feat: sync skills to CesiumJS 1.144 with optimized wording

[jdehorty]

What

Syncs the skills package to CesiumJS 1.144 (released 2026-08-01; no breaking changes, no deprecations), scoped to skills/** only — 17 files, +1623/−168. No eval infra, config, docs, or README changes (follow-up planned for those).

Consolidates three streams:

1. feat/eval-and-optimization skill improvements — the optimization-loop wording the tracked all-pass baselines were rendered from: hover/selection recipes, camera framing guidance, tileset render-readiness patterns, provider hygiene, time/property recipes, and expanded quick-reference tables across all 14 domain skills.

2. CesiumJS 1.144 release sync (from justin/czmjs-1144-sync-eval @ 7d95669):

   Area	Addition	Skill
   Camera	Composable Controller framework: ControllerHost, ScreenSpaceMapCameraController, ScreenSpaceElevatorCameraController, HybridScreenSpacePanCameraController, ScreenSpaceTiltOrbitCameraController, ScreenSpaceZoomCameraController, MouseButton, ScreenSpaceInputBindings, Viewer.addController	cesiumjs-camera
   Interaction	Scene.snap + SceneSnapResult experimental edge-snap picking	cesiumjs-interaction
   3D Tiles	UrlTemplate3DTilesDataProvider base class; automatic terrain draping of clamped vector tile polylines/polygons	cesiumjs-3d-tiles
   Imagery	Feature-info picking for imagery draped on 3D Tiles	cesiumjs-imagery
   Models	BENTLEY_materials_planar_fill support, KHR_mesh_primitive_restart, edge-visibility typed-array loading (~19x less JS heap)	cesiumjs-models-particles
   Materials	Texture.defaultColor placeholder color (runtime-public, untyped)	cesiumjs-materials-shaders

3. API-verification refinements — every 1.144 claim was checked against the published cesium@1.144 npm package (Cesium.d.ts + runtime import), the official CHANGES.md, and engine sources:

   • 1.144 adds exactly 12 exports to Cesium.d.ts; 11 are runtime-real. GRID_TARGET_SEGMENTS_PER_CELL is a declaration-only leak (undefined at runtime) — noted so agents never import it.
   • KHR_mesh_primitive_restart is not ratified (Khronos glTF PR #2569 still open) — described as the multi-vendor KHR successor still in review.
   • The fifth controller (ScreenSpaceZoomCameraController) is public even though the release notes list four.
   • Controller tuning knobs documented per class exactly as shipped: panSpeed/inertiaEnabled/inertialDecay (map/elevator), dampingEnabled (tilt-orbit/zoom), angleThreshold + nested controllers (hybrid).
   • Scene.snap defaults (width = 25, height = width), Model-pipeline-only scope, WebGL2 EXT_color_buffer_float requirement, and SceneSnapResult = { object, position, screenPosition, isEdge } all match the shipped JSDoc.
   • Draped-imagery pick fallback verified provider-generic in engine source (covers WMS and WMTS enablePickFeatures).
   • All 14 skill version baselines bumped to 1.144.

Conflict handling vs main

main merged #24 (frame named targets with bounding spheres) after the eval branch forked; the camera skill here is a three-way merge that preserves both: #24's flyToBoundingSphere target-framing guidance and quick-reference rows are intact, and the new Controller-framework note explicitly routes input-remap tasks back to ScreenSpaceCameraController's *EventTypes.

Evaluation evidence (feat/eval-and-optimization pipeline)

Vetted bytes == shipped bytes (the validated tree is byte-identical to this PR's skills/).

• Tier 1 — deterministic gate (npm run gate, same script as eval-branch CI): PASS — evaluation + scenario manifests validate, skill contract passes for all 15 skills against the 582-symbol registry, 227 unit tests pass, deterministic fixture scorecard 100%, fixture polarity verification passes, canonical eval surface + tracked public surface scans pass.
• Tier 2 — live eval: all 47 scenarios across the 6 content-changed skills (camera 14, imagery 15, interaction 5, 3d-tiles 5, models-particles 4, materials-shaders 4) re-generated from this PR's skill text (codegen: copilot / gpt-5.6-sol / low) and rendered under CesiumJS 1.144, then scored deterministically:
  • 46/47 cases all-checks-pass; combined audit score 99.3% (140/141 checks; threshold 95%).
  • The single failing case is cesiumjs-camera/eval-006 (a regression_critical: false scenario) and it is environment-caused, control-proven: only its tile-settle/screenshot gates fail ("globe tiles still streaming" past the 45 s budget); its substantive checks (no_console_errors, code_runs, all pattern checks) pass on every roll. Re-rendering the byte-identical Aug-24 source that previously settled in 5–9 s fails the same way under the host load present during validation (load average 24–29 with an external Chrome pegging a core; SwiftShader tile generation is CPU-bound). Re-run on a quiet host to get 47/47.
• Regression bar: the previous tracked baselines (rendered from the eval-branch skills) are 47/47 all-pass; case-by-case comparison shows no check-set differences in 46 cases. Transient infra flakes observed during the runs (a cesium.com CDN ERR_CERT_AUTHORITY_INVALID blip, Ion preflight timeouts, and one codegen style roll that hoisted baseLayer into a variable) all pass on retry and are unrelated to skill wording.
• The 8 skills whose only change is the version-baseline line were not re-rendered (per ADR-0005/0007 cost tiering); their existing all-pass baselines were rendered from this same content minus that one line.

Wording hardening informed by eval evidence

• Interaction eval-003 flaked in an earlier run (codegen chose rectangle graphics where the contract expects polygon): the hover recipe now pins graphics-type fidelity ("polygon over a region" ⇒ polygon graphics with PolygonHierarchy, not rectangle) and adds the entity material-swap variant.
• Camera eval-010 protection: the framework blockquote keeps rotateEventTypes/tiltEventTypes/zoomEventTypes canonical for remap tasks, so the new controllers cannot misdirect them.
• Imagery eval-002 flaked once on a source-regex (baseLayer: layer hoisted into a variable — functionally identical); fresh roll passes. Flagged as a scenario-regex hardening opportunity on the eval branch (not a skills issue).

Follow-ups (intentionally out of scope)

• README.md, .claude-plugin/marketplace.json, and docs/DOMAINS.md still say 1.143/~551 on main; a companion docs/metadata PR will bump them (content already prepared).
• Eval-branch companion Done: feat/eval-and-optimization now carries the 1.144 sync (7d95669), the wording corrections (6f20882), and an alignment commit (4d29273) making its skills/ byte-identical to this PR, so the operational eval branch renders baselines from the same wording that ships to main (pushed to origin).

Reviewer checklist

• New 1.144 sections (camera framework / snap / draping / draped picking / planar fill / defaultColor)
• "Not ratified" phrasing for KHR_mesh_primitive_restart
• Camera skill merge: Fix camera target framing guidance #24 bounding-sphere guidance + Controller framework coexist correctly
• Skills-only scope is acceptable (docs/metadata bump as follow-up)