@inneranimalmedia/agentsam-loading-scene

Provider-agnostic runtime visualization for AgentSam.

The package turns normalized runtime events into one persistent controller/world/canvas and nine reusable full-frame Computational Hyperspace studies. Runtime semantics choose the study automatically and the renderer cross-fades between studies without rebuilding the canvas, controller, world, or RAF.

Nine fullscreen studies

| Study | Runtime use | | --- | --- | | Interdimensional Runway | idle / boot | | Signal Through the Void | reading / thinking / external waiting | | Infinite Layer Stack | context loading / asset generation | | Quantum Navigation Field | tool execution / parallel execution | | Warped Merkle Space | indexing / verification / compaction | | Event Horizon Compute Field | build / deployment / manufacturing compile / receipt persistence / success / error |\n| Vector Trace Lattice | raster-to-vector tracing / path convergence |\n| Spectral Gamut Warper | color-profile normalization / alpha-color preparation |\n| Thread Density Matrix | embroidery pre-digitization handoff |

Each study is an infinite Canvas 2D loop and scales to the consumer container, so the same component works in a browser side panel, full-screen work surface, modal, or embedded editor.

Architecture

runtime events
  -> provider-agnostic adapter
  -> LoadingSceneController
  -> SceneState + semantic intent
  -> semantic study selection
  -> seamless cross-fade
  -> one persistent canvas / one RAF

React

import { LoadingSceneController, computationalHyperspace } from "@inneranimalmedia/agentsam-loading-scene";
import { LoadingScene, LoadingSceneStatus } from "@inneranimalmedia/agentsam-loading-scene/react";

const controller = new LoadingSceneController(computationalHyperspace);

<LoadingScene
  controller={controller}
  preset={computationalHyperspace}
  study="auto"
  reducedMotion="auto"
>
  <YourRealResult />
</LoadingScene>

<LoadingSceneStatus controller={controller} preset={computationalHyperspace} />

study="auto" is the production mode. Consumers emit truthful runtime activity; the package chooses and transitions among the nine studies.

For galleries, fixtures, and targeted previews, pin one study with study="runway", signal, layers, quantum, merkle, horizon, vector, gamut, or thread.

Canonical AgentSam activity

The package directly accepts the SDK runtime contract agentsam.activity.v1 through adaptAgentSamActivity(). progress.current / progress.total is projected as deterministic metric points and can drive both the progress indicator and scene narration.

const sceneEvent = adaptAgentSamActivity(activityEvent);
controller.handle(sceneEvent);

The adapter maps observe/context/plan/execute/verify/compact/waiting/complete phases and tool/artifact/file/approval/verification events to the same semantic scene vocabulary. This lets CLI, web, desktop, lead chat and co-worker surfaces consume one run activity stream instead of inventing product-specific progress.

Performance contract

• one persistent canvas
• one requestAnimationFrame loop
• conserved persistent World
• seamless cross-fades
• DPR cap
• document visibility pause/resume
• deterministic disposal
• reduced-motion support
• no image/video assets
• no WebGL or shaders
• no fabricated progress

Preview

npm run build -w @inneranimalmedia/agentsam-loading-scene
npm run dev -w hyperspace-demo

Open http://127.0.0.1:5181.

The demo can expose all nine runtime families and manual full-screen study controls. Use ?debugRuntimeVisuals=1 for runtime devtools or ?study=runway (or any other study id) to force one scene.

Gates

npm run typecheck -w @inneranimalmedia/agentsam-loading-scene
npm run test -w @inneranimalmedia/agentsam-loading-scene
npm run build -w @inneranimalmedia/agentsam-loading-scene
npm run build -w hyperspace-demo
npm pack --dry-run -w @inneranimalmedia/agentsam-loading-scene