280 lines
8.7 KiB
TypeScript
280 lines
8.7 KiB
TypeScript
import { describe, expect, it } from "vitest";
|
|
import {
|
|
DICTATE_MODULE_READ_STATE_TAG,
|
|
buildDictateModuleReadIndex,
|
|
runDictateTurn,
|
|
} from "../src/dictate/index.js";
|
|
import { loadModuleCatalog } from "../src/skills/creation-flow.js";
|
|
import type { AgentDriver } from "../src/runtime/driver.js";
|
|
|
|
async function setup() {
|
|
const catalog = await loadModuleCatalog("dialogue/world-simulator");
|
|
expect(catalog).toBeTruthy();
|
|
const moduleReadIndex = await buildDictateModuleReadIndex({
|
|
skillPackRoot: "dialogue/world-simulator",
|
|
catalog: catalog!,
|
|
});
|
|
return { catalog: catalog!, moduleReadIndex };
|
|
}
|
|
|
|
describe("dictate read_module", () => {
|
|
it("rejects insert before reading the current prompt version", async () => {
|
|
const { catalog, moduleReadIndex } = await setup();
|
|
const tags = new Map<string, string>();
|
|
const driver: AgentDriver = {
|
|
async run(input) {
|
|
const outcome = await input.handleStep([
|
|
{
|
|
id: "i1",
|
|
name: "insert",
|
|
arguments: JSON.stringify({
|
|
position: "设计.美学纲领与交互范式",
|
|
content: "{}",
|
|
self_score: {
|
|
dims: [
|
|
{ name: "核心体验", score: 5 },
|
|
{ name: "背景与规则", score: 5 },
|
|
],
|
|
},
|
|
}),
|
|
},
|
|
]);
|
|
if (outcome.kind === "continue") {
|
|
const payload = JSON.parse(outcome.results[0]!.content);
|
|
expect(payload.error).toContain('read_module("aesthetics-interaction")');
|
|
}
|
|
return { iterations: 1, stop: { kind: "text", content: "未写入" } };
|
|
},
|
|
};
|
|
|
|
const result = await runDictateTurn({
|
|
llm: {} as never,
|
|
dialogue: [],
|
|
moduleCatalog: catalog,
|
|
moduleReadIndex,
|
|
driver,
|
|
handlers: handlers(tags),
|
|
});
|
|
expect(result.wroteTags).toEqual([]);
|
|
});
|
|
|
|
it("allows reading multiple modules then inserting both in the same step", async () => {
|
|
const { catalog, moduleReadIndex } = await setup();
|
|
const tags = new Map<string, string>();
|
|
const driver: AgentDriver = {
|
|
async run(input) {
|
|
const outcome = await input.handleStep([
|
|
{
|
|
id: "r1",
|
|
name: "read_module",
|
|
arguments: JSON.stringify({ module_id: "美学纲领与交互范式" }),
|
|
},
|
|
{
|
|
id: "r2",
|
|
name: "read_module",
|
|
arguments: JSON.stringify({ module_id: "protagonist" }),
|
|
},
|
|
{
|
|
id: "i1",
|
|
name: "insert",
|
|
arguments: JSON.stringify({
|
|
position: "设计.美学纲领与交互范式",
|
|
content: "{}",
|
|
self_score: {
|
|
dims: [
|
|
{ name: "核心体验", score: 5 },
|
|
{ name: "背景与规则", score: 3, gap: "规则因果仍需确认" },
|
|
],
|
|
},
|
|
}),
|
|
},
|
|
{
|
|
id: "i2",
|
|
name: "insert",
|
|
arguments: JSON.stringify({
|
|
position: "设计.主角设定",
|
|
content: "{}",
|
|
self_score: {
|
|
dims: [
|
|
{ name: "宽松度", score: 5 },
|
|
{ name: "必要度", score: 5 },
|
|
],
|
|
},
|
|
}),
|
|
},
|
|
]);
|
|
if (outcome.kind === "continue") {
|
|
const read = JSON.parse(outcome.results[0]!.content);
|
|
expect(read.task).toBeTruthy();
|
|
expect(read.version).toMatch(/^[a-f0-9]{16}$/);
|
|
expect(JSON.parse(outcome.results[2]!.content).ok).toBe(true);
|
|
expect(JSON.parse(outcome.results[3]!.content).ok).toBe(true);
|
|
}
|
|
return { iterations: 1, stop: { kind: "text", content: "已写入" } };
|
|
},
|
|
};
|
|
|
|
const result = await runDictateTurn({
|
|
llm: {} as never,
|
|
dialogue: [],
|
|
moduleCatalog: catalog,
|
|
moduleReadIndex,
|
|
driver,
|
|
handlers: handlers(tags),
|
|
});
|
|
expect(result.wroteTags).toEqual([
|
|
"设计.美学纲领与交互范式",
|
|
"设计.主角设定",
|
|
]);
|
|
expect(tags.get(DICTATE_MODULE_READ_STATE_TAG)).toContain("protagonist");
|
|
});
|
|
|
|
it("returns presentation references for reply-format", async () => {
|
|
const { moduleReadIndex } = await setup();
|
|
const module = moduleReadIndex.get("reply-format");
|
|
expect(module?.presentation_references?.shells.length).toBeGreaterThan(0);
|
|
expect(module?.presentation_references?.markdown_safe_subset).toContain(
|
|
"正文安全子集",
|
|
);
|
|
expect(module?.presentation_references?.iframe_contract).toContain(
|
|
"present.onData",
|
|
);
|
|
});
|
|
|
|
it("requires another read when the prompt version changes", async () => {
|
|
const { catalog, moduleReadIndex } = await setup();
|
|
const tags = new Map<string, string>();
|
|
const driver: AgentDriver = {
|
|
async run(input) {
|
|
await input.handleStep([
|
|
{
|
|
id: "r1",
|
|
name: "read_module",
|
|
arguments: JSON.stringify({ module_id: "aesthetics-interaction" }),
|
|
},
|
|
]);
|
|
moduleReadIndex.get("aesthetics-interaction")!.version = "new-version";
|
|
const outcome = await input.handleStep([
|
|
{
|
|
id: "i1",
|
|
name: "insert",
|
|
arguments: JSON.stringify({
|
|
position: "设计.美学纲领与交互范式",
|
|
content: "{}",
|
|
}),
|
|
},
|
|
]);
|
|
if (outcome.kind === "continue") {
|
|
expect(JSON.parse(outcome.results[0]!.content).error).toContain(
|
|
"当前提示版本",
|
|
);
|
|
}
|
|
return { iterations: 2, stop: { kind: "text", content: "需重读" } };
|
|
},
|
|
};
|
|
const result = await runDictateTurn({
|
|
llm: {} as never,
|
|
dialogue: [],
|
|
moduleCatalog: catalog,
|
|
moduleReadIndex,
|
|
driver,
|
|
handlers: handlers(tags),
|
|
});
|
|
expect(result.wroteTags).toEqual([]);
|
|
});
|
|
|
|
it("validates score dimensions and stores low-score gaps as metadata", async () => {
|
|
const { catalog, moduleReadIndex } = await setup();
|
|
const tags = new Map<string, string>();
|
|
let metadata: unknown;
|
|
const driver: AgentDriver = {
|
|
async run(input) {
|
|
await input.handleStep([
|
|
{
|
|
id: "r1",
|
|
name: "read_module",
|
|
arguments: JSON.stringify({ module_id: "aesthetics-interaction" }),
|
|
},
|
|
]);
|
|
const rejected = await input.handleStep([
|
|
{
|
|
id: "i1",
|
|
name: "insert",
|
|
arguments: JSON.stringify({
|
|
position: "设计.美学纲领与交互范式",
|
|
content: "{}",
|
|
self_score: {
|
|
dims: [
|
|
{ name: "核心体验", score: 3 },
|
|
{ name: "背景与规则", score: 5 },
|
|
],
|
|
},
|
|
}),
|
|
},
|
|
]);
|
|
if (rejected.kind === "continue") {
|
|
expect(JSON.parse(rejected.results[0]!.content).error).toContain(
|
|
"必须写 gap",
|
|
);
|
|
}
|
|
await input.handleStep([
|
|
{
|
|
id: "i2",
|
|
name: "insert",
|
|
arguments: JSON.stringify({
|
|
position: "设计.美学纲领与交互范式",
|
|
content: "{}",
|
|
self_score: {
|
|
dims: [
|
|
{ name: "核心体验", score: 3, gap: "发生方式仍偏宽" },
|
|
{ name: "背景与规则", score: 5 },
|
|
],
|
|
},
|
|
}),
|
|
},
|
|
]);
|
|
return { iterations: 3, stop: { kind: "text", content: "已写入" } };
|
|
},
|
|
};
|
|
await runDictateTurn({
|
|
llm: {} as never,
|
|
dialogue: [],
|
|
moduleCatalog: catalog,
|
|
moduleReadIndex,
|
|
driver,
|
|
handlers: {
|
|
...handlers(tags),
|
|
writeProduct: (tag, content, order, meta) => {
|
|
tags.set(tag, content);
|
|
metadata = { order, ...meta };
|
|
},
|
|
},
|
|
});
|
|
expect(metadata).toMatchObject({
|
|
layer: "final",
|
|
selfScore: {
|
|
dims: expect.arrayContaining([
|
|
expect.objectContaining({ name: "核心体验", score: 3 }),
|
|
]),
|
|
},
|
|
});
|
|
});
|
|
});
|
|
|
|
function handlers(tags: Map<string, string>) {
|
|
return {
|
|
listProducts: () =>
|
|
[...tags.entries()]
|
|
.filter(([tag]) => tag.startsWith("设计.") || tag.startsWith("用户."))
|
|
.map(([tag, content]) => ({ tag, content })),
|
|
writeProduct: (tag: string, content: string) => tags.set(tag, content),
|
|
deleteProduct: (tag: string) => tags.delete(tag),
|
|
readTag: (tag: string) => tags.get(tag),
|
|
writeTag: (tag: string, content: string) => {
|
|
tags.set(tag, content);
|
|
},
|
|
deleteTag: (tag: string) => tags.delete(tag),
|
|
clearDialogue: () => undefined,
|
|
};
|
|
}
|