Skip to content

A local-first multi-agent app development platform, built on PenguinHarness

Draft-screen examples: a rhythm game, an investment Copilot, and a scheduled-tasks folder (#399)

Yaowei Zheng07046e8ec8a5
diff --git a/changelog/unreleased/2026-08-21-draft-examples.md b/changelog/unreleased/2026-08-21-draft-examples.md
new file mode 100644
index 00000000..0ac6c538
--- /dev/null
+++ b/changelog/unreleased/2026-08-21-draft-examples.md
@@ -0,0 +1,26 @@
+# Draft-screen examples: a rhythm game, an investment Copilot, and a scheduled-tasks folder
+
+- **Date:** 2026-08-21
+- **Type:** feature
+- **Scope:** `web`
+- **PR:** [#399](https://github.com/Prism-Shadow/penguin-harness/pull/399)
+
+[中文版](2026-08-21-draft-examples.zh.md)
+
+The home screen's example catalog grows by five cards and one folder. Build web apps gains a Muse Dash-style rhythm runner; Build and optimize agents gains a conversational investment analyst as its first entry; and a new Create scheduled tasks folder holds three examples that set up real schedules.
+
+Each of the five is a one- or two-sentence request rather than a full working brief — what to build and the constraints without which the result would be the wrong thing, with the file layouts, field lists and step-by-step headings left for the Agent to work out or ask about.
+
+## The new examples
+
+- **Rhythm runner mini game** (Build web apps, `web-design`) — a Muse Dash-style rhythm runner: a penguin runs forward on its own, notes drawn as music-note icons fly in locked to the beat, and each hit is graded Perfect / Great / Miss with a combo score.
+- **Conversational investment analyst** (Build and optimize agents, `penguin-sdk` + `web-design`) — a stock-market Copilot on the Penguin SDK, modelled on perplexity.ai/finance, that pulls live market data every five minutes from startup and lists the strongest-trending stocks and sector comparison on its home page, with the evidence behind every call. Its stock-lookup tool takes a company name rather than only a ticker, and says so when a company has no match or is not listed instead of inventing a quote. Framed as analysis of public data, not investment advice.
+- **Create scheduled tasks** (new folder, no pinned skills) — three schedules, each saying when it fires:
+ - *A 9am daily planning check-in* — every day at 9am, in the same conversation, planning the day and reviewing yesterday's progress.
+ - *Daily GitHub project digest* — every morning, one repo's issues, PRs and CI, ending in recommendations ranked by priority.
+ - *Friday memory review* — every Friday evening, in the same conversation, going through what is worth remembering from the week and writing the confirmed items into Memory.
+
+## Layout and tests
+
+- Folders are no longer all the same length (4 / 4 / 3), so opening the schedules folder moves what sits below the block by one row. The registry docstring and the draft view's layout comments now state the rule that actually holds — folders stay within one row of each other — and `example-tasks.test.ts` enforces it.
+- The same test now also checks that every folder and example has non-empty copy in both dictionaries, that the five new prompts stay under their length ceiling and keep their core (the rhythm game synced to music; the Copilot's Penguin SDK, refresh-from-startup, home-page trending list and name-resolving lookup tool; each schedule's trigger time and, for two of them, that they run in the current conversation), and that the investment brief keeps its not-advice framing.
diff --git a/changelog/unreleased/2026-08-21-draft-examples.zh.md b/changelog/unreleased/2026-08-21-draft-examples.zh.md
new file mode 100644
index 00000000..3a2fb6d2
--- /dev/null
+++ b/changelog/unreleased/2026-08-21-draft-examples.zh.md
@@ -0,0 +1,26 @@
+# 首页示例:音游、投资 Copilot 与定时任务分类
+
+- **Date:** 2026-08-21
+- **Type:** feature
+- **Scope:** `web`
+- **PR:** [#399](https://github.com/Prism-Shadow/penguin-harness/pull/399)
+
+[English](2026-08-21-draft-examples.md)
+
+首页示例目录新增五张卡片与一个分类。「搭建网页应用」新增一个喵斯快跑式的节奏跑酷游戏;「搭建和优化智能体」的第一条新增对话式投资分析助理;新分类「创建定时任务」下三条示例创建的是真实的定时任务。
+
+五条示例的提示词都是一两句话的请求,而不是完整的工作简报:只写要做什么、以及少了就会做错的约束;文件结构、字段清单与分步小节交给 Agent 自行判断或来问。
+
+## 新增示例
+
+- **音乐节奏跑酷小游戏**(搭建网页应用,`web-design`)——喵斯快跑式的节奏跑酷:企鹅主角自动向前跑,音符画成音符图标、严格踩着节拍飞来,每次击打给出 Perfect / Great / Miss 判定与连击计分。
+- **对话式投资分析助理**(搭建和优化智能体,`penguin-sdk` + `web-design`)——基于 Penguin SDK、形态参考 perplexity.ai/finance 的股市 Copilot:启动后每 5 分钟实时抓取行情,首页直接列出近期走势较好的股票与板块强弱,每个判断都给出依据;查股工具支持按公司名(中文亦可)提问,查不到或未上市就直说而不是编一个报价。整体定位为基于公开数据的分析,而非投资建议。
+- **创建定时任务**(新分类,不固定技能)——三条定时任务,各自写明触发时间:
+ - *每天早 9 点的计划对话*——每天早上 9 点在当前会话里聊当天计划,并回顾昨天的进展。
+ - *每天汇总 GitHub 项目状态*——每天早上汇总一个仓库的 Issue、PR 与 CI,结尾给出按优先级排序的建议。
+ - *每周五晚回顾并记录 Memory*——每周五傍晚在当前会话里过一遍这周值得长期记住的事,确认后写进 Memory。
+
+## 布局与测试
+
+- 各分类不再等长(4 / 4 / 3),打开定时任务分类时其下方内容会移动一行。注册表文档注释与首页布局注释改为陈述当前真正成立的规则——各分类长度相差不超过一行——并由 `example-tasks.test.ts` 守住。
+- 该测试同时检查:每个分类与示例在两份词典里都有非空文案;五条新提示词不超过长度上限且保留各自的核心(音游与音乐同步;Copilot 的 Penguin SDK、启动后周期刷新、首页走势榜与按公司名查询的查股工具;每条定时任务的触发时间,其中两条还要落在当前会话);投资示例保留「不是投资建议」的定位。
diff --git a/packages/web/src/features/chat/draft-view.tsx b/packages/web/src/features/chat/draft-view.tsx
index 7c5009a8..8bb8e11c 100644
--- a/packages/web/src/features/chat/draft-view.tsx
+++ b/packages/web/src/features/chat/draft-view.tsx
@@ -124,12 +124,15 @@ function saveAppliedRouteKey(field: RouteStateField, key: string): void {
* Agents entry uses (NAV_ICONS.agents) — deliberately not a generic refresh loop, because the
* app already has one glyph that means "agent" and a folder of agent examples should wear it.
* Duplicated as a literal rather than imported: sidebar.tsx imports from chat-page.tsx, which
- * renders this file, so importing it back would close an import cycle.
+ * renders this file, so importing it back would close an import cycle. schedules: a clock face
+ * with hands — the plainest mark for "fires on a timer", and distinct from the hourglass that
+ * already means a Session is waiting.
*/
const FOLDER_GLYPHS: Record<ExampleFolderId, string> = {
webapps:
"M3 6a2 2 0 0 1 2-2h14a2 2 0 0 1 2 2v12a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2V6zM3 9h18M6 6.5h.01M9 6.5h.01",
agents: "M12 3v3m-6 4a6 6 0 0 1 12 0v5a3 3 0 0 1-3 3H9a3 3 0 0 1-3-3v-5zm3 3h.01M15 13h.01",
+ schedules: "M12 2a10 10 0 1 0 0 20 10 10 0 1 0 0-20M12 6.5V12l3.5 2",
};
export function DraftView({
@@ -687,9 +690,10 @@ export function DraftView({
/**
* The open example folder — bookmark-style, and ALWAYS exactly one: selecting another closes
* the previous, and clicking the open one is a no-op rather than collapsing it. Never
- * nullable on purpose. With every folder the same length, "one open" is what makes the
- * block's height a constant: the examples area can neither collapse to bare folder rows nor
- * grow, so nothing below it shifts as folders are switched.
+ * nullable on purpose. With the folders kept within one row of each other, "one open" is
+ * what keeps the block's height near-constant: the examples area can neither collapse to
+ * bare folder rows nor grow to the whole catalog, so switching folders moves what sits
+ * below it by at most one row.
*/
const [openFolder, setOpenFolder] = useState<ExampleFolderId>(EXAMPLE_FOLDERS[0].id);
@@ -763,13 +767,14 @@ export function DraftView({
{/* Example tasks: one-click canned builds showing off the one-sentence → app flow.
Bookmark-style folders with ALWAYS exactly one open — selecting another closes the
- previous, and the open one cannot be collapsed. The block is therefore a FIXED
- height: two folder rows plus one folder's rows, whichever folder that is (they are
- kept the same length). Nothing below shifts when folders are switched, and no
- scroll container is needed — a scrollbar inside a six-line showcase reads as a
- defect. Each example is a single-line title; its one-sentence description rides in
- the row tooltip rather than a second line. Rows are disabled until
- agents/models/skills are resolved (onSend would silently no-op without an Agent). */}
+ previous, and the open one cannot be collapsed. The block is therefore three
+ folder rows plus one folder's rows, with the folders kept within one row of each
+ other (3–4 examples each), so switching folders moves what sits below by at most
+ one row and no scroll container is needed — a scrollbar inside a seven-line
+ showcase reads as a defect. Each example is a single-line title; its one-sentence
+ description rides in the row tooltip rather than a second line. Rows are disabled
+ until agents/models/skills are resolved (onSend would silently no-op without an
+ Agent). */}
<div className="mt-6 space-y-1">
{EXAMPLE_FOLDERS.map((folder) => {
const open = folder.id === openFolder;
diff --git a/packages/web/src/features/chat/example-tasks.ts b/packages/web/src/features/chat/example-tasks.ts
index ebe02574..c372acc2 100644
--- a/packages/web/src/features/chat/example-tasks.ts
+++ b/packages/web/src/features/chat/example-tasks.ts
@@ -6,12 +6,15 @@
* folder plus the open folder's own rows, never the whole catalog. Adding an example means
* appending it to the folder it belongs to, not lengthening the page.
*
- * Keep the folders similarly sized: the draft page reserves no scroll area for this block, so
- * a folder much longer than its siblings is what would make the height jump between them.
+ * Keep the folders within one row of each other (3–4 examples each today): the draft page
+ * reserves no scroll area for this block, so its height is the folder rows plus the open
+ * folder's rows — a folder much longer than its siblings is what would make that height jump
+ * as folders are switched.
*
- * Copy and full prompts live in the active locale dictionary at `S.chat.exampleFolders[id]`
- * and `S.chat.exampleTasks[id]`. Skills listed here are pinned only when the selected Agent has
- * them installed; an empty list sends the prompt unchanged.
+ * Copy and prompts live in the active locale dictionary at `S.chat.exampleFolders[id]` and
+ * `S.chat.exampleTasks[id]`, which is also where the rule for how long a prompt may get is
+ * stated. Skills listed here are pinned only when the selected Agent has them installed; an
+ * empty list sends the prompt unchanged.
*/
export const EXAMPLE_FOLDERS = [
{
@@ -20,16 +23,26 @@ export const EXAMPLE_FOLDERS = [
{ id: "game", skills: ["web-design"] },
{ id: "gamecenter", skills: ["web-design"] },
{ id: "lol", skills: ["web-design"] },
+ { id: "rhythmRunner", skills: ["web-design"] },
],
},
{
id: "agents",
tasks: [
+ { id: "investmentCopilot", skills: ["penguin-sdk", "web-design"] },
{ id: "rag", skills: ["penguin-sdk", "web-design"] },
{ id: "agentBenchmarkBuild", skills: [] },
{ id: "agentOptimization", skills: [] },
],
},
+ {
+ id: "schedules",
+ tasks: [
+ { id: "dailyPlan", skills: [] },
+ { id: "githubDigest", skills: [] },
+ { id: "memoryReview", skills: [] },
+ ],
+ },
] as const;
export type ExampleFolder = (typeof EXAMPLE_FOLDERS)[number];
diff --git a/packages/web/src/lib/strings-en.ts b/packages/web/src/lib/strings-en.ts
index 043f3208..2fcdb6a5 100644
--- a/packages/web/src/lib/strings-en.ts
+++ b/packages/web/src/lib/strings-en.ts
@@ -1164,6 +1164,7 @@ export const en: Strings = {
exampleFolders: {
webapps: "Build web apps",
agents: "Build and optimize agents",
+ schedules: "Create scheduled tasks",
},
exampleTasks: {
game: {
@@ -1231,6 +1232,28 @@ Penguin visual style (see the web-design skill), dark by default. On phones the
When done, open index.html in a browser and self-test once.`,
},
+ rhythmRunner: {
+ label: "Rhythm runner mini game",
+ desc: "A Muse Dash-style rhythm runner: a penguin lead, notes locked to the beat, graded Perfect / Great / Miss",
+ prompt:
+ "Build a Muse Dash-style rhythm runner: a penguin runs forward on its own, and notes — " +
+ "drawn as music-note icons — fly in locked to the beat for me to hit. Show Perfect / " +
+ "Great / Miss for each hit, score combos, and let the difficulty climb as the track goes " +
+ "on. Pure front end, one file, playable straight from file://.",
+ },
+ investmentCopilot: {
+ label: "Conversational investment analyst",
+ desc: "A conversational market Copilot on the Penguin SDK: the home page lists what is trending, with the market factors behind every call",
+ prompt:
+ "Build a conversational stock-market Copilot on the Penguin SDK, along the lines of " +
+ "perplexity.ai/finance: from startup it pulls live market data every 5 minutes, and the " +
+ "home page lists the stocks trending strongest lately alongside how the sectors compare. " +
+ "Every call has to name the market factors behind it — policy, sector news, fund flows, " +
+ "earnings, macro data — rather than technical indicators. Analysis of public data, not " +
+ "investment advice. Its stock-lookup tool has to handle questions like \"look up Zhipu's " +
+ 'stock for me": resolve a company name to its ticker itself, and when there is no match ' +
+ "or the company is not listed, say so rather than inventing a quote.",
+ },
rag: {
label: "Build a Claude Code docs RAG agent",
desc: "Collect the claude-code-docs repo into a conversational RAG knowledge app with source citations",
@@ -1277,6 +1300,32 @@ Scenarios:
- desired_score: \`>=95\`
- candidate_round_limit: \`5\``,
},
+ dailyPlan: {
+ label: "A 9am daily planning check-in",
+ desc: "09:00 every day: talk through the day's plan in this same chat, and review yesterday's progress",
+ prompt:
+ "Set up a scheduled task: every day at 9am, in this same conversation, plan today's work " +
+ "with me. Read back over the conversation first and say what yesterday's plan got done and " +
+ "where it stuck, then offer an ordered shortlist for today with a line of reasoning each, " +
+ "and write up what I confirm as a checklist.",
+ },
+ githubDigest: {
+ label: "Daily GitHub project digest",
+ desc: "A daily pass over one repo's issues, PRs and CI, ending in prioritized recommendations",
+ prompt:
+ "Set up a scheduled task: every morning, use gh to digest one GitHub repo's issues, PRs " +
+ "and CI — surface what has stalled, what is waiting on review and what is failing — and " +
+ "end with recommendations ranked by priority, each saying why it sits where it does.",
+ },
+ memoryReview: {
+ label: "Friday memory review",
+ desc: "Friday evening: go through what is worth remembering from the week and write it into Memory",
+ prompt:
+ "Set up a scheduled task: every Friday evening, in this same conversation, go through " +
+ "what is worth remembering from the week with me. Check the existing memory index first so " +
+ "nothing is duplicated, then take it item by item — what to record, what to revise — and " +
+ "write what I confirm into Memory.",
+ },
},
sessionList: "Sessions",
defaultSessionTitle: "New chat",
diff --git a/packages/web/src/lib/strings.ts b/packages/web/src/lib/strings.ts
index 9cf60fab..1ac3015c 100644
--- a/packages/web/src/lib/strings.ts
+++ b/packages/web/src/lib/strings.ts
@@ -1126,11 +1126,15 @@ export const zh = {
exampleFolders: {
webapps: "搭建网页应用",
agents: "搭建和优化智能体",
+ schedules: "创建定时任务",
},
/**
- * Example task cards on the draft screen: one click auto-submits the canned prompt. These
- * are the FULL working prompts — descriptions stay short, but the submitted instructions
- * remain detailed because execution quality depends on them.
+ * Example task cards on the draft screen: one click loads the canned prompt for the user.
+ * Keep a prompt SHORT — a short paragraph, around 100 Chinese characters, carrying what to
+ * build plus the constraints the result would be wrong without. File layouts, field lists,
+ * step-by-step headings and self-test instructions are what the Agent works out or asks
+ * about, so they stay out; the older briefs below are still far longer and are being trimmed
+ * to match.
*/
exampleTasks: {
game: {
@@ -1195,6 +1199,24 @@ Penguin 视觉风格(见 web-design 技能),默认深色。手机端侧边
完成后在浏览器打开 index.html 自测一次。`,
},
+ rhythmRunner: {
+ label: "音乐节奏跑酷小游戏",
+ desc: "喵斯快跑式的音乐节奏跑酷:企鹅主角,音符踩着节拍飞来,判定分 Perfect / Great / Miss",
+ prompt:
+ "做一个喵斯快跑(Muse Dash)式的音乐节奏跑酷小游戏:主角是一只企鹅,自动向前跑;" +
+ "音符画成音符图标,严格踩着节拍飞来,玩家按键击打,判定显示 Perfect / Great / Miss 三档," +
+ "连击计分,难度随曲子推进。纯前端单文件,file:// 直接打开即玩。",
+ },
+ investmentCopilot: {
+ label: "对话式投资分析助理",
+ desc: "用 Penguin SDK 做对话式看盘 Copilot:首页列出近期走势较好的股票,每个判断都说清市场因素",
+ prompt:
+ "用 Penguin SDK 做一个对话式的股市 Copilot,形态参考 perplexity.ai/finance:启动后每 5 分钟实时抓取大盘行情," +
+ "首页直接列出近期走势较好的股票和板块强弱,每个判断都要说清背后的市场因素——政策、行业消息、" +
+ "资金流向、财报或宏观数据,而不是技术指标," +
+ "只做分析不是投资建议。它的查股工具要能答「帮我查一下智谱的股票」这类问题:" +
+ "按公司名(中文也行)自己对应到股票代码,查不到或没上市就直说,不要编。",
+ },
rag: {
label: "构建 Claude Code 文档 RAG 智能体",
desc: "收集 claude-code-docs 仓库,生成可对话、带来源引用的 RAG 知识应用",
@@ -1240,6 +1262,28 @@ Benchmark:
- desired_score:\`>=95\`
- candidate_round_limit:\`5\``,
},
+ dailyPlan: {
+ label: "每天早 9 点的计划对话",
+ desc: "每天 09:00 在同一个会话里聊当天计划,并回顾昨天的进展",
+ prompt:
+ "建一个定时任务:每天早上 9 点在这个会话里和我聊今天的计划。" +
+ "先回看上文说清昨天定的事做完了多少、哪些卡住,再给我一份排好序的今日候选、每条一句理由," +
+ "我确认后写成清单。",
+ },
+ githubDigest: {
+ label: "每天汇总 GitHub 项目状态",
+ desc: "定时跑一遍仓库的 Issue、PR 与 CI,日报结尾给出按优先级排序的建议",
+ prompt:
+ "建一个定时任务:每天早上用 gh 汇总一个 GitHub 仓库的 Issue、PR 与 CI 状态," +
+ "挑出停滞的、待评审的和挂掉的,结尾给出按优先级排序的建议,每条说清为什么排在这个位置。",
+ },
+ memoryReview: {
+ label: "每周五晚回顾并记录 Memory",
+ desc: "周五傍晚一起过一遍这周值得长期记住的事,确认后由你写进 Memory",
+ prompt:
+ "建一个定时任务:每周五傍晚在这个会话里和我过一遍这周值得长期记住的事。" +
+ "先看已有记忆索引避免重复,再逐条问我该记什么、要不要改已有的,我确认后你写进 Memory。",
+ },
},
sessionList: "Session",
defaultSessionTitle: "新对话",
diff --git a/packages/web/test/example-tasks.test.ts b/packages/web/test/example-tasks.test.ts
index 5f394a01..bed2fb3d 100644
--- a/packages/web/test/example-tasks.test.ts
+++ b/packages/web/test/example-tasks.test.ts
@@ -1,5 +1,5 @@
import { describe, expect, it } from "vitest";
-import { EXAMPLE_TASKS } from "../src/features/chat/example-tasks";
+import { EXAMPLE_FOLDERS, EXAMPLE_TASKS } from "../src/features/chat/example-tasks";
import { buildSkillsMessage } from "../src/features/chat/skill-use";
import { en } from "../src/lib/strings-en";
import { zh } from "../src/lib/strings";
@@ -101,3 +101,155 @@ describe("draft example tasks", () => {
},
);
});
+
+describe("draft example catalog", () => {
+ it("gives every folder and every example copy in both dictionaries", () => {
+ const ids = EXAMPLE_TASKS.map((task) => task.id);
+ expect(new Set(ids).size).toBe(ids.length);
+ for (const dict of [zh, en]) {
+ for (const folder of EXAMPLE_FOLDERS) {
+ expect(dict.chat.exampleFolders[folder.id]).not.toBe("");
+ }
+ for (const id of ids) {
+ const copy = dict.chat.exampleTasks[id];
+ expect(copy.label).not.toBe("");
+ expect(copy.desc).not.toBe("");
+ // The prompt is the request; the description is only the row's tooltip for it.
+ expect(copy.prompt.length).toBeGreaterThan(copy.desc.length);
+ }
+ }
+ });
+
+ // The draft page reserves no scroll area: its height is the folder rows plus the open
+ // folder's rows, so a folder much longer than its siblings makes that height jump.
+ it("keeps the folders within one row of each other", () => {
+ const sizes = EXAMPLE_FOLDERS.map((folder) => folder.tasks.length);
+ expect(Math.max(...sizes) - Math.min(...sizes)).toBeLessThanOrEqual(1);
+ });
+});
+
+describe("scheduled-task examples", () => {
+ const scheduleIds = ["dailyPlan", "githubDigest", "memoryReview"] as const;
+
+ it("files them all in the schedules folder", () => {
+ const folder = EXAMPLE_FOLDERS.find((candidate) => candidate.id === "schedules");
+ expect(folder?.tasks.map((task) => task.id)).toEqual([...scheduleIds]);
+ });
+
+ // Two things a shortened schedule brief cannot lose. WHEN it fires: a schedule with no time
+ // in it is not a schedule, and the Agent has nothing to ask about it that would not be a
+ // guess. And, for the two check-ins that need the user to answer, that they run in the
+ // conversation the user is already in — the default is a fresh Session per run, which for
+ // those two would be wrong. How the schedule is written (TOML under agent_state/schedule/,
+ // `enabled`, `start_at`) is in the Agent's own Scheduled Tasks section; the prompt no longer
+ // repeats it. `period` is a fixed interval (30m / 12h / 7d), so cron syntax is always wrong.
+ it.each([
+ {
+ id: "dailyPlan",
+ zhMarkers: ["每天", "9 点", "这个会话"],
+ enMarkers: ["every day", "9am", "this same conversation"],
+ },
+ { id: "githubDigest", zhMarkers: ["每天"], enMarkers: ["every morning"] },
+ {
+ id: "memoryReview",
+ zhMarkers: ["每周五", "这个会话"],
+ enMarkers: ["every Friday", "this same conversation"],
+ },
+ ] as const)("$id says when it fires in both locales", ({ id, zhMarkers, enMarkers }) => {
+ const zhPrompt = zh.chat.exampleTasks[id].prompt;
+ const enPrompt = en.chat.exampleTasks[id].prompt;
+ for (const marker of zhMarkers) {
+ expect(zhPrompt).toContain(marker);
+ }
+ for (const marker of enMarkers) {
+ expect(enPrompt).toContain(marker);
+ }
+ expect(zhPrompt).not.toContain("cron");
+ expect(enPrompt).not.toContain("cron");
+ });
+});
+
+/**
+ * These five are written to be read by the user before they run, so each is a short paragraph —
+ * around 100 Chinese characters — carrying what to build plus the constraints the result would be
+ * wrong without, and none of the step-by-step detail the Agent works out or asks about. The caps
+ * below are a ceiling against re-inflation, not a target. Only these five are covered: the older
+ * briefs in the catalog are still far longer and are being shortened separately.
+ */
+describe("short example prompts", () => {
+ const shortExamples = [
+ {
+ id: "rhythmRunner",
+ zhMarkers: ["喵斯快跑", "企鹅", "音符图标", "节拍", "Perfect", "Great", "Miss"],
+ enMarkers: ["Muse Dash", "penguin", "music-note icons", "beat", "Perfect", "Great", "Miss"],
+ },
+ {
+ id: "investmentCopilot",
+ zhMarkers: [
+ "Penguin SDK",
+ "perplexity.ai/finance",
+ "对话式",
+ "启动后",
+ "实时抓取",
+ "首页",
+ "走势较好的股票",
+ "市场因素",
+ "查股工具",
+ "公司名",
+ "不要编",
+ ],
+ enMarkers: [
+ "Penguin SDK",
+ "perplexity.ai/finance",
+ "conversational",
+ "from startup",
+ "live market",
+ "home page",
+ "trending strongest",
+ "market factors",
+ "stock-lookup tool",
+ "company name",
+ "not listed",
+ ],
+ },
+ { id: "dailyPlan", zhMarkers: ["计划"], enMarkers: ["plan"] },
+ { id: "githubDigest", zhMarkers: ["GitHub", "优先级"], enMarkers: ["GitHub", "priority"] },
+ { id: "memoryReview", zhMarkers: ["Memory"], enMarkers: ["Memory"] },
+ ] as const;
+
+ it.each(shortExamples)("$id stays within the length ceiling", ({ id }) => {
+ // investmentCopilot sits highest by a wide margin: seven requirements plus a reference URL and
+ // a quoted sample question, both literals no rewording compresses. The English ceiling is the
+ // wider one because English runs about three times the character count of the same Chinese,
+ // not because the English is allowed to say more.
+ expect(zh.chat.exampleTasks[id].prompt.length).toBeLessThanOrEqual(230);
+ expect(en.chat.exampleTasks[id].prompt.length).toBeLessThanOrEqual(700);
+ });
+
+ // What is left after the cut: what to build, and the constraints without which the result
+ // would be the wrong thing (a rhythm game is a game synced to music; the investment Copilot
+ // is a Penguin SDK app that refreshes from startup, surfaces trending stocks on its home page,
+ // and gives the assistant a name-resolving stock-lookup tool).
+ it.each(shortExamples)("$id keeps its core in both locales", ({ id, zhMarkers, enMarkers }) => {
+ for (const marker of zhMarkers) {
+ expect(zh.chat.exampleTasks[id].prompt).toContain(marker);
+ }
+ for (const marker of enMarkers) {
+ expect(en.chat.exampleTasks[id].prompt).toContain(marker);
+ }
+ });
+});
+
+describe("investment copilot example", () => {
+ // Short as it is, the brief stays analysis with stated evidence, never a call to act.
+ it.each([
+ { locale: "zh", prompt: zh.chat.exampleTasks.investmentCopilot.prompt, marker: "不是投资建议" },
+ {
+ locale: "en",
+ prompt: en.chat.exampleTasks.investmentCopilot.prompt,
+ marker: "not investment advice",
+ },
+ ])("$locale keeps the not-advice framing", ({ prompt, marker }) => {
+ expect(prompt).toContain(marker);
+ });
+});