
cast:
  alice: {reference: alice}

For a multi-panel document use scene.schema.json instead.

That comment is understood by the YAML language server used by VS Code, Neovim, Helix, Emacs (lsp-mode), IntelliJ and Zed, among others. You get:

  • Completion on every key, and on every enum value — all ten shot types, all four balloon kinds, all nine placement zones.

  • Hover documentation, because the docstrings on the models become description fields in the schema. Hovering shot: explains that it is an upper bound on tightness, not a promise.

  • Inline errors for unknown keys, wrong types and out-of-range values, as you type.

The VS Code extension#

A small extension bundles the same schemas plus a side-by-side preview:

  1. Download scenet.vsix from the latest release.

  2. In VS Code: Extensions → ⋯ → Install from VSIX…

It contributes:

  • Schema association for *.panel.yaml and *.scene.yaml, so no $schema comment is needed.

  • Syntax highlighting for .script comic scripts.

  • Scenet: Preview panel, which compiles the current document and shows the SVG beside it.

The preview shells out to the scenet command, so it needs the compiler installed. If it is not on your PATH, set scenet.executable — uv run scenet works, as does an absolute path.

It depends on the Red Hat YAML extension, which VS Code will offer to install for you.

Generating the schema yourself#

scenet schema -o panel.schema.json
scenet schema --scene -o scene.schema.json

Both go to stdout without -o. This is the same command CI runs, and the same one that produces the copies shipped in the extension.

From Python:

import json

from scenet import PanelIR

schema = PanelIR.model_json_schema()

assert schema["$defs"]["ShotType"]["enum"][0] == "long_shot"

# The docstrings travel with it -- which is where the hover text comes from.
assert "upper bound" in schema["$defs"]["ShotType"]["description"]

document = json.dumps(schema, indent=2, sort_keys=True)

Why it is generated and not written#

A hand-written schema is a second definition of the language, and second definitions drift. Within a release or two the editor is suggesting a key the compiler rejects, or failing to suggest one it accepts, and nobody notices because nothing checks.

Here there is one definition — the pydantic models — and the schema is a projection of it. The prose in the models is the prose in your editor’s hover. The enum in the models is the enum in your completion list. Adding a construct to the language updates both by construction.