openclaw-voc-skill/claude-code/fmode-image-set/mcp/src/features/image-set/reference-first-visual-observer.test.mjs

import assert from "node:assert/strict";
import test from "node:test";
import { mkdtemp, rm, writeFile } from "node:fs/promises";
import { join } from "node:path";
import { tmpdir } from "node:os";
import { observeGeneratedReferences, observeOriginalVisual, ORIGINAL_OBSERVER_INSTRUCTION } from "./reference-first-visual-observer.mjs";

const product = { category: "鞋", form: "自然直立", colors: ["黑色"], material_and_light_cues: ["哑光鞋面"], identity_cues: ["白色鞋底", "黑色鞋带"], natural_support_and_use: ["地面穿着"] };

test("original observer makes one descriptive Qwen call and reuses by source hash", async () => {
  const directory = await mkdtemp(join(tmpdir(), "fmode-observer-"));
  try {
    const source = join(directory, "source.png"); const output = join(directory, "observation.json");
    await writeFile(source, "image"); let calls = 0;
    const options = { runId: "run-20260804210000-aabbcc", source, outputFile: output, config: { provider: "qwen", model: "qwen3-vl-plus", apiKey: "x" }, uploadImage: async () => ({ public_url: "https://image.test/source.png" }), requestJson: async ({ instruction }) => { calls += 1; assert.match(instruction, /不是创意策划器/); return { value: { product }, usage: { total_tokens: 10 } }; } };
    assert.equal((await observeOriginalVisual(options)).model_calls, 1);
    assert.equal((await observeOriginalVisual(options)).model_calls, 0);
    assert.equal(calls, 1); assert.match(ORIGINAL_OBSERVER_INSTRUCTION, /禁止选择页面模板/);
  } finally { await rm(directory, { recursive: true, force: true }); }
});

test("reference observer observes both assets in one model call", async () => {
  let calls = 0;
  const result = await observeGeneratedReferences({ runId: "run-20260804210000-aabbcc", sources: ["https://image.test/three.png", "https://image.test/board.png"], config: { provider: "qwen", model: "qwen3-vl-plus", apiKey: "x" }, requestJson: async ({ imageUrls }) => { calls += 1; assert.equal(imageUrls.length, 2); return { value: { assets: [{ product }, { product, actual_composition_and_scenes: ["四场景"] }] } }; } });
  assert.equal(result.model_calls, 1); assert.equal(calls, 1); assert.equal(result.observation.assets[1].asset_role, "setting_board_visual");
});