Build and revise procedural 3D assets with your coding agent. Local MCP server, CLI and TypeScript engine with rendered review, editable source and GLB export.
JavaScript 3d3d-assets3d-modelingagent-skillsagent-toolsai-agentsasset-pipelinecligame-assetsglbgltfmcpmcp-servermodel-context-protocolprocedural-generationprocedural-modelingtext-to-3dthreejstypescript
/**
* ScriptedModel — a deterministic offline Strands `Model` for agent-loop tests.
*
* Plays back a fixed list of turns: each turn is either a set of tool calls
* (with complete JSON inputs) or a final text. Events are emitted in the exact
* shapes the SDK's `streamAggregated` consumes (verified against
* @strands-agents/sdk 1.10.0 dist/src/models/model.js), so a REAL `Agent` runs
* its genuine loop — hooks, concurrent tool execution, usage accumulation —
* against it with no network and no provider SDK.
*
* Test helper only (lives under __tests__ so it is coverage-exempt).
*/
import { Model, type Message, type StreamOptions, type Usage } from '@strands-agents/sdk';
import type { BaseModelConfig, ModelStreamEvent } from '@strands-agents/sdk';
export interface ScriptedToolCall {
name: string;
input?: unknown;
id?: string;
}
export type ScriptedTurn =
| { toolCalls: ScriptedToolCall[]; usage?: Usage }
| { text: string; usage?: Usage };
export class ScriptedModel extends Model {
/** `options.systemPrompt` captured per stream() call, in order. */
readonly seenSystemPrompts: unknown[] = [];
readonly messageSnapshots: unknown[] = [];
private turnIndex = 0;
constructor(private readonly turns: ScriptedTurn[]) {
super();
}
override updateConfig(_config: BaseModelConfig): void {
// scripted — nothing to configure
}
override getConfig(): BaseModelConfig {
return { modelId: 'scripted-test-model' };
}
override async *stream(
_messages: Message[],
options?: StreamOptions,
): AsyncIterable<ModelStreamEvent> {
this.seenSystemPrompts.push(options?.systemPrompt);
this.messageSnapshots.push(JSON.parse(JSON.stringify(_messages)));
const turn: ScriptedTurn = this.turns[this.turnIndex] ?? { text: '(script exhausted)' };
this.turnIndex += 1;
yield { type: 'modelMessageStartEvent', role: 'assistant' };
if ('toolCalls' in turn) {
let i = 0;
for (const call of turn.toolCalls) {
yield {
type: 'modelContentBlockStartEvent',
start: {
type: 'toolUseStart',
name: call.name,
toolUseId: call.id ?? `scripted_${this.turnIndex}_${i}`,
},
};
yield {
type: 'modelContentBlockDeltaEvent',
delta: { type: 'toolUseInputDelta', input: JSON.stringify(call.input ?? {}) },
};
yield { type: 'modelContentBlockStopEvent' };
i += 1;
}
} else {
yield { type: 'modelContentBlockStartEvent' };
yield { type: 'modelContentBlockDeltaEvent', delta: { type: 'textDelta', text: turn.text } };
yield { type: 'modelContentBlockStopEvent' };
}
if (turn.usage) yield { type: 'modelMetadataEvent', usage: turn.usage };
yield {
type: 'modelMessageStopEvent',
stopReason: 'toolCalls' in turn ? 'toolUse' : 'endTurn',
};
}
}