新建作品改为直接进入对话落盘，按本局创作方案写能力，并把楼层呈现和零层卡分开。
输入栏同时可以按分组切换模型。 Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Cursor
+155
-32
@@ -73,7 +73,7 @@ describe("dictate context", () => {
|
||||
expect(isDictateModeValue("recipe")).toBe(false);
|
||||
});
|
||||
|
||||
it("system prompt requires insert toolcall and relative order", () => {
|
||||
it("system prompt starts from a plan and keeps generic write tools", () => {
|
||||
const prompt = buildDictateSystemPrompt({
|
||||
recipeName: "数据化跑团体验",
|
||||
recipeBrief:
|
||||
@@ -81,9 +81,8 @@ describe("dictate context", () => {
|
||||
moduleGuide:
|
||||
"【能力 · 何时落盘】\n- 美学纲领与交互范式：…\n 落盘：insert 「设计.美学纲领与交互范式」\n 何时用：还不能回答站位",
|
||||
});
|
||||
expect(prompt).toContain("设计.正文组成");
|
||||
expect(prompt).toContain("设计.开场白");
|
||||
expect(prompt).toContain("输出.开场白");
|
||||
expect(prompt).toContain("creation-plan.v1");
|
||||
expect(prompt).toContain("设计.本局创作方案");
|
||||
expect(prompt).toContain("insert");
|
||||
expect(prompt).toContain("delete");
|
||||
expect(prompt).toContain("declare_variable");
|
||||
@@ -93,29 +92,15 @@ describe("dictate context", () => {
|
||||
expect(prompt).toContain("同 position 再 insert");
|
||||
expect(prompt).toContain("数据化跑团体验");
|
||||
expect(prompt).toMatch(/order/);
|
||||
expect(prompt).toContain("仅允许 toolcall");
|
||||
expect(prompt).toContain("设计.模仿范例");
|
||||
expect(prompt).toContain("设计.模仿要点");
|
||||
expect(prompt).toContain("设计.故事推进");
|
||||
expect(prompt).toContain("设计.叙事指南");
|
||||
expect(prompt).toContain("大纲扩写");
|
||||
expect(prompt).toContain("扮演停笔关掉");
|
||||
expect(prompt).toContain("只挂载这些产物");
|
||||
expect(prompt).toContain("【能力 · 何时落盘】");
|
||||
expect(prompt).toContain("设计.美学纲领与交互范式");
|
||||
expect(prompt).toContain("同轮可多次 insert");
|
||||
expect(prompt).toContain("不抢跑未齐条件的下游");
|
||||
expect(prompt).toContain("也不自动开跑 D");
|
||||
expect(prompt).toContain("快穿短局");
|
||||
expect(prompt).toContain("一核短局");
|
||||
expect(prompt).toMatch(/舞台骨架/);
|
||||
expect(prompt).toMatch(/覆盖 C/);
|
||||
expect(prompt).toContain("设计.主角设定");
|
||||
expect(prompt).toContain("reply_module");
|
||||
expect(prompt).toContain("对用户可见回复");
|
||||
expect(prompt).toContain("有 N 个 reply_module");
|
||||
expect(prompt).toContain("〔可反复〕须拆分");
|
||||
expect(prompt).toContain("设计.具体实例#");
|
||||
expect(prompt).toContain("通常 2～3 个");
|
||||
expect(prompt).toContain("给 2～3 个选项来问");
|
||||
expect(prompt).toContain("不是 DAG");
|
||||
expect(prompt).toContain("不是用户已选路径");
|
||||
expect(prompt).toContain("自然回应");
|
||||
expect(prompt).not.toContain("体验锚定（程序打分）");
|
||||
expect(prompt).not.toContain("## 本轮落什么");
|
||||
});
|
||||
|
||||
it("formats module catalog for dictate with land tags and when", async () => {
|
||||
@@ -209,13 +194,12 @@ describe("dictate context", () => {
|
||||
});
|
||||
const brief = index.get("设计.美学纲领与交互范式");
|
||||
expect(brief).toBeTruthy();
|
||||
expect(brief!).toContain("# 模块 · 美学纲领与交互范式");
|
||||
expect(brief!).toContain("回答时请附带本模块");
|
||||
expect(brief!).toContain("# 本轮关注 · 美学纲领与交互范式");
|
||||
expect(brief!).toContain("不要求固定标题");
|
||||
expect(brief!).toContain("本模块内容包括");
|
||||
expect(brief!).toContain("## 写入用户回复时做什么");
|
||||
expect(brief!).toContain("## 探测与追问（本模块口径）");
|
||||
expect(brief!).toContain("展示优于提问");
|
||||
expect(brief!).toMatch(/自评关注|交互范式|美学纲领/);
|
||||
expect(brief!).toContain("同轮多个产物时合并回应");
|
||||
expect(brief!).toMatch(/自评关注|核心体验|背景与规则/);
|
||||
|
||||
const alias = index.get("设计.开场白");
|
||||
expect(alias).toContain("开场白与开场变量");
|
||||
@@ -357,6 +341,145 @@ describe("dictate context", () => {
|
||||
expect(products.map((p) => p.tag)).toEqual(["设计.具体实例#lei-ying"]);
|
||||
});
|
||||
|
||||
it("splits a concrete-instance records array into one product per record", async () => {
|
||||
const products: { tag: string; content: string }[] = [];
|
||||
const body = JSON.stringify({
|
||||
schema: "context-fragment.v1",
|
||||
技能: "具体实例",
|
||||
brief: "十位目标",
|
||||
正文: {
|
||||
rule_id: "目标女配",
|
||||
batch_id: "target-npcs-roster",
|
||||
本批数量: 3,
|
||||
批次条件: ["待清算的恶女名录"],
|
||||
records: [
|
||||
{ 姓名: "苏晚", 身份: "同学" },
|
||||
{ 姓名: "林乔", 身份: "上司" },
|
||||
{ 姓名: "苏晚", 身份: "同名另一人" },
|
||||
],
|
||||
},
|
||||
});
|
||||
const driver: AgentDriver = {
|
||||
async run(input) {
|
||||
const outcome = await input.handleStep([
|
||||
{
|
||||
id: "c1",
|
||||
name: "insert",
|
||||
arguments: JSON.stringify({
|
||||
position: "设计.具体实例#target-npcs-roster",
|
||||
content: body,
|
||||
}),
|
||||
},
|
||||
]);
|
||||
expect(outcome.kind).toBe("continue");
|
||||
if (outcome.kind === "continue") {
|
||||
const payload = JSON.parse(outcome.results[0]!.content) as {
|
||||
ok?: boolean;
|
||||
split?: number;
|
||||
positions?: string[];
|
||||
};
|
||||
expect(payload.ok).toBe(true);
|
||||
expect(payload.split).toBe(3);
|
||||
expect(payload.positions).toEqual([
|
||||
"设计.具体实例#苏晚",
|
||||
"设计.具体实例#林乔",
|
||||
"设计.具体实例#苏晚-2",
|
||||
]);
|
||||
}
|
||||
return { iterations: 1, stop: { kind: "text", content: "已拆开" } };
|
||||
},
|
||||
};
|
||||
const result = await runDictateTurn({
|
||||
llm: {} as never,
|
||||
dialogue: [{ role: "user", text: "把这三位分开落盘" }],
|
||||
driver,
|
||||
repeatableFamilies: ["设计.具体实例"],
|
||||
handlers: {
|
||||
listProducts: () => products,
|
||||
writeProduct: (tag, content) => {
|
||||
products.push({ tag, content });
|
||||
},
|
||||
deleteProduct: () => false,
|
||||
readTag: () => undefined,
|
||||
writeTag: () => undefined,
|
||||
deleteTag: () => false,
|
||||
clearDialogue: () => undefined,
|
||||
},
|
||||
});
|
||||
expect(result.wroteTags).toEqual([
|
||||
"设计.具体实例#苏晚",
|
||||
"设计.具体实例#林乔",
|
||||
"设计.具体实例#苏晚-2",
|
||||
]);
|
||||
const first = JSON.parse(products[0]!.content) as {
|
||||
brief: string;
|
||||
正文: { 规则?: string; 名称?: string; 记录?: { 姓名?: string }; batch_id?: string };
|
||||
};
|
||||
expect(first.brief).toBe("苏晚");
|
||||
expect(first.正文.规则).toBe("目标女配");
|
||||
expect(first.正文.名称).toBe("苏晚");
|
||||
expect(first.正文.记录?.姓名).toBe("苏晚");
|
||||
expect(first.正文.batch_id).toBeUndefined();
|
||||
});
|
||||
|
||||
it("prepares runtime tools before persisting an opening", async () => {
|
||||
const events: string[] = [];
|
||||
const driver: AgentDriver = {
|
||||
async run(input) {
|
||||
const prepared = await input.handleStep([
|
||||
{
|
||||
id: "prepare",
|
||||
name: "prepare_opening",
|
||||
arguments: "{}",
|
||||
},
|
||||
]);
|
||||
expect(prepared.kind).toBe("continue");
|
||||
const outcome = await input.handleStep([
|
||||
{
|
||||
id: "opening",
|
||||
name: "insert",
|
||||
arguments: JSON.stringify({
|
||||
position: "设计.开场白",
|
||||
content: "@玩家推开门。",
|
||||
}),
|
||||
},
|
||||
]);
|
||||
expect(outcome.kind).toBe("continue");
|
||||
if (outcome.kind === "continue") {
|
||||
const payload = JSON.parse(outcome.results[0]!.content) as {
|
||||
runtime_prepared_before_opening?: boolean;
|
||||
preparation_mode?: string;
|
||||
};
|
||||
expect(payload.runtime_prepared_before_opening).toBe(true);
|
||||
expect(payload.preparation_mode).toBe("prepare_opening");
|
||||
}
|
||||
return {
|
||||
text: "开场已写入",
|
||||
iterations: 1,
|
||||
stop: { kind: "text", content: "开场已写入" },
|
||||
};
|
||||
},
|
||||
};
|
||||
|
||||
await runDictateTurn({
|
||||
llm: {} as never,
|
||||
dialogue: [{ role: "user", text: "生成开场" }],
|
||||
driver,
|
||||
handlers: {
|
||||
listProducts: () => [],
|
||||
prepareOpening: () => events.push("prepare"),
|
||||
writeProduct: () => events.push("write"),
|
||||
deleteProduct: () => false,
|
||||
readTag: () => undefined,
|
||||
writeTag: () => undefined,
|
||||
deleteTag: () => false,
|
||||
clearDialogue: () => undefined,
|
||||
},
|
||||
});
|
||||
|
||||
expect(events).toEqual(["prepare", "write"]);
|
||||
});
|
||||
|
||||
it("insert tool result includes reply_module from lookup", async () => {
|
||||
let capturedSpec: string | undefined;
|
||||
const driver: AgentDriver = {
|
||||
@@ -378,7 +501,7 @@ describe("dictate context", () => {
|
||||
reply_module?: string;
|
||||
};
|
||||
expect(payload.ok).toBe(true);
|
||||
expect(payload.reply_module).toContain("回答时请附带本模块");
|
||||
expect(payload.reply_module).toContain("自然回复");
|
||||
capturedSpec = payload.reply_module;
|
||||
}
|
||||
return {
|
||||
@@ -601,7 +724,7 @@ describe("dictate context", () => {
|
||||
};
|
||||
expect(a.reply_module).toContain("美学纲领");
|
||||
expect(b.reply_module).toContain("主角设定");
|
||||
expect(a.reply_hint).toMatch(/多个 reply_module|本步已带回模块/);
|
||||
expect(a.reply_hint).toMatch(/多个产物|本步已带回模块/);
|
||||
expect(b.reply_hint).toContain("本步已带回模块");
|
||||
}
|
||||
return {
|
||||
|
||||
Reference in New Issue
Block a user