u1s1-cli 0.14.0 → 0.16.0

package/bench/suite.json

@@ -0,0 +1,193 @@
+[
+  {
+    "name": "quick",
+    "description": "快速烟雾测试 — 几秒钟测完,日常切模型后随手验证",
+    "questions": [
+      {
+        "id": "hello",
+        "category": "基础",
+        "prompt": "只说一句「你好,我是 u1s1」就行,不要多余的话。",
+        "maxScore": 5,
+        "checks": [
+          { "type": "contains", "value": "你好", "ignoreCase": false, "points": 2, "description": "包含「你好」" },
+          { "type": "contains", "value": "u1s1", "ignoreCase": true, "points": 2, "description": "包含「u1s1」" },
+          { "type": "max_length", "value": 50, "points": 1, "description": "不超过 50 字(啰嗦扣分)" }
+        ]
+      },
+      {
+        "id": "reverse",
+        "category": "编码",
+        "prompt": "写一个 TypeScript 函数,把字符串里的单词顺序反转(不是字符反转)。示例: 'hello world' → 'world hello'。只给代码,不要解释。",
+        "maxScore": 10,
+        "checks": [
+          { "type": "contains_code_block", "value": true, "points": 2, "description": "包含代码块" },
+          { "type": "contains", "value": "reverse", "ignoreCase": false, "points": 2, "description": "函数名包含 reverse" },
+          { "type": "contains", "value": "split", "ignoreCase": false, "points": 2, "description": "用了 split" },
+          { "type": "contains", "value": "reverse(", "ignoreCase": false, "points": 2, "description": "调用了数组 reverse" },
+          { "type": "contains", "value": "join", "ignoreCase": false, "points": 2, "description": "用了 join" }
+        ]
+      }
+    ]
+  },
+  {
+    "name": "full",
+    "description": "全面评测 — 覆盖编程、推理、中文理解、指令遵循",
+    "questions": [
+      {
+        "id": "fizzbuzz",
+        "category": "编码",
+        "prompt": "用 TypeScript 写 fizzbuzz,从 1 到 100,3 的倍数打印 fizz,5 的倍数打印 buzz,同时是 3 和 5 的倍数打印 fizzbuzz。只给代码,不要解释。",
+        "maxScore": 10,
+        "checks": [
+          { "type": "contains_code_block", "value": true, "points": 1, "description": "包含代码" },
+          { "type": "contains", "value": "% 3", "ignoreCase": false, "points": 2, "description": "判断 3 的倍数" },
+          { "type": "contains", "value": "% 5", "ignoreCase": false, "points": 2, "description": "判断 5 的倍数" },
+          { "type": "contains", "value": "fizzbuzz", "ignoreCase": false, "points": 2, "description": "输出了 fizzbuzz" },
+          { "type": "contains", "value": "100", "ignoreCase": false, "points": 1.5, "description": "循环到 100" },
+          { "type": "not_contains", "value": "解释", "ignoreCase": true, "points": 1.5, "description": "没有多余解释" }
+        ]
+      },
+      {
+        "id": "sort-algo",
+        "category": "编码",
+        "prompt": "用 TypeScript 实现一个快速排序(原地排序,不创建新数组)。只给代码,不要解释。",
+        "maxScore": 10,
+        "checks": [
+          { "type": "contains_code_block", "value": true, "points": 1, "description": "包含代码" },
+          { "type": "contains", "value": "partition", "ignoreCase": false, "points": 2, "description": "实现了 partition 函数" },
+          { "type": "contains", "value": "pivot", "ignoreCase": false, "points": 2, "description": "选定了 pivot" },
+          { "type": "contains", "value": "quickSort", "ignoreCase": false, "points": 2, "description": "递归调用 quickSort" },
+          { "type": "not_contains", "value": "concat", "ignoreCase": false, "points": 2, "description": "不是非原地排序(没用 concat)" },
+          { "type": "not_contains", "value": "解释", "ignoreCase": true, "points": 1, "description": "没有多余解释" }
+        ]
+      },
+      {
+        "id": "react-component",
+        "category": "编码",
+        "prompt": "写一个 React 计数器组件(TypeScript),包含:+ 按钮、- 按钮、重置按钮,显示当前数值。用 useState。只给代码,不要解释。",
+        "maxScore": 10,
+        "checks": [
+          { "type": "contains_code_block", "value": true, "points": 1, "description": "包含代码" },
+          { "type": "contains", "value": "useState", "ignoreCase": false, "points": 2, "description": "用了 useState" },
+          { "type": "contains", "value": "button", "ignoreCase": true, "points": 1.5, "description": "包含按钮元素" },
+          { "type": "contains", "value": "+", "ignoreCase": false, "points": 1, "description": "有加号按钮" },
+          { "type": "contains", "value": "-", "ignoreCase": false, "points": 1, "description": "有减号按钮" },
+          { "type": "contains", "value": "重置", "ignoreCase": false, "points": 1.5, "description": "有重置按钮" },
+          { "type": "contains", "value": "React", "ignoreCase": true, "points": 1, "description": "导入 React" },
+          { "type": "contains", "value": "FC", "ignoreCase": false, "points": 1, "description": "用了 React.FC 类型" }
+        ]
+      },
+      {
+        "id": "regex",
+        "category": "编码",
+        "prompt": "写一个 JavaScript 正则表达式,匹配中国大陆手机号(11 位,1 开头,第二位 3-9)。只给正则,不要解释。",
+        "maxScore": 10,
+        "checks": [
+          { "type": "valid_regex", "value": true, "points": 3, "description": "正则语法有效" },
+          { "type": "contains", "value": "1[3-9]", "ignoreCase": false, "points": 3, "description": "匹配 1 开头+第二位 3-9" },
+          { "type": "contains", "value": "\\d{9}", "ignoreCase": false, "points": 2, "description": "匹配后 9 位数字" },
+          { "type": "contains", "value": "^", "ignoreCase": false, "points": 1, "description": "有开头锚点" },
+          { "type": "contains", "value": "$", "ignoreCase": false, "points": 1, "description": "有结尾锚点" }
+        ]
+      },
+      {
+        "id": "reasoning-1",
+        "category": "推理",
+        "prompt": "有三个箱子:一个只装苹果,一个只装橘子,一个混装。所有标签都贴错了。你只能从一个箱子里拿一个水果看,就能推断出所有箱子的正确内容。请问该从哪个箱子拿?为什么?",
+        "maxScore": 10,
+        "checks": [
+          { "type": "contains", "value": "混装", "ignoreCase": false, "points": 3, "description": "回答从混装箱拿" },
+          { "type": "contains", "value": "标签", "ignoreCase": false, "points": 2, "description": "提到标签贴错" },
+          { "type": "contains", "value": "推理", "ignoreCase": true, "points": 2, "description": "有推理过程" },
+          { "type": "min_length", "value": 100, "points": 3, "description": "解释够详细(>=100 字)" }
+        ]
+      },
+      {
+        "id": "reasoning-2",
+        "category": "推理",
+        "prompt": "一个人花 8 元买了一只鸡,9 元卖出,10 元买回,11 元卖出。他赚了多少钱?一步步算。",
+        "maxScore": 10,
+        "checks": [
+          { "type": "contains", "value": "2", "ignoreCase": false, "points": 4, "description": "答案正确(赚 2 元)" },
+          { "type": "contains", "value": "8", "ignoreCase": false, "points": 1.5, "description": "提到了 8 元买入" },
+          { "type": "contains", "value": "11", "ignoreCase": false, "points": 1.5, "description": "提到了 11 元卖出" },
+          { "type": "contains", "value": "利润", "ignoreCase": true, "points": 1.5, "description": "有利润计算过程" },
+          { "type": "min_length", "value": 80, "points": 1.5, "description": "有分步计算过程" }
+        ]
+      },
+      {
+        "id": "zh-explain",
+        "category": "中文",
+        "prompt": "用一句话给完全不懂编程的人解释什么是「递归」。说得通俗一点。",
+        "maxScore": 10,
+        "checks": [
+          { "type": "contains", "value": "套娃", "ignoreCase": false, "points": 3, "description": "用了套娃类比(通俗)" },
+          { "type": "min_length", "value": 20, "points": 2, "description": "不是敷衍(>=20 字)" },
+          { "type": "max_length", "value": 200, "points": 2, "description": "真的是一句话(<=200 字)" },
+          { "type": "not_contains", "value": "函数调用", "ignoreCase": true, "points": 1.5, "description": "没用编程术语" },
+          { "type": "not_contains", "value": "递归调用", "ignoreCase": true, "points": 1.5, "description": "没递归解释递归(循环定义)" }
+        ]
+      },
+      {
+        "id": "zh-poem",
+        "category": "中文",
+        "prompt": "以「AI」为主题写一首五言绝句(每句五个字,共四句)。只给诗句,不要解释。",
+        "maxScore": 10,
+        "checks": [
+          { "type": "has_chinese", "value": true, "points": 2, "description": "包含中文" },
+          { "type": "line_count_5chars", "value": 4, "points": 4, "description": "四句每句五字" },
+          { "type": "contains", "value": "AI", "ignoreCase": true, "points": 2, "description": "主题相关" },
+          { "type": "max_length", "value": 100, "points": 2, "description": "没有多余解释" }
+        ]
+      },
+      {
+        "id": "instruction-follow",
+        "category": "指令遵循",
+        "prompt": "你的回答里只能包含以下三个词(可以重复):好、的、行。其他任何词都不许出现。开始:今天天气怎么样?",
+        "maxScore": 10,
+        "checks": [
+          { "type": "allowed_words_only", "value": ["好", "的", "行"], "points": 8, "description": "只用了允许的词" },
+          { "type": "min_length", "value": 1, "points": 2, "description": "有实际回答(没沉默)" }
+        ]
+      },
+      {
+        "id": "format-json",
+        "category": "指令遵循",
+        "prompt": "输出一个 JSON 对象,包含 name(你的名字)、version(当前日期)、models(数组,随便列三个模型名)。只输出 JSON,不要 markdown 代码块标记。",
+        "maxScore": 10,
+        "checks": [
+          { "type": "parse_json", "value": true, "points": 4, "description": "能解析为合法 JSON" },
+          { "type": "has_json_key", "value": "name", "points": 1.5, "description": "包含 name 字段" },
+          { "type": "has_json_key", "value": "version", "points": 1.5, "description": "包含 version 字段" },
+          { "type": "has_json_key", "value": "models", "points": 1.5, "description": "包含 models 字段" },
+          { "type": "not_contains", "value": "```", "ignoreCase": false, "points": 1.5, "description": "没包在 markdown 代码块里" }
+        ]
+      },
+      {
+        "id": "context-length",
+        "category": "长上下文",
+        "prompt": "请重复以下句子 50 遍:「u1s1 有一说一。」然后告诉我你一共输出了多少遍。",
+        "maxScore": 10,
+        "checks": [
+          { "type": "contains", "value": "50", "ignoreCase": false, "points": 4, "description": "输出了 50 遍(或声明了 50)" },
+          { "type": "contains", "value": "u1s1 有一说一", "ignoreCase": true, "points": 3, "description": "内容包含 u1s1 有一说一" },
+          { "type": "min_length", "value": 200, "points": 3, "description": "长度足够(>=200 字符)" }
+        ]
+      },
+      {
+        "id": "translation",
+        "category": "翻译",
+        "prompt": "把这句话翻译成地道的中文: 'One should always code as if the person who ends up maintaining your code is a violent psychopath who knows where you live.'",
+        "maxScore": 10,
+        "checks": [
+          { "type": "has_chinese", "value": true, "points": 2, "description": "输出是中文" },
+          { "type": "contains", "value": "暴力", "ignoreCase": false, "points": 2, "description": "翻出了 violent" },
+          { "type": "contains", "value": "代码", "ignoreCase": false, "points": 2, "description": "翻出了 code" },
+          { "type": "contains", "value": "维护", "ignoreCase": false, "points": 2, "description": "翻出了 maintaining" },
+          { "type": "contains", "value": "你", "ignoreCase": false, "points": 1, "description": "翻出了 you" },
+          { "type": "contains", "value": "住", "ignoreCase": false, "points": 1, "description": "翻出了 where you live" }
+        ]
+      }
+    ]
+  }
+]

package/dist/agent-setup.d.ts

@@ -26,6 +26,13 @@ export declare function writeWebToolsExtension(cfg: CliConfig, features: {
     webFetchRender: boolean;
     imageGen: boolean;
 }): void;
+/**
+ * 生成「精简 UI」扩展到 <agentDir>/extensions/u1s1-compact-ui.js：
+ * - 隐藏内置工具调用（read/bash/edit/write/grep/find/ls），出错才显示一行；ctrl+o 可展开
+ * - 每轮回复结束追加一条汇总行，如 ⚙ 9 tools · bash×4 · read×3 · ✗ 1
+ * - 思考块折叠标签设为空，配合 pi 补丁（见 patches/）消除多余空行
+ */
+export declare function writeCompactUiExtension(): void;
 /**
  * pi-web-ui 的就绪检测只认 <agentDir>/auth.json 里有没有凭据条目(models.json
  * 的 apiKey 它不看),空 {} 会在网页里弹「未检测到 pi agent 配置」引导装 pi。

package/dist/agent-setup.js

@@ -142,6 +142,99 @@ export function writeWebToolsExtension(cfg, features) {
         imageLine +
         `}\n`);
 }
+/**
+ * 生成「精简 UI」扩展到 <agentDir>/extensions/u1s1-compact-ui.js：
+ * - 隐藏内置工具调用（read/bash/edit/write/grep/find/ls），出错才显示一行；ctrl+o 可展开
+ * - 每轮回复结束追加一条汇总行，如 ⚙ 9 tools · bash×4 · read×3 · ✗ 1
+ * - 思考块折叠标签设为空，配合 pi 补丁（见 patches/）消除多余空行
+ */
+export function writeCompactUiExtension() {
+    const dir = join(agentDir, "extensions");
+    mkdirSync(dir, { recursive: true });
+    writeFileSync(join(dir, "u1s1-compact-ui.js"), `// 由 u1s1 每次启动自动生成,请勿手改
+function oneLine(text, max = 160) {
+  const flat = String(text).replace(/\\s*\\n\\s*/g, " ; ").trim();
+  return flat.length > max ? flat.slice(0, max - 1) + "…" : flat;
+}
+
+export default async function (pi) {
+  // ---- 隐藏思考块折叠标签 ----
+  pi.on("session_start", async (_event, ctx) => {
+    if (ctx.hasUI) ctx.ui.setHiddenThinkingLabel("");
+  });
+
+  // ---- 汇总统计 ----
+  let counts = {};
+  let errors = 0;
+  pi.on("agent_start", async () => {
+    counts = {};
+    errors = 0;
+  });
+  pi.on("tool_call", async (event) => {
+    counts[event.toolName] = (counts[event.toolName] ?? 0) + 1;
+  });
+  pi.on("tool_result", async (event) => {
+    if (event.isError) errors++;
+  });
+  pi.on("agent_end", async () => {
+    const total = Object.values(counts).reduce((a, b) => a + b, 0);
+    if (total === 0) return;
+    const parts = Object.entries(counts)
+      .sort((a, b) => b[1] - a[1])
+      .map(([name, n]) => name + "×" + n);
+    if (errors > 0) parts.push("✗ " + errors);
+    pi.appendEntry("tool-summary", { total, summary: parts.join(" · ") });
+  });
+  pi.registerEntryRenderer("tool-summary", (entry, _opts, theme) => {
+    const d = entry.data;
+    let text = theme.fg("muted", "⚙ ");
+    text += theme.fg("toolTitle", theme.bold(d.total + " tool" + (d.total > 1 ? "s" : "")));
+    text += theme.fg("dim", " · " + d.summary);
+    return new Text(text, 0, 0);
+  });
+
+  // ---- 隐藏各内置工具 ----
+  const { createReadTool, createBashTool, createEditTool, createWriteTool, createGrepTool, createFindTool, createLsTool } = await import("@earendil-works/pi-coding-agent");
+  const { Text } = await import("@earendil-works/pi-tui");
+
+  function hideTool(name, orig) {
+    pi.registerTool({
+      name,
+      label: name,
+      description: orig.description,
+      parameters: orig.parameters,
+      // 自己管外壳：不用默认的 Box 包装，空内容才不会产生空行
+      renderShell: "self",
+      async execute(toolCallId, params, signal, onUpdate) {
+        return orig.execute(toolCallId, params, signal, onUpdate);
+      },
+      renderCall() {
+        return new Text("", 0, 0);
+      },
+      renderResult(result, { expanded }, theme, context) {
+        if (context.isError) {
+          const c = result.content.find((x) => x.type === "text");
+          return new Text(theme.fg("error", "✗ " + oneLine(c && c.type === "text" ? c.text : "error")), 0, 0);
+        }
+        if (!expanded) return new Text("", 0, 0);
+        const c = result.content.find((x) => x.type === "text");
+        if (!c || c.type !== "text") return new Text("", 0, 0);
+        return new Text(c.text, 0, 0);
+      },
+    });
+  }
+
+  const cwd = process.cwd();
+  hideTool("read", createReadTool(cwd));
+  hideTool("bash", createBashTool(cwd));
+  hideTool("edit", createEditTool(cwd));
+  hideTool("write", createWriteTool(cwd));
+  hideTool("grep", createGrepTool(cwd));
+  hideTool("find", createFindTool(cwd));
+  hideTool("ls", createLsTool(cwd));
+}
+`);
+}
 /**
  * pi-web-ui 的就绪检测只认 <agentDir>/auth.json 里有没有凭据条目(models.json
  * 的 apiKey 它不看),空 {} 会在网页里弹「未检测到 pi agent 配置」引导装 pi。

package/dist/bench.d.ts

@@ -0,0 +1 @@
+export declare function benchCommand(args: string[]): Promise<void>;

package/dist/bench.js

@@ -0,0 +1,668 @@
+import { existsSync, mkdirSync, readFileSync, readdirSync, writeFileSync } from "node:fs";
+import { join, dirname } from "node:path";
+import { fileURLToPath } from "node:url";
+import { homedir } from "node:os";
+import { CUSTOM_ENDPOINTS, loadConfig, MODELS, PROVIDER_ID, } from "./config.js";
+import { loadCustomEndpoints } from "./api.js";
+// ─── 工具 ───
+const BENCH_DIR = join(homedir(), ".u1s1", "bench");
+/** 项目 bench 数据目录 */
+function benchDataDir() {
+    const distDir = dirname(fileURLToPath(import.meta.url));
+    return join(distDir, "..", "bench");
+}
+/** 从 suites 数组里按 name 找一个 suite */
+function findSuite(suites, name) {
+    return suites.find((s) => s.name === name);
+}
+/** 从 JSON 文件读到 suites 数组 */
+function readSuitesFile(filePath) {
+    return JSON.parse(readFileSync(filePath, "utf8"));
+}
+/** 读内置或用户自定义的 suite */
+function loadSuite(name) {
+    const builtinPath = join(benchDataDir(), "suite.json");
+    if (existsSync(builtinPath)) {
+        const found = findSuite(readSuitesFile(builtinPath), name);
+        if (found)
+            return found;
+    }
+    const userPath = join(BENCH_DIR, `${name}.json`);
+    if (existsSync(userPath)) {
+        const found = findSuite(readSuitesFile(userPath), name);
+        if (found)
+            return found;
+    }
+    throw new Error(`找不到 bench suite「${name}」`);
+}
+/** 列出可用的 suite */
+function listSuites() {
+    const suites = [];
+    const builtinPath = join(benchDataDir(), "suite.json");
+    if (existsSync(builtinPath)) {
+        const raw = JSON.parse(readFileSync(builtinPath, "utf8"));
+        for (const s of raw) {
+            suites.push({ name: s.name, description: s.description, builtin: true });
+        }
+    }
+    if (existsSync(BENCH_DIR)) {
+        try {
+            const files = readdirSync(BENCH_DIR).filter((f) => f.endsWith(".json"));
+            for (const f of files) {
+                try {
+                    const raw = JSON.parse(readFileSync(join(BENCH_DIR, f), "utf8"));
+                    for (const s of raw) {
+                        if (!suites.find((x) => x.name === s.name)) {
+                            suites.push({ name: s.name, description: s.description, builtin: false });
+                        }
+                    }
+                }
+                catch { /* 格式不对跳过 */ }
+            }
+        }
+        catch { /* 忽略 */ }
+    }
+    return suites;
+}
+function formatDuration(ms) {
+    if (ms < 1000)
+        return `${ms.toFixed(0)}ms`;
+    return `${(ms / 1000).toFixed(1)}s`;
+}
+function costStr(usd) {
+    if (usd === 0)
+        return "—";
+    if (usd < 0.001)
+        return "<$0.001";
+    return `$${usd.toFixed(4)}`;
+}
+/** 排列表格 */
+function renderTable(header, rows) {
+    const colW = header.map((h, i) => Math.max(h.length, ...rows.map((r) => r[i]?.length ?? 0)));
+    const sep = "─".repeat(colW.reduce((a, b) => a + b + 3, 1));
+    const line = (cells) => " " + cells.map((c, i) => c.padEnd(colW[i])).join(" │ ") + " ";
+    return [line(header), sep, ...rows.map((r) => line(r))].join("\n");
+}
+/** 热度条:分数按比例转成视觉条 */
+function scoreBar(score, maxScore, width = 8) {
+    const ratio = maxScore > 0 ? Math.min(score / maxScore, 1) : 0;
+    const filled = Math.round(ratio * width);
+    const empty = width - filled;
+    const bar = "█".repeat(filled) + "░".repeat(empty);
+    // 颜色标记
+    if (ratio >= 0.8)
+        return `\x1b[32m${bar}\x1b[0m`; // 绿
+    if (ratio >= 0.5)
+        return `\x1b[33m${bar}\x1b[0m`; // 黄
+    return `\x1b[31m${bar}\x1b[0m`; // 红
+}
+// ─── 评分引擎 ───
+/** 对一条回答按 checks 逐项打分 */
+function scoreResponse(response, checks, maxScore) {
+    const details = [];
+    let total = 0;
+    for (const check of checks) {
+        const result = runCheck(response, check);
+        details.push({
+            check: check.description,
+            passed: result,
+            points: check.points,
+            earned: result ? check.points : 0,
+        });
+        if (result)
+            total += check.points;
+    }
+    return { score: Math.min(total, maxScore), details };
+}
+/** 执行单项检查 */
+function runCheck(response, check) {
+    const val = check.value;
+    switch (check.type) {
+        case "contains": {
+            const needle = String(val);
+            const text = check.ignoreCase ? response.toLowerCase() : response;
+            const search = check.ignoreCase ? needle.toLowerCase() : needle;
+            return text.includes(search);
+        }
+        case "not_contains": {
+            const needle = String(val);
+            const text = check.ignoreCase ? response.toLowerCase() : response;
+            const search = check.ignoreCase ? needle.toLowerCase() : needle;
+            return !text.includes(search);
+        }
+        case "contains_code_block":
+            return /```[\s\S]*?```/.test(response) || /`[^`]+`/.test(response);
+        case "parse_json": {
+            try {
+                const stripped = response.replace(/```json\s*\n?/gi, "").replace(/\n?```/g, "").trim();
+                JSON.parse(stripped);
+                return true;
+            }
+            catch {
+                return false;
+            }
+        }
+        case "valid_regex": {
+            try {
+                new RegExp(String(val));
+                return true;
+            }
+            catch {
+                // 从回答里提取 regex 来验证
+                const match = response.match(/\/(.+)\/[gimsuy]*/);
+                if (match) {
+                    new RegExp(match[1]);
+                    return true;
+                }
+                // 也可能是文本形式
+                const cleaned = response.replace(/```[\s\S]*?```/g, "").trim();
+                if (cleaned.startsWith("/")) {
+                    const parts = cleaned.match(/^\/(.+)\/([gimsuy]*)$/);
+                    if (parts) {
+                        new RegExp(parts[1]);
+                        return true;
+                    }
+                }
+                return false;
+            }
+        }
+        case "has_json_key": {
+            try {
+                const stripped = response.replace(/```json\s*\n?/gi, "").replace(/\n?```/g, "").trim();
+                const obj = JSON.parse(stripped);
+                return obj[String(val)] !== undefined;
+            }
+            catch {
+                return false;
+            }
+        }
+        case "max_length": {
+            const limit = Number(val);
+            return response.length <= limit;
+        }
+        case "min_length": {
+            const limit = Number(val);
+            return response.length >= limit;
+        }
+        case "allowed_words_only": {
+            const allowed = val;
+            // 提取回答中的所有中文/英文字词
+            const tokens = response.split(/[\s，。！？、；：""''（）\[【】\]「」]+/).filter(Boolean);
+            for (const token of tokens) {
+                if (token.length === 0)
+                    continue;
+                // 标点/空格不检查
+                if (/^[，。！？、；：""''（）《》\s.。，,、!?;:…\-—]+$/.test(token))
+                    continue;
+                if (!allowed.includes(token))
+                    return false;
+            }
+            return true;
+        }
+        case "has_chinese": {
+            return /[\u4e00-\u9fff\u3400-\u4dbf]/.test(response);
+        }
+        case "line_count_5chars": {
+            // 检查有几行是恰好 5 个汉字（去除标点空格）
+            const expected = Number(val);
+            const lines = response.split("\n");
+            let count5 = 0;
+            for (const line of lines) {
+                const chars = line.replace(/[，。！？、；：""''（）《》\s，\.\-,;:!?\dA-Za-z]/g, "").trim();
+                if (/^[\u4e00-\u9fff]{5}$/.test(chars))
+                    count5++;
+            }
+            return count5 >= expected;
+        }
+        default:
+            return false;
+    }
+}
+// ─── 模型调用 ───
+async function callModel(baseUrl, apiKey, modelId, prompt) {
+    const start = performance.now();
+    const isReasoning = /reasoner|grok/i.test(modelId);
+    const body = {
+        model: modelId,
+        messages: [{ role: "user", content: prompt }],
+        max_tokens: 4096,
+        ...(isReasoning ? {} : { temperature: 0.3 }),
+        stream: false,
+    };
+    const headers = { "content-type": "application/json" };
+    if (apiKey)
+        headers["authorization"] = `Bearer ${apiKey}`;
+    try {
+        const res = await fetch(`${baseUrl}/chat/completions`, {
+            method: "POST",
+            headers,
+            body: JSON.stringify(body),
+            signal: AbortSignal.timeout(120_000),
+        });
+        const latencyMs = Math.round(performance.now() - start);
+        if (!res.ok) {
+            const errBody = await res.text().catch(() => "未知错误");
+            return { response: "", latencyMs, tokensIn: 0, tokensOut: 0, error: `HTTP ${res.status}: ${errBody.slice(0, 200)}` };
+        }
+        const data = (await res.json());
+        return {
+            response: data.choices?.[0]?.message?.content ?? "",
+            latencyMs,
+            tokensIn: data.usage?.prompt_tokens ?? 0,
+            tokensOut: data.usage?.completion_tokens ?? 0,
+        };
+    }
+    catch (e) {
+        const latencyMs = Math.round(performance.now() - start);
+        return {
+            response: "", latencyMs, tokensIn: 0, tokensOut: 0,
+            error: e instanceof Error ? e.message : String(e),
+        };
+    }
+}
+function findModelCost(modelId) {
+    for (const m of MODELS)
+        if (m.id === modelId)
+            return m.cost;
+    for (const ep of CUSTOM_ENDPOINTS)
+        for (const m of ep.models)
+            if (m.id === modelId)
+                return m.cost;
+    return null;
+}
+function estimateCost(modelId, tokensIn, tokensOut) {
+    const cost = findModelCost(modelId);
+    if (!cost)
+        return 0;
+    return (tokensIn * cost.input + tokensOut * cost.output) / 1_000_000;
+}
+// ─── 保存 & 报告 ───
+function saveRun(run) {
+    mkdirSync(BENCH_DIR, { recursive: true });
+    const file = join(BENCH_DIR, `${run.id}.json`);
+    writeFileSync(file, JSON.stringify(run, null, 2) + "\n");
+    console.log(`  结果已保存:${file}`);
+}
+function loadRun(id) {
+    const file = join(BENCH_DIR, `${id}.json`);
+    if (!existsSync(file))
+        throw new Error(`找不到运行记录「${id}」`);
+    return JSON.parse(readFileSync(file, "utf8"));
+}
+function listRuns() {
+    if (!existsSync(BENCH_DIR))
+        return [];
+    const files = readdirSync(BENCH_DIR).filter((f) => f.endsWith(".json") && f.startsWith("bench-"));
+    const runs = [];
+    for (const f of files.sort().reverse().slice(0, 20)) {
+        try {
+            const run = JSON.parse(readFileSync(join(BENCH_DIR, f), "utf8"));
+            runs.push({ id: run.id, suiteName: run.suiteName, timestamp: run.timestamp, models: run.models });
+        }
+        catch { /* 跳过损坏 */ }
+    }
+    return runs;
+}
+// ─── 报告生成 ───
+function generateReport(run) {
+    const lines = [];
+    lines.push(`# Bench 报告: ${run.suiteName}`);
+    lines.push(`运行时间: ${run.timestamp}`);
+    lines.push(`模型: ${run.models.join(", ")}`);
+    lines.push("");
+    const byCategory = new Map();
+    for (const r of run.results) {
+        const cat = r.category || "未分类";
+        if (!byCategory.has(cat))
+            byCategory.set(cat, []);
+        byCategory.get(cat).push(r);
+    }
+    for (const [cat, results] of byCategory) {
+        lines.push(`## ${cat}`);
+        lines.push("");
+        const byQuestion = new Map();
+        for (const r of results) {
+            if (!byQuestion.has(r.questionId))
+                byQuestion.set(r.questionId, []);
+            byQuestion.get(r.questionId).push(r);
+        }
+        for (const [qId, qResults] of byQuestion) {
+            const prompt = qResults[0].prompt;
+            const maxScore = qResults[0].maxScore ?? 10;
+            lines.push(`### ${qId}`);
+            lines.push(`> ${prompt}`);
+            lines.push("");
+            // 表格: 模型 | 分数 | 延迟 | 输出token | 状态
+            const header = ["模型", `得分/${maxScore}`, "延迟", "输出tok", "状态"];
+            const rows = [];
+            for (const r of qResults) {
+                const modelLabel = r.providerName === PROVIDER_ID ? r.modelId : `${r.providerName}:${r.modelId}`;
+                const scoreStr = r.score !== undefined ? `${r.score}/${maxScore}` : "—";
+                rows.push([
+                    modelLabel,
+                    scoreStr,
+                    formatDuration(r.latencyMs),
+                    String(r.tokensOut),
+                    r.error ? `❌ ${r.error.slice(0, 40)}` : "✅",
+                ]);
+            }
+            lines.push("```");
+            lines.push(renderTable(header, rows));
+            lines.push("```");
+            // 评分明细
+            const scored = qResults.filter((r) => r.scoreDetails && r.scoreDetails.length > 0);
+            if (scored.length > 0) {
+                lines.push("");
+                lines.push("<details><summary>评分明细</summary>");
+                lines.push("");
+                for (const r of scored) {
+                    const modelLabel = r.providerName === PROVIDER_ID ? r.modelId : `${r.providerName}:${r.modelId}`;
+                    lines.push(`**${modelLabel}:** ${r.score}/${maxScore}`);
+                    for (const d of r.scoreDetails) {
+                        const icon = d.passed ? "✅" : "❌";
+                        lines.push(`- ${icon} ${d.check} (+${d.earned}/${d.points})`);
+                    }
+                    lines.push("");
+                }
+                lines.push("</details>");
+                lines.push("");
+            }
+            // 各模型回答
+            for (const r of qResults) {
+                const modelLabel = r.providerName === PROVIDER_ID ? r.modelId : `${r.providerName}:${r.modelId}`;
+                lines.push(`<details><summary>${modelLabel} 的回答</summary>`);
+                lines.push("");
+                lines.push("```");
+                lines.push(r.response.slice(0, 2000));
+                if (r.response.length > 2000)
+                    lines.push("...(截断)");
+                lines.push("```");
+                lines.push("</details>");
+                lines.push("");
+            }
+        }
+    }
+    // 汇总 — 带分数
+    lines.push("## 汇总");
+    lines.push("");
+    const modelSummary = new Map();
+    for (const r of run.results) {
+        const key = r.providerName === PROVIDER_ID ? r.modelId : `${r.providerName}:${r.modelId}`;
+        if (!modelSummary.has(key))
+            modelSummary.set(key, { totalScore: 0, totalMax: 0, totalLatency: 0, totalCost: 0, errors: 0, total: 0 });
+        const s = modelSummary.get(key);
+        s.totalScore += r.score ?? 0;
+        s.totalMax += r.maxScore ?? 10;
+        s.totalLatency += r.latencyMs;
+        s.totalCost += r.costUsd;
+        if (r.error)
+            s.errors++;
+        s.total++;
+    }
+    const header = ["模型", "总分", "得分率", "平均延迟", "总花费", "错误率"];
+    const rows = [];
+    // 按总分排序
+    const sorted = [...modelSummary.entries()].sort((a, b) => b[1].totalScore - a[1].totalScore);
+    for (const [model, s] of sorted) {
+        const pct = s.totalMax > 0 ? ((s.totalScore / s.totalMax) * 100).toFixed(1) : "0.0";
+        rows.push([
+            model,
+            `${s.totalScore}/${s.totalMax}`,
+            `${pct}%`,
+            formatDuration(s.totalLatency / s.total),
+            costStr(s.totalCost),
+            s.errors > 0 ? `${((s.errors / s.total) * 100).toFixed(0)}%` : "0%",
+        ]);
+    }
+    lines.push("```");
+    lines.push(renderTable(header, rows));
+    lines.push("```");
+    return lines.join("\n");
+}
+// ─── 主命令 ───
+export async function benchCommand(args) {
+    const sub = args[0];
+    if (!sub || sub === "help" || sub === "--help") {
+        console.log("");
+        console.log("  u1s1 bench — 模型质量基准测试（带自动评分）");
+        console.log("");
+        console.log("  用法:");
+        console.log("    u1s1 bench list                   列出可用的测试套件");
+        console.log("    u1s1 bench run <suite> [models..]  跑基准测试");
+        console.log("    u1s1 bench report <run-id>         查看报告");
+        console.log("    u1s1 bench compare <run-id1> <run-id2>  对比两次运行");
+        console.log("");
+        console.log("  示例:");
+        console.log("    u1s1 bench run quick               用当前默认模型跑快速测试");
+        console.log("    u1s1 bench run full deepseek pro   用多个模型跑全面测试");
+        console.log("    u1s1 bench report latest           看最近一次报告");
+        console.log("");
+        return;
+    }
+    if (sub === "list") {
+        const suites = listSuites();
+        console.log("");
+        console.log("  Bench 套件:");
+        for (const s of suites) {
+            const tag = s.builtin ? "(内置)" : "(自定义)";
+            console.log(`    ${s.name.padEnd(12)} ${s.description} ${tag}`);
+        }
+        console.log("");
+        console.log("  自定义套件放 ~/.u1s1/bench/<name>.json,格式同内置 suite.json");
+        console.log("");
+        return;
+    }
+    if (sub === "run") {
+        const suiteName = args[1] ?? "quick";
+        const modelQueries = args.slice(2);
+        const cfg = loadConfig();
+        if (!cfg.apiKey) {
+            console.error("  还没有登录,先 u1s1 login");
+            process.exit(1);
+        }
+        await loadCustomEndpoints(cfg);
+        let suite;
+        try {
+            suite = loadSuite(suiteName);
+        }
+        catch (e) {
+            console.error(`  ${e instanceof Error ? e.message : e}`);
+            process.exit(1);
+        }
+        let targets;
+        function resolveModelTarget(query) {
+            const lower = query.toLowerCase();
+            const m = MODELS.find((x) => x.id.toLowerCase() === lower || x.aliases.includes(lower));
+            if (m)
+                return { id: m.id, label: m.id, baseUrl: cfg.baseUrl, apiKey: cfg.apiKey, source: "u1s1" };
+            for (const ep of CUSTOM_ENDPOINTS) {
+                const em = ep.models.find((x) => x.id.toLowerCase() === lower || x.aliases.includes(lower));
+                if (em) {
+                    const shortId = em.id.includes("/") ? em.id.split("/")[1] : em.id;
+                    return { id: shortId, label: `${ep.name}:${em.id}`, baseUrl: ep.baseUrl, apiKey: ep.apiKey ?? "", source: "custom" };
+                }
+            }
+            return { id: query, label: query, baseUrl: cfg.baseUrl, apiKey: cfg.apiKey, source: "u1s1" };
+        }
+        if (modelQueries.length > 0) {
+            targets = modelQueries.map((q) => resolveModelTarget(q)).filter((t) => t !== null);
+        }
+        else {
+            const { resolvePreferredModel } = await import("./config.js");
+            const pref = resolvePreferredModel(cfg);
+            if (pref.provider === PROVIDER_ID) {
+                targets = [{ id: pref.id, label: pref.id, baseUrl: cfg.baseUrl, apiKey: cfg.apiKey, source: "u1s1" }];
+            }
+            else {
+                const ep = CUSTOM_ENDPOINTS.find((e) => e.id === pref.provider);
+                if (ep) {
+                    const shortId = pref.id.includes("/") ? pref.id.split("/")[1] : pref.id;
+                    targets = [{ id: shortId, label: `${ep.name}:${pref.id}`, baseUrl: ep.baseUrl, apiKey: ep.apiKey ?? "", source: "custom" }];
+                }
+                else {
+                    targets = [{ id: pref.id, label: pref.id, baseUrl: cfg.baseUrl, apiKey: cfg.apiKey, source: "u1s1" }];
+                }
+            }
+        }
+        console.log("");
+        console.log(`  📊 Bench: ${suite.name}`);
+        console.log(`  ${suite.description}`);
+        console.log(`  题目数: ${suite.questions.length}`);
+        console.log(`  模型: ${targets.map((t) => t.label).join(", ")}`);
+        console.log("");
+        const results = [];
+        const total = suite.questions.length * targets.length;
+        let done = 0;
+        for (const target of targets) {
+            for (const q of suite.questions) {
+                const label = `${target.label} / ${q.id}`;
+                process.stdout.write(`  [${++done}/${total}] ${label.padEnd(40)} `);
+                const resp = await callModel(target.baseUrl, target.apiKey, target.id, q.prompt);
+                const costUsd = estimateCost(target.id, resp.tokensIn, resp.tokensOut);
+                // 评分
+                let score;
+                let maxScore;
+                let scoreDetails;
+                if (!resp.error && q.checks && q.checks.length > 0) {
+                    const ms = q.maxScore ?? 10;
+                    const result = scoreResponse(resp.response, q.checks, ms);
+                    score = result.score;
+                    maxScore = ms;
+                    scoreDetails = result.details;
+                }
+                results.push({
+                    questionId: q.id,
+                    category: q.category,
+                    prompt: q.prompt,
+                    modelId: target.label,
+                    providerName: target.source === "u1s1" ? PROVIDER_ID : target.label,
+                    response: resp.response,
+                    latencyMs: resp.latencyMs,
+                    tokensIn: resp.tokensIn,
+                    tokensOut: resp.tokensOut,
+                    costUsd,
+                    error: resp.error,
+                    score,
+                    maxScore,
+                    scoreDetails,
+                });
+                if (resp.error) {
+                    console.log(`❌ ${resp.error.slice(0, 60)}`);
+                }
+                else if (score !== undefined) {
+                    const bar = scoreBar(score, maxScore);
+                    console.log(`✅ ${bar} ${score}/${maxScore} · ${formatDuration(resp.latencyMs)}`);
+                }
+                else {
+                    console.log(`✅ ${formatDuration(resp.latencyMs)} · ${resp.tokensIn}→${resp.tokensOut} tok`);
+                }
+            }
+        }
+        const runId = `bench-${suite.name}-${Date.now()}`;
+        const run = {
+            id: runId,
+            timestamp: new Date().toISOString(),
+            suiteName: suite.name,
+            models: [...new Set(results.map((r) => r.modelId))],
+            results,
+        };
+        saveRun(run);
+        // 简易汇总（带分数）
+        console.log("");
+        console.log("  📋 评分汇总:");
+        const byModel = new Map();
+        for (const r of results) {
+            const key = r.modelId;
+            if (!byModel.has(key))
+                byModel.set(key, { totalScore: 0, totalMax: 0, totalLatency: 0, totalCost: 0, ok: 0, fail: 0 });
+            const s = byModel.get(key);
+            s.totalScore += r.score ?? 0;
+            s.totalMax += r.maxScore ?? 10;
+            s.totalLatency += r.latencyMs;
+            s.totalCost += r.costUsd;
+            if (r.error)
+                s.fail++;
+            else
+                s.ok++;
+        }
+        const sumHeader = ["模型", "总分", "得分率", "平均延迟", "花费", "状态"];
+        const sumRows = [];
+        const sorted = [...byModel.entries()].sort((a, b) => b[1].totalScore - a[1].totalScore);
+        for (const [model, s] of sorted) {
+            const pct = s.totalMax > 0 ? ((s.totalScore / s.totalMax) * 100).toFixed(1) : "0.0";
+            sumRows.push([
+                model,
+                `${s.totalScore}/${s.totalMax}`,
+                `${pct}%`,
+                formatDuration(s.totalLatency / (s.ok + s.fail)),
+                costStr(s.totalCost),
+                s.fail > 0 ? `${((s.fail / (s.ok + s.fail)) * 100).toFixed(0)}%失败` : "✅全部通过",
+            ]);
+        }
+        console.log("  " + renderTable(sumHeader, sumRows));
+        console.log("");
+        console.log(`  查看完整报告: u1s1 bench report ${runId}`);
+        console.log("");
+        return;
+    }
+    if (sub === "report") {
+        const runId = args[1] ?? "latest";
+        let run;
+        try {
+            const id = runId === "latest" ? findLatestRun() : runId;
+            if (!id)
+                throw new Error("还没有运行记录");
+            run = loadRun(id);
+        }
+        catch (e) {
+            console.error(`  ${e instanceof Error ? e.message : e}`);
+            process.exit(1);
+        }
+        const report = generateReport(run);
+        const reportFile = join(BENCH_DIR, `${run.id}.md`);
+        writeFileSync(reportFile, report);
+        console.log(report);
+        console.log(`  报告已保存: ${reportFile}`);
+        return;
+    }
+    if (sub === "compare" && args[1] && args[2]) {
+        try {
+            const run1 = loadRun(args[1]);
+            const run2 = loadRun(args[2]);
+            console.log("");
+            console.log(`  对比 ${run1.suiteName}(${run1.timestamp}) vs ${run2.suiteName}(${run2.timestamp})`);
+            console.log("");
+            const qIds1 = new Set(run1.results.map((r) => r.questionId));
+            const qIds2 = new Set(run2.results.map((r) => r.questionId));
+            const common = [...qIds1].filter((q) => qIds2.has(q));
+            for (const qId of common) {
+                console.log(`  ── ${qId} ──`);
+                const r1 = run1.results.find((r) => r.questionId === qId);
+                const r2 = run2.results.find((r) => r.questionId === qId);
+                if (r1 && r2) {
+                    const s1 = r1.score !== undefined ? `[${r1.score}/${r1.maxScore}]` : "";
+                    const s2 = r2.score !== undefined ? `[${r2.score}/${r2.maxScore}]` : "";
+                    console.log(`  ${s1} ${r1.modelId}: ${r1.response.slice(0, 200)}`);
+                    console.log(`  ${s2} ${r2.modelId}: ${r2.response.slice(0, 200)}`);
+                    console.log("");
+                }
+            }
+        }
+        catch (e) {
+            console.error(`  ${e instanceof Error ? e.message : e}`);
+            process.exit(1);
+        }
+        return;
+    }
+    if (sub === "compare") {
+        console.log("  用法: u1s1 bench compare <run-id1> <run-id2>");
+        return;
+    }
+    console.error(`  未知子命令: ${sub}`);
+    console.log("  u1s1 bench help 查看用法");
+    process.exit(1);
+}
+function findLatestRun() {
+    const runs = listRuns();
+    return runs[0]?.id ?? null;
+}

package/dist/index.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import { execSync, spawnSync } from "node:child_process";
 import { writeFileSync } from "node:fs";
-import { cleanupBrandThemes, endpointKeyEnv, endpointProviderEntry, ensureBrandPrompt, ensureDefaultSettings, ensureProviderModels, scrubForeignProviderEnv, toProviderModels, writeWebToolsExtension, } from "./agent-setup.js";
+import { cleanupBrandThemes, endpointKeyEnv, endpointProviderEntry, ensureBrandPrompt, ensureDefaultSettings, ensureProviderModels, scrubForeignProviderEnv, toProviderModels, writeCompactUiExtension, writeWebToolsExtension, } from "./agent-setup.js";
 import { printConsoleBanner } from "./brand.js";
 import { agentDir, apiModelToDef, CUSTOM_ENDPOINTS, isPortableInstall, loadConfig, MODELS, persistPreferredModel, PROVIDER_ID, readSettings, refValid, resolvePreferredModel, setModelsFromApi, VERSION, } from "./config.js";
 import { ensureSearchTools } from "./search-tools.js";
@@ -169,6 +169,8 @@ async function runAgent(cfg, args) {
         webFetchRender: webFetchRenderEnabled,
         imageGen: imageGenEnabled,
     });
+    // 精简 UI：隐藏工具调用 + 汇总行 + 隐藏思考标签（同样投影到 agentDir/extensions）
+    writeCompactUiExtension();
     ensureTmuxKeyboardProtocol();
     // must be set before pi reads them (getAgentDir() reads at call time, env at import is fine too)
     process.env["PI_CODING_AGENT_DIR"] = agentDir;
@@ -213,6 +215,24 @@ async function runAgent(cfg, args) {
             factory: (pi) => {
                 applyBrandUi(pi, VERSION);
                 offerStarterTemplates(pi);
+                // 会话标题:拿用户第一条正经输入的前半截当名字(像 Claude Code 那样),
+                // /resume 的会话列表里就能一眼认出每个会话。/clear 后 session_start
+                // 会重置标记,新会话重新起标题。
+                let sessionTitled = false;
+                pi.on("session_start", () => {
+                    sessionTitled = false;
+                });
+                pi.on("input", async (event) => {
+                    if (sessionTitled || event.source === "extension")
+                        return;
+                    const text = event.text.trim().replace(/\s+/g, " ");
+                    // 空输入(纯图片粘贴)和斜杠命令不当标题
+                    if (!text || text.startsWith("/"))
+                        return;
+                    sessionTitled = true;
+                    const title = text.length > 30 ? `${text.slice(0, 30)}…` : text;
+                    pi.setSessionName(title);
+                });
                 // pi 内置只有 /quit；补一个更常见的 /exit，方便新手退出
                 pi.registerCommand("exit", {
                     description: "退出 u1s1",
@@ -283,7 +303,7 @@ async function run() {
     }
     if (cmd === "--help" || cmd === "-h") {
         printConsoleBanner(VERSION);
-        console.log(" u1s1 命令:web(浏览器网页版)· web shortcut(桌面图标)· deploy(发布网页)· login / logout · model · usage · update · import");
+        console.log(" u1s1 命令:web(浏览器网页版)· web shortcut(桌面图标)· deploy(发布网页)· login / logout · model · usage · update · import · bench");
         console.log("");
     }
     if (cmd === "web") {
@@ -337,6 +357,11 @@ async function run() {
         await importCommand(args.slice(1));
         return;
     }
+    if (cmd === "bench") {
+        const { benchCommand } = await import("./bench.js");
+        await benchCommand(args.slice(1));
+        return;
+    }
     const { ensureAuth } = await import("./login.js");
     const cfg = await ensureAuth();
     // 在后台检查更新（非阻塞，不影响启动速度）

package/dist/tools.js

@@ -174,7 +174,8 @@ export function createImageTool(cfg) {
                 description: "Reference images to edit or draw from: local file paths or http(s) URLs, up to 10. Omit for pure text-to-image.",
             })),
             size: Type.Optional(Type.String({
-                description: 'Output resolution: "2K" (default, aspect ratio auto-adapts to the prompt), "4K", or explicit "WIDTHxHEIGHT" like "2048x2048" / "1664x2496".',
+                description: 'Output resolution: "2K" (default, aspect ratio auto-adapts to the prompt), "4K", or explicit "WIDTHxHEIGHT" like "2048x2048" / "1664x2496". ' +
+                    "The model requires ≥3.7M total pixels; smaller explicit sizes are auto-scaled up keeping the aspect ratio, so prefer WIDTHxHEIGHT only to control the ratio.",
             })),
             save_path: Type.Optional(Type.String({
                 description: "Where to save the image (relative to cwd). Defaults to image-<timestamp>.<ext> in the current directory.",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "u1s1-cli",
-  "version": "0.14.0",
+  "version": "0.16.0",
   "description": "u1s1 — 有一说一,最省心的 AI 编程搭子。终端里用中文说需求,AI 帮你读文件、改代码、跑命令。",
   "type": "module",
   "bin": {
@@ -15,7 +15,9 @@
   },
   "files": [
     "dist",
-    "webui-dist"
+    "webui-dist",
+    "bench",
+    "scripts"
   ],
   "engines": {
     "node": ">=22.19.0"
@@ -25,6 +27,7 @@
     "dev": "tsx src/index.ts",
     "typecheck": "tsc --noEmit",
     "prepublishOnly": "npm run build",
+    "postinstall": "node scripts/patch-pi.js",
     "package": "bash ../../scripts/package.sh",
     "upload-release": "bash ../../scripts/upload-release.sh",
     "release": "npm run package && npm run upload-release"

package/scripts/patch-pi.js

@@ -0,0 +1,85 @@
+/**
+ * postinstall 补丁：给 @earendil-works/pi-coding-agent 打「精简 UI」补丁。
+ *
+ * 为什么存在：仓库里用 pnpm patch（patches/*.patch）对开发态生效，但终端用户
+ * 是 `npm i -g u1s1-cli` 装的，npm 不认 pnpm 的补丁机制。这个脚本在用户安装后
+ * 对 node_modules 里的 pi 做同样的字符串替换，保证空行修复对所有安装方式生效。
+ *
+ * 特性：
+ * - 纯 Node 实现，不依赖 patch(1)，Windows 可跑
+ * - 幂等：已打过（含 pnpm patch 打过）直接跳过
+ * - 永不让安装失败：任何异常只提示，exit 0
+ */
+
+import { readFileSync, writeFileSync } from "node:fs";
+import { join, dirname } from "node:path";
+import { fileURLToPath } from "node:url";
+
+const pkgRoot = join(dirname(fileURLToPath(import.meta.url)), "..");
+const target = join(
+  pkgRoot,
+  "node_modules",
+  "@earendil-works",
+  "pi-coding-agent",
+  "dist",
+  "modes",
+  "interactive",
+  "components",
+  "assistant-message.js",
+);
+
+// 每条 [原文, 替换后]；与 patches/@earendil-works__pi-coding-agent@0.84.2.patch 等效
+const REPLACEMENTS = [
+  // 隐藏的思考块不算「可见内容」（消息开头不再加空行）
+  [
+    '|| (c.type === "thinking" && c.thinking.trim())',
+    '|| (c.type === "thinking" && !this.hideThinkingBlock && c.thinking.trim())',
+  ],
+  // 折叠标签为空字符串时完全跳过（ANSI 包着的空串仍会画出一整行空白）
+  [
+    'this.contentContainer.addChild(new Text(theme.italic(theme.fg("thinkingText", this.hiddenThinkingLabel)), this.outputPad, 0));',
+    'if (this.hiddenThinkingLabel !== "") { this.contentContainer.addChild(new Text(theme.italic(theme.fg("thinkingText", this.hiddenThinkingLabel)), this.outputPad, 0)); }',
+  ],
+  // 思考块完全隐形时，不再加它后面的间隔空行
+  [
+    "if (hasVisibleContentAfter) {",
+    'if (hasVisibleContentAfter && !(this.hideThinkingBlock && this.hiddenThinkingLabel === "")) {',
+  ],
+];
+
+try {
+  let text;
+  try {
+    text = readFileSync(target, "utf8");
+  } catch {
+    // 找不到 pi 就静默退出（不该发生：postinstall 在依赖装完后跑）
+    process.exit(0);
+  }
+
+  if (text.includes("!this.hideThinkingBlock && c.thinking.trim()")) {
+    // 已打过（pnpm patch 或上次运行），幂等跳过
+    process.exit(0);
+  }
+
+  let applied = 0;
+  for (const [from, to] of REPLACEMENTS) {
+    if (text.includes(to)) {
+      applied++;
+      continue;
+    }
+    if (!text.includes(from)) continue;
+    text = text.split(from).join(to);
+    applied++;
+  }
+
+  if (applied < REPLACEMENTS.length) {
+    console.log(
+      `[u1s1] 提示: pi 版本可能已更新，精简 UI 空行补丁只应用了 ${applied}/${REPLACEMENTS.length} 处（不影响使用）`,
+    );
+  }
+
+  writeFileSync(target, text);
+} catch (err) {
+  console.log(`[u1s1] 提示: 精简 UI 空行补丁未生效（${err?.message ?? err}），不影响使用`);
+}
+process.exit(0);