Spot it. Save it. Free, local-first Chrome extension for images, videos and prompts. Explore 2,000+ curated cases and connect your library to AI Agents via MCP. 提示词导演
JavaScript ai-artai-videochrome-extensionclaude-codecodexhiggsfieldjimenglocal-firstmcpmidjourneymoodboardprompt-engineeringprompt-libraryprompt-managerreference-libraryseedancevisual-bookmarksweb-clipper
import test from "node:test";
import assert from "node:assert/strict";
import { aiConfigurationFromStorage, resolveTextTaskSettings, resolveVisionTaskSettings, resolveVideoAnalysisTask } from "../extension/ai-runtime.js";
import { analyzeTextDetailedWithDeepSeek } from "../extension/deepseek.js";
import { analyzeImageWithVision } from "../extension/vision.js";
import { analyzeVideoWithChatCompletions } from "../extension/video-analysis.js";
import { createDefaultFacetCatalog } from "../extension/facets.js";
const model = "glm-5.3-flash";
const tasks = ["textTags", "imageAnalysis", "videoAnalysis"];
const configuration = aiConfigurationFromStorage({
aiProviderRegistry: { providers: { zhipu: {
apiKey: "offline-fixture-not-a-key", consent: true,
models: Object.fromEntries(tasks.map(task => [task, model])),
discoveredModels: [{ id: model, status: "available", confidence: "declared", tasks }]
} } },
aiTaskAssignments: Object.fromEntries(tasks.map(task => [task, { providerId: "zhipu", model }]))
});
const catalog = createDefaultFacetCatalog();
const routes = {
text: fetchImpl => analyzeTextDetailedWithDeepSeek({ text: "离线文字夹具" }, catalog, resolveTextTaskSettings("textTags", configuration), fetchImpl),
image: fetchImpl => analyzeImageWithVision({ settings: resolveVisionTaskSettings("imageAnalysis", configuration), catalog,
imageDataUrl: "data:image/png;base64,AQID" }, fetchImpl),
video: fetchImpl => analyzeVideoWithChatCompletions({ ...resolveVideoAnalysisTask(configuration), catalog,
mode: "visual-reconstruction", requestId: "offline-budget-fixture", videoBlob: new Blob([new Uint8Array([1, 2, 3])], { type: "video/mp4" }) }, { fetchImpl })
};
for (const [kind, run] of Object.entries(routes)) {
test(`${kind}: GLM sends its documented output capacity without weakening thinking or retrying truncation`, async () => {
const requests = [];
// Synthetic transport boundary: this is not a real model-quality or token-usage receipt.
await assert.rejects(run(async (_url, options) => {
requests.push(JSON.parse(options.body));
return new Response(JSON.stringify({ choices: [{ finish_reason: "length", message: { content: "离线截断夹具" } }] }), { status: 200 });
}), error => error.code === "output_truncated");
assert.equal(requests.length, 1, "a provider truncation must not trigger another billable request");
const request = requests[0];
// Independent source: https://docs.bigmodel.cn/cn/guide/models/vlm/glm-5.3-flash (128K).
assert.equal(request.max_tokens, 131072);
assert.equal(request.model, model);
assert.notEqual(request.thinking?.type, "disabled");
assert.ok(!request.reasoning_effort || request.reasoning_effort === "max");
if (kind === "image") assert.ok(request.messages.some(message => Array.isArray(message.content) && message.content.some(part => part.type === "image_url")));
if (kind === "video") assert.equal(request.messages[0].content.find(part => part.type === "video_url").video_url.url, "AQID");
});
}