Example Selection By Capability

Use docs/examples/capabilities.json when a user asks for a visual, feature, data kind, or domain tag and the best starting example is not obvious by filename.

Lookup Order

Search capability groups in this order:

1. visuals for requests such as point, marker, image, mesh, text, labels, sphere, or volume;
2. features for layout, controllers, picking, updates, colorbars, scale bars, lighting, or blending;
3. composites for semantic objects such as polygon and graph;
4. tags for workflows, domains, techniques, interaction modes, or real-data labels;
5. data_kinds when the user specifies synthetic, simulated, prepared, or real data.

Each capability entry lists examples with id, title, category, status, agent_copy_safe, source, and page. After finding candidates, open docs/examples/examples.json for build, smoke, validation, dataset, and summary details.

Selection Rules

Prefer candidates in this order:

1. agent_copy_safe: true;
2. status: supported;
3. atomic categories: visual, feature, then composite;
4. showcase examples only for composed workflows, real-data stories, or polished demos.

When no copy-safe candidate exists, say so and use the example as reference material instead of copying it directly.

Common Routes

Request	Capability path
Scatter or point cloud basics	visuals.point, then tags.dense-point-cloud if needed
Image with color scale	visuals.image, features.colorbar, features.colormap-scale
Axes and labels	features.axes-2d, features.axis-labels
Pan and zoom	features.controller-panzoom, tags.panzoom
3D mesh interaction	visuals.mesh, features.controller-arcball, features.mesh-texture
Picking or probing	features.pick-point, features.pick-hover, features.image-probe
Real or prepared datasets	data_kinds.real, data_kinds.prepared, then inspect dataset
Scientific showcase	tags.scientific, then prefer the matching data kind

Guardrails

Do not treat tag coverage as API support by itself. Confirm support with the example's status, documentation page, and validation command.

Do not turn a showcase into a minimal template unless the user explicitly needs that workflow. Showcases may include dataset loading, GUI controls, capture setup, or experimental runtime paths that are inappropriate for small generated snippets.