@bty/customer-service-cli 0.4.7 → 0.4.8

package/README.md

@@ -473,6 +473,60 @@ cs-cli change-consumer delivery complete <delivery_id> --status completed \
 | --- | --- |
 | `monitor workspaces [--has-agent] [--page N] [--page-size N]` | 运营工作空间列表（含合同、客户信息） |
 
+### 测试集 (`testset`)
+
+测试集驱动 Agent 回归验证：测试集（testset）下挂多条用例（case），执行后产生批次（batch）。CLI 让上层 Agent / 工程师本地 debug 无需开浏览器即可走完「触发 → 拉结果 → 改用例 → 重跑」闭环。所有测试集相关命令收敛在 `testset` 主语下（ADR-001b）。
+
+#### 测试集 CRUD
+
+| 命令 | 说明 |
+| --- | --- |
+| `testset list --customer-agent-config-id <id>` | 列出测试集（分页，Agent ID 必填） |
+| `testset show <id>` | 测试集详情 |
+| `testset update <id> [--name ...] [--description ...] [--data @file.json]` | 更新测试集元数据（至少一项） |
+| `testset delete <id>` | 软删除测试集 |
+| `testset copy <id> [--name <副本名>]` | 克隆测试集（含用例，**不带历史批次**） |
+| `testset get-eval-prompt <id> [--raw]` | 查看评估 Prompt；`--raw` 输出裸文本到 stdout |
+| `testset set-eval-prompt <id> --prompt <s\|@file>` | 设置评估 Prompt（`@file` 读纯文本，不 JSON.parse） |
+
+#### 用例（`testset case`，全部必填 `--testset`）
+
+| 命令 | 说明 |
+| --- | --- |
+| `testset case list --testset <id>` | 列出测试集下用例（分页） |
+| `testset case show <case_id> --testset <id>` | 查看单条用例（内部翻页 + 本地 filter） |
+| `testset case create --testset <id> --data @file.json` | 新增用例（`case_content` 必填，至少 1 条对话） |
+| `testset case update <case_id> --testset <id> --data @file.json` | 更新用例 |
+| `testset case delete <case_id> --testset <id>` | 软删除用例 |
+
+#### 批次只读（`testset batch`）
+
+| 命令 | 说明 |
+| --- | --- |
+| `testset batch list [--testset <id>] [--status <s>]` | 列出批次（缺 `--testset` 时列全 workspace） |
+| `testset batch show <batch_id>` | 批次详情（含执行记录） |
+| `testset batch status <batch_id>` | 轻量批次状态（含 `is_terminal / total_count / pass_count / fail_count`） |
+
+#### 触发命令（`testset run / run-case / export`）
+
+| 命令 | 说明 |
+| --- | --- |
+| `testset run --testset <id> --agent <id> [--wait] [--timeout <sec>]` | 触发整批回归。默认异步立刻返回 `batch_id`；`--wait` 切同步阻塞 poll 到终态 |
+| `testset run-case --batch <id> --case <id>` | 单跑用例（覆盖写回该批次） |
+| `testset export --testset <id> --batch <id> --output <path>` | 导出批次为 xlsx 落盘 |
+
+**⚠️ `run --wait` 协议（详见 ADR-002）**：
+
+- Poll 间隔 3000ms，默认超时 600 秒；`--timeout <sec>` 覆盖（**单位是秒**，与全局 `--request-timeout` 的毫秒**严格区分**）
+- 终态判定锚定后端 `is_terminal` 字段（不枚举 `status` 字面量，避免后端新增字面量时 CLI 卡死）
+- **退出码中立**：批次 pass_count=0、status=failed、超时未完成全部 `exit 0`；只有 CLI 自身失败（鉴权 / 网络 / 参数错误 / poll 连续 3 次失败）才非 0
+- Poll 单次失败容错：连续 < 3 次抛错继续，3 次中断；中断时 `lastError` 是 `APIError(2/401)` → `exit 2`，否则 `exit 3`
+- 超时仍 `exit 0` + 输出 `{success:true, data:{batch_id, status:<最后状态>, is_terminal:false, timeout:true, elapsed_sec, ...}}`，调用方判 `is_terminal` 而非 `status` 字面量
+
+**长文本输入**：所有可能超过命令行长度的字段统一 `@文件路径` 前缀（沿用现有 `@file` 约定，详见 ADR-003）。例如 `case create --data @case.json` 走 JSON 解析；`testset set-eval-prompt --prompt @prompt.md` 走纯文本。
+
+**xlsx 二进制输出**：`export --output <path>` 必填路径，stdout 仅承载 JSON 报告 `{success:true, data:{path, bytes}}`，不支持 stdout pipe（避免与 JSON-by-default 冲突）。
+
 ## 输出格式
 
 默认输出 JSON：

package/dist/bin.js

@@ -276,7 +276,7 @@ function toExitCode(err) {
   return 1;
 }
 function createRequest(globalTimeout) {
-  return async function request(baseUrl, path4, options) {
+  return async function request(baseUrl, path5, options) {
     const headers = {
       "Content-Type": "application/json",
       ...options.headers
@@ -304,7 +304,7 @@ function createRequest(globalTimeout) {
     if (workspaceId) {
       headers["workspace-id"] = workspaceId;
     }
-    let url = `${baseUrl}${path4}`;
+    let url = `${baseUrl}${path5}`;
     if (options.query) {
       const params = new URLSearchParams();
       for (const [key, value] of Object.entries(options.query)) {
@@ -3447,6 +3447,645 @@ function registerSACommand(program2) {
   });
 }
 
+// src/commands/testset.ts
+import fs7 from "fs";
+
+// src/utils/file-output.ts
+import fs6 from "fs";
+import path4 from "path";
+async function writeBinaryToFile(filePath, buffer) {
+  try {
+    const dir = path4.dirname(filePath);
+    fs6.mkdirSync(dir, { recursive: true });
+    fs6.writeFileSync(filePath, buffer);
+  } catch (err) {
+    const msg = err instanceof Error ? err.message : String(err);
+    throw new APIError(1, `\u6587\u4EF6\u5199\u5165\u5931\u8D25: ${msg}`);
+  }
+  return { path: filePath, bytes: buffer.length };
+}
+
+// src/client/testset-api.ts
+function unwrapPaginated(raw, fallbackPageSize) {
+  return {
+    items: Array.isArray(raw?.data) ? raw?.data : [],
+    total: typeof raw?.total === "number" ? raw.total : 0,
+    page: typeof raw?.page_no === "number" ? raw.page_no : 1,
+    pageSize: typeof raw?.page_size === "number" ? raw.page_size : fallbackPageSize
+  };
+}
+async function listTestSets(query) {
+  const request = createRequest();
+  const raw = await request(
+    getCustomerServiceUrl(),
+    "/v1/test_sets",
+    {
+      method: "GET",
+      query: {
+        customer_agent_config_id: query.customerAgentConfigId,
+        keyword: query.keyword,
+        page: query.page,
+        page_size: query.pageSize
+      }
+    }
+  );
+  return unwrapPaginated(raw, query.pageSize ?? 20);
+}
+async function getTestSet(id) {
+  const request = createRequest();
+  return request(getCustomerServiceUrl(), `/v1/test_sets/${id}`, {
+    method: "GET"
+  });
+}
+async function updateTestSet(id, body) {
+  const request = createRequest();
+  return request(getCustomerServiceUrl(), `/v1/test_sets/${id}`, {
+    method: "PUT",
+    body
+  });
+}
+async function deleteTestSet(id) {
+  const request = createRequest();
+  return request(getCustomerServiceUrl(), `/v1/test_sets/${id}`, {
+    method: "DELETE"
+  });
+}
+async function duplicateTestSet(id, body) {
+  const request = createRequest();
+  return request(getCustomerServiceUrl(), `/v1/test_sets/${id}/duplicate`, {
+    method: "POST",
+    body
+  });
+}
+async function getEvalPrompt(id) {
+  const request = createRequest();
+  return request(getCustomerServiceUrl(), `/v1/test_sets/${id}/evaluate-prompt`, {
+    method: "GET"
+  });
+}
+async function setEvalPrompt(id, body) {
+  const request = createRequest();
+  return request(getCustomerServiceUrl(), `/v1/test_sets/${id}/evaluate-prompt`, {
+    method: "PUT",
+    body
+  });
+}
+async function listCases(testSetId, query = {}) {
+  const request = createRequest();
+  const raw = await request(
+    getCustomerServiceUrl(),
+    `/v1/test_sets/${testSetId}/cases`,
+    {
+      method: "GET",
+      query: {
+        page: query.page,
+        page_size: query.pageSize
+      }
+    }
+  );
+  return unwrapPaginated(raw, query.pageSize ?? 20);
+}
+async function createCase(testSetId, body) {
+  const request = createRequest();
+  return request(getCustomerServiceUrl(), `/v1/test_sets/${testSetId}/cases`, {
+    method: "POST",
+    body
+  });
+}
+async function updateCase(testSetId, caseId, body) {
+  const request = createRequest();
+  return request(getCustomerServiceUrl(), `/v1/test_sets/${testSetId}/cases/${caseId}`, {
+    method: "PUT",
+    body
+  });
+}
+async function deleteCase(testSetId, caseId) {
+  const request = createRequest();
+  return request(getCustomerServiceUrl(), `/v1/test_sets/${testSetId}/cases/${caseId}`, {
+    method: "DELETE"
+  });
+}
+async function listBatches(query) {
+  const request = createRequest();
+  const raw = await request(
+    getCustomerServiceUrl(),
+    "/v1/test_sets/execution_batches",
+    {
+      method: "GET",
+      query: {
+        test_set_id: query.testSetId,
+        customer_agent_config_id: query.customerAgentConfigId,
+        status: query.status,
+        keyword: query.keyword,
+        page: query.page,
+        page_size: query.pageSize
+      }
+    }
+  );
+  return unwrapPaginated(raw, query.pageSize ?? 20);
+}
+async function getBatch(batchId) {
+  const request = createRequest();
+  return request(getCustomerServiceUrl(), `/v1/test_sets/execution_batches/${batchId}`, {
+    method: "GET"
+  });
+}
+async function getBatchStatus(batchId) {
+  const request = createRequest();
+  return request(
+    getCustomerServiceUrl(),
+    `/v1/test_sets/execution_batches/${batchId}/status`,
+    {
+      method: "GET"
+    }
+  );
+}
+async function executeBatch(testSetId, customerAgentConfigId) {
+  const request = createRequest();
+  return request(getCustomerServiceUrl(), "/v1/test_sets/execute", {
+    method: "POST",
+    body: {
+      test_set_id: testSetId,
+      customer_agent_config_id: customerAgentConfigId
+    }
+  });
+}
+async function rerunCase(batchId, caseId) {
+  const request = createRequest();
+  return request(
+    getCustomerServiceUrl(),
+    `/v1/test_sets/execution_batches/${batchId}/cases/${caseId}/rerun`,
+    {
+      method: "POST"
+    }
+  );
+}
+function buildAuthAndWorkspaceHeaders() {
+  const creds = readCredentials();
+  if (!creds) {
+    throw new APIError(2, "\u672A\u767B\u5F55\uFF0C\u8BF7\u8FD0\u884C: cs-cli auth login");
+  }
+  if (isTokenExpired(creds.expiresAt)) {
+    clearCredentials();
+    throw new APIError(2, "Token \u5DF2\u8FC7\u671F\uFF0C\u8BF7\u8FD0\u884C: cs-cli auth login");
+  }
+  const envLock = readEnvLockState();
+  const config = readConfig();
+  let workspaceId;
+  if (envLock.workspaceId) {
+    assertNoWorkspaceOverride(getRuntimeWorkspaceId());
+    workspaceId = envLock.workspaceId;
+  } else {
+    workspaceId = getRuntimeWorkspaceId() ?? config?.defaultWorkspaceId;
+  }
+  if (!workspaceId) {
+    throw new APIError(1, "\u672A\u8BBE\u7F6E\u5DE5\u4F5C\u7A7A\u95F4\uFF0C\u8BF7\u8FD0\u884C: cs-cli config set-workspace <id>");
+  }
+  return {
+    Authorization: `Bearer ${creds.accessToken}`,
+    "workspace-id": workspaceId
+  };
+}
+async function exportBatchToFile(testSetId, batchId, outputPath) {
+  const headers = buildAuthAndWorkspaceHeaders();
+  const baseUrl = getCustomerServiceUrl();
+  const timeoutMs = getRuntimeRequestTimeoutMs() ?? 6e4;
+  const url = `${baseUrl}/v1/test_sets/${testSetId}/export?batch_id=${encodeURIComponent(batchId)}`;
+  const response = await fetch(url, {
+    method: "GET",
+    headers,
+    signal: AbortSignal.timeout(timeoutMs)
+  });
+  if (!response.ok) {
+    if (response.status === 401) clearCredentials();
+    throw new APIError(response.status, `\u5BFC\u51FA\u5931\u8D25 HTTP ${response.status}`);
+  }
+  const contentType = response.headers.get("content-type") ?? "";
+  if (!contentType.toLowerCase().includes("spreadsheetml")) {
+    throw new APIError(1, `\u5BFC\u51FA\u54CD\u5E94 Content-Type \u975E xlsx: ${contentType || "<\u7A7A>"}`);
+  }
+  const arrayBuf = await response.arrayBuffer();
+  const buf = Buffer.from(arrayBuf);
+  await writeBinaryToFile(outputPath, buf);
+  return { path: outputPath, bytes: buf.byteLength };
+}
+
+// src/commands/batch.ts
+function registerBatchCommand(rootProgram, parent = rootProgram) {
+  const batch = parent.command("batch").description(
+    "\u6267\u884C\u6279\u6B21\uFF08ExecutionBatch\uFF09\u7BA1\u7406 \u2014\u2014 \u6D4B\u8BD5\u96C6\u56DE\u5F52\u6267\u884C\u540E\u7684\u6279\u6B21\u5217\u8868 / \u8BE6\u60C5 / \u72B6\u6001\u67E5\u8BE2\u3002"
+  );
+  batch.command("list").description("\u5217\u51FA\u6267\u884C\u6279\u6B21\uFF08--testset \u53EF\u9009\uFF1B\u7F3A\u7701\u5217\u5F53\u524D workspace \u5168\u90E8\u6279\u6B21\uFF09").option("--testset <id>", "\u6309\u6D4B\u8BD5\u96C6\u7B5B\u9009").option("--customer-agent-config-id <id>", "\u6309 Agent \u914D\u7F6E\u7B5B\u9009").option("--status <status>", "\u6309\u72B6\u6001\u7B5B\u9009\uFF08\u5982 running / finished / failed\uFF09").option("--keyword <text>", "\u6309\u5173\u952E\u8BCD\u6A21\u7CCA\u641C\u7D22").option("--page <number>", "\u9875\u7801", "1").option("--page-size <number>", "\u6BCF\u9875\u6570\u91CF", "20").action(async (opts) => {
+    try {
+      const result = await listBatches({
+        testSetId: opts.testset,
+        customerAgentConfigId: opts.customerAgentConfigId,
+        status: opts.status,
+        keyword: opts.keyword,
+        page: Number(opts.page),
+        pageSize: Number(opts.pageSize)
+      });
+      formatOutput(
+        {
+          success: true,
+          data: result.items,
+          pagination: {
+            page: result.page,
+            pageSize: result.pageSize,
+            total: result.total
+          }
+        },
+        rootProgram.opts().table
+      );
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  batch.command("show").description("\u67E5\u770B\u6279\u6B21\u8BE6\u60C5\uFF08summary\uFF09").argument("<batch_id>", "\u6279\u6B21 ID").action(async (batchId) => {
+    try {
+      const data = await getBatch(batchId);
+      formatOutput({ success: true, data }, rootProgram.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  batch.command("status").description("\u67E5\u770B\u6279\u6B21\u5B9E\u65F6\u72B6\u6001\uFF08\u542B is_terminal \u5B57\u6BB5\uFF09").argument("<batch_id>", "\u6279\u6B21 ID").action(async (batchId) => {
+    try {
+      const data = await getBatchStatus(batchId);
+      formatOutput({ success: true, data }, rootProgram.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+}
+
+// src/commands/case.ts
+var SHOW_PAGE_SIZE = 200;
+var SHOW_MAX_PAGES = 1e3;
+function registerCaseCommand(rootProgram, parent = rootProgram) {
+  const caseCmd = parent.command("case").description(
+    "\u6D4B\u8BD5\u7528\u4F8B\uFF08TestCase\uFF09\u7BA1\u7406 \u2014\u2014 \u5355\u6761 case \u7684 CRUD\uFF0C\u6240\u6709\u5B50\u547D\u4EE4\u9700\u8981 --testset \u951A\u5B9A\u5F52\u5C5E\u6D4B\u8BD5\u96C6\u3002"
+  );
+  caseCmd.command("list").description("\u5217\u51FA\u6307\u5B9A\u6D4B\u8BD5\u96C6\u4E0B\u7684\u7528\u4F8B").requiredOption("--testset <id>", "\u6D4B\u8BD5\u96C6 ID\uFF08\u5FC5\u586B\uFF09").option("--page <number>", "\u9875\u7801", "1").option("--page-size <number>", "\u6BCF\u9875\u6570\u91CF", "20").action(async (opts) => {
+    try {
+      const result = await listCases(opts.testset, {
+        page: Number(opts.page),
+        pageSize: Number(opts.pageSize)
+      });
+      formatOutput(
+        {
+          success: true,
+          data: result.items,
+          pagination: {
+            page: result.page,
+            pageSize: result.pageSize,
+            total: result.total
+          }
+        },
+        rootProgram.opts().table
+      );
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  caseCmd.command("show").description("\u67E5\u770B\u5355\u6761\u7528\u4F8B\u8BE6\u60C5\uFF08\u5185\u90E8\u7528 list+filter \u5B9E\u73B0\uFF0C\u81EA\u52A8\u7FFB\u9875\uFF09").argument("<case_id>", "\u7528\u4F8B ID").requiredOption("--testset <id>", "\u6D4B\u8BD5\u96C6 ID\uFF08\u5FC5\u586B\uFF09").action(async (caseId, opts) => {
+    try {
+      let page = 1;
+      let accumulated = 0;
+      while (page <= SHOW_MAX_PAGES) {
+        const result = await listCases(opts.testset, {
+          page,
+          pageSize: SHOW_PAGE_SIZE
+        });
+        const items = result.items ?? [];
+        const hit = items.find((it) => it?.case_id === caseId);
+        if (hit) {
+          formatOutput({ success: true, data: hit }, rootProgram.opts().table);
+          return;
+        }
+        accumulated += items.length;
+        const isLastPage = items.length === 0 || accumulated >= result.total;
+        if (isLastPage) break;
+        page += 1;
+      }
+      outputError(1, `\u7528\u4F8B ${caseId} \u4E0D\u5B58\u5728`);
+      process.exit(1);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  caseCmd.command("create").description("\u65B0\u5EFA\u7528\u4F8B").requiredOption("--testset <id>", "\u6D4B\u8BD5\u96C6 ID\uFF08\u5FC5\u586B\uFF09").requiredOption("--data <json|@file>", "\u7528\u4F8B JSON body \u6216 @file \u8DEF\u5F84\uFF08\u5FC5\u586B\uFF09").action(async (opts) => {
+    try {
+      const body = parseDataOption(opts.data);
+      const data = await createCase(opts.testset, body);
+      formatOutput({ success: true, data }, rootProgram.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  caseCmd.command("update").description("\u66F4\u65B0\u7528\u4F8B").argument("<case_id>", "\u7528\u4F8B ID").requiredOption("--testset <id>", "\u6D4B\u8BD5\u96C6 ID\uFF08\u5FC5\u586B\uFF09").requiredOption("--data <json|@file>", "\u66F4\u65B0 JSON body \u6216 @file \u8DEF\u5F84\uFF08\u5FC5\u586B\uFF09").action(async (caseId, opts) => {
+    try {
+      const body = parseDataOption(opts.data);
+      const data = await updateCase(opts.testset, caseId, body);
+      formatOutput({ success: true, data }, rootProgram.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  caseCmd.command("delete").description("\u5220\u9664\u7528\u4F8B\uFF08\u4E0D\u5F39\u4E8C\u6B21\u786E\u8BA4\uFF09").argument("<case_id>", "\u7528\u4F8B ID").requiredOption("--testset <id>", "\u6D4B\u8BD5\u96C6 ID\uFF08\u5FC5\u586B\uFF09").action(async (caseId, opts) => {
+    try {
+      const data = await deleteCase(opts.testset, caseId);
+      formatOutput({ success: true, data }, rootProgram.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+}
+
+// src/commands/export.ts
+function registerExportCommand(rootProgram, parent = rootProgram) {
+  parent.command("export").description(
+    "\u5BFC\u51FA\u6279\u6B21\u7ED3\u679C\u4E3A xlsx \u6587\u4EF6\uFF08GET /test_sets/{id}/export?batch_id=<id>\uFF0C\u843D\u76D8\u5230 --output\uFF09"
+  ).requiredOption("--testset <id>", "\u6D4B\u8BD5\u96C6 ID\uFF08\u5FC5\u586B\uFF09").requiredOption("--batch <id>", "\u6279\u6B21 ID\uFF08\u5FC5\u586B\uFF09").requiredOption("--output <path>", "\u843D\u76D8\u6587\u4EF6\u8DEF\u5F84\uFF08\u5FC5\u586B\uFF0C\u7236\u76EE\u5F55\u81EA\u52A8 mkdir -p\uFF09").action(async (opts) => {
+    try {
+      const result = await exportBatchToFile(opts.testset, opts.batch, opts.output);
+      formatOutput({ success: true, data: result }, rootProgram.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+}
+
+// src/commands/run-case.ts
+function registerRunCaseCommand(rootProgram, parent = rootProgram) {
+  parent.command("run-case").description(
+    "\u91CD\u8DD1\u6307\u5B9A\u6279\u6B21\u5185\u7684\u5355\u6761\u7528\u4F8B\uFF08POST /test_sets/execution_batches/{B}/cases/{C}/rerun\uFF09"
+  ).requiredOption("--batch <id>", "\u6279\u6B21 ID\uFF08\u5FC5\u586B\uFF09").requiredOption("--case <id>", "\u7528\u4F8B ID\uFF08\u5FC5\u586B\uFF09").action(async (opts) => {
+    try {
+      const data = await rerunCase(opts.batch, opts.case);
+      formatOutput({ success: true, data }, rootProgram.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+}
+
+// src/utils/poll.ts
+var defaultSleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
+async function pollUntilDone(opts) {
+  const {
+    intervalMs,
+    timeoutSec,
+    fetch: fetchFn,
+    isDone,
+    maxConsecutiveFailures = 3,
+    now = Date.now,
+    sleep = defaultSleep
+  } = opts;
+  const start = now();
+  let lastResult;
+  let lastError;
+  let consecutiveFailures = 0;
+  while (true) {
+    const elapsedMs = now() - start;
+    if (elapsedMs >= timeoutSec * 1e3) {
+      return {
+        done: false,
+        timedOut: true,
+        lastResult,
+        elapsedSec: Math.floor(elapsedMs / 1e3)
+      };
+    }
+    try {
+      const result = await fetchFn();
+      lastResult = result;
+      consecutiveFailures = 0;
+      if (isDone(result)) {
+        return { done: true, result };
+      }
+    } catch (err) {
+      lastError = err;
+      consecutiveFailures += 1;
+      if (consecutiveFailures >= maxConsecutiveFailures) {
+        return { done: false, failed: true, lastError };
+      }
+    }
+    const afterFetchElapsedMs = now() - start;
+    if (afterFetchElapsedMs >= timeoutSec * 1e3) {
+      return {
+        done: false,
+        timedOut: true,
+        lastResult,
+        elapsedSec: Math.floor(afterFetchElapsedMs / 1e3)
+      };
+    }
+    await sleep(intervalMs);
+  }
+}
+
+// src/commands/run.ts
+var DEFAULT_POLL_INTERVAL_MS = 3e3;
+var DEFAULT_TIMEOUT_SEC = 600;
+function extractBatchId(resp) {
+  if (resp && typeof resp === "object" && "batch_id" in resp) {
+    const v = resp.batch_id;
+    if (typeof v === "string") return v;
+  }
+  return "";
+}
+function registerRunCommand(rootProgram, parent = rootProgram) {
+  parent.command("run").description("\u89E6\u53D1\u6D4B\u8BD5\u96C6\u56DE\u5F52\u6267\u884C\uFF08\u9ED8\u8BA4\u5F02\u6B65\uFF1B--wait \u5207\u540C\u6B65\u8F6E\u8BE2\uFF0C\u8D85\u65F6\u4ECD exit 0 / ADR-002\uFF09").requiredOption("--testset <id>", "\u6D4B\u8BD5\u96C6 ID\uFF08\u5FC5\u586B\uFF09").requiredOption("--agent <id>", "Agent \u914D\u7F6E ID\uFF08\u5FC5\u586B\uFF0C\u5BF9\u5E94\u540E\u7AEF customer_agent_config_id\uFF09").option("--wait", "\u540C\u6B65\u7B49\u5F85\u7EC8\u6001\uFF08poll /status\uFF0C3s \u95F4\u9694\uFF0C--timeout \u63A7\u5236\u4E0A\u9650\uFF0C\u5355\u4F4D\uFF1A\u79D2\uFF09", false).option(
+    "--timeout <sec>",
+    "\u540C\u6B65\u7B49\u5F85\u4E0A\u9650\uFF08\u5355\u4F4D\uFF1A\u79D2\uFF1B\u4E0E\u5168\u5C40 --request-timeout \u7684\u6BEB\u79D2\u4E0D\u540C\uFF09",
+    String(DEFAULT_TIMEOUT_SEC)
+  ).action(async (opts) => {
+    let exitCode = 0;
+    try {
+      const triggerResp = await executeBatch(opts.testset, opts.agent);
+      const batchId = extractBatchId(triggerResp);
+      if (!opts.wait) {
+        formatOutput({ success: true, data: { batch_id: batchId } }, rootProgram.opts().table);
+      } else {
+        const timeoutSec = Number(opts.timeout) || DEFAULT_TIMEOUT_SEC;
+        const startedAt = Date.now();
+        const pollResult = await pollUntilDone({
+          intervalMs: DEFAULT_POLL_INTERVAL_MS,
+          timeoutSec,
+          fetch: () => getBatchStatus(batchId),
+          isDone: (r) => r?.is_terminal === true
+        });
+        if (pollResult.done) {
+          const elapsedSec = Math.floor((Date.now() - startedAt) / 1e3);
+          formatOutput(
+            {
+              success: true,
+              data: {
+                ...pollResult.result,
+                elapsed_sec: elapsedSec
+              }
+            },
+            rootProgram.opts().table
+          );
+        } else if ("timedOut" in pollResult && pollResult.timedOut) {
+          const last = pollResult.lastResult ?? {};
+          formatOutput(
+            {
+              success: true,
+              data: {
+                batch_id: batchId,
+                status: last.status,
+                is_terminal: false,
+                timeout: true,
+                elapsed_sec: pollResult.elapsedSec,
+                total_count: last.total_count,
+                pass_count: last.pass_count,
+                fail_count: last.fail_count
+              }
+            },
+            rootProgram.opts().table
+          );
+        } else if ("failed" in pollResult && pollResult.failed) {
+          reportCaughtError(pollResult.lastError);
+          const mapped = toExitCode(pollResult.lastError);
+          exitCode = mapped === 1 ? 3 : mapped;
+        }
+      }
+    } catch (err) {
+      reportCaughtError(err);
+      exitCode = toExitCode(err);
+    }
+    process.exit(exitCode);
+  });
+}
+
+// src/commands/testset.ts
+function readPromptInput(value) {
+  if (value.startsWith("@")) {
+    const filePath = value.slice(1);
+    if (!filePath) {
+      throw new Error("File path cannot be empty after @");
+    }
+    return fs7.readFileSync(filePath, "utf-8");
+  }
+  return value;
+}
+function registerTestsetCommand(program2) {
+  const testset = program2.command("testset").description(
+    "\u6D4B\u8BD5\u96C6\uFF08TestSet\uFF09\u7BA1\u7406 \u2014\u2014 \u5BF9\u8BDD\u56DE\u5F52\u6D4B\u8BD5\u7528\u4F8B\u96C6\u5408\uFF0C\u914D\u5408 batch / run / export \u5F62\u6210\u95ED\u73AF\u3002"
+  );
+  testset.command("list").description("\u5217\u51FA\u6D4B\u8BD5\u96C6").requiredOption("--customer-agent-config-id <id>", "Agent \u914D\u7F6E ID\uFF08\u5FC5\u586B\uFF0C\u4E0E\u540E\u7AEF\u5951\u7EA6\u5BF9\u9F50\uFF09").option("--keyword <text>", "\u6309\u540D\u79F0\u5173\u952E\u8BCD\u6A21\u7CCA\u641C\u7D22").option("--page <number>", "\u9875\u7801", "1").option("--page-size <number>", "\u6BCF\u9875\u6570\u91CF", "20").action(async (opts) => {
+    try {
+      const result = await listTestSets({
+        customerAgentConfigId: opts.customerAgentConfigId,
+        keyword: opts.keyword,
+        page: Number(opts.page),
+        pageSize: Number(opts.pageSize)
+      });
+      formatOutput(
+        {
+          success: true,
+          data: result.items,
+          pagination: {
+            page: result.page,
+            pageSize: result.pageSize,
+            total: result.total
+          }
+        },
+        program2.opts().table
+      );
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  testset.command("show").description("\u67E5\u770B\u5355\u4E2A\u6D4B\u8BD5\u96C6\u8BE6\u60C5").argument("<id>", "\u6D4B\u8BD5\u96C6 ID").action(async (id) => {
+    try {
+      const data = await getTestSet(id);
+      formatOutput({ success: true, data }, program2.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  testset.command("update").description("\u66F4\u65B0\u6D4B\u8BD5\u96C6\uFF08\u81F3\u5C11\u63D0\u4F9B --name / --description / --data \u4E4B\u4E00\uFF1B--data \u4F18\u5148\u5408\u5E76\uFF09").argument("<id>", "\u6D4B\u8BD5\u96C6 ID").option("--name <text>", "\u6D4B\u8BD5\u96C6\u540D\u79F0").option("--description <text>", "\u63CF\u8FF0").option("--data <json|@file>", "\u5B8C\u6574 JSON body\uFF08\u8986\u76D6 --name / --description\uFF09").action(async (id, opts) => {
+    try {
+      let body = {};
+      if (opts.name !== void 0) body.test_set_name = opts.name;
+      if (opts.description !== void 0) body.description = opts.description;
+      if (opts.data !== void 0) {
+        const parsed = parseDataOption(opts.data);
+        body = { ...body, ...parsed };
+      }
+      if (Object.keys(body).length === 0) {
+        outputError(1, "\u81F3\u5C11\u63D0\u4F9B --name/--description/--data \u4E4B\u4E00");
+        process.exit(1);
+      }
+      const data = await updateTestSet(id, body);
+      formatOutput({ success: true, data }, program2.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  testset.command("delete").description("\u5220\u9664\u6D4B\u8BD5\u96C6\uFF08\u76F4\u63A5\u6267\u884C\uFF0C\u4E0D\u5F39\u4E8C\u6B21\u786E\u8BA4\uFF09").argument("<id>", "\u6D4B\u8BD5\u96C6 ID").action(async (id) => {
+    try {
+      const data = await deleteTestSet(id);
+      formatOutput({ success: true, data }, program2.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  testset.command("copy").description("\u590D\u5236\u6D4B\u8BD5\u96C6").argument("<id>", "\u6E90\u6D4B\u8BD5\u96C6 ID").option("--name <text>", "\u526F\u672C\u540D\u79F0").action(async (id, opts) => {
+    try {
+      const data = await duplicateTestSet(id, { test_set_name: opts.name });
+      formatOutput({ success: true, data }, program2.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  testset.command("get-eval-prompt").description("\u67E5\u770B\u6D4B\u8BD5\u96C6\u7684\u8BC4\u4EF7\u63D0\u793A\u8BCD").argument("<id>", "\u6D4B\u8BD5\u96C6 ID").option("--raw", "\u88F8\u6587\u672C\u8F93\u51FA\u5230 stdout\uFF08\u4E0D\u5E26 JSON \u5305\u88C5\uFF09", false).action(async (id, opts) => {
+    try {
+      const data = await getEvalPrompt(id);
+      if (opts.raw) {
+        const text = data?.prompt_template ?? "";
+        process.stdout.write(text);
+        return;
+      }
+      formatOutput({ success: true, data }, program2.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  testset.command("set-eval-prompt").description("\u8BBE\u7F6E\u6D4B\u8BD5\u96C6\u7684\u8BC4\u4EF7\u63D0\u793A\u8BCD\uFF08\u652F\u6301 @file \u8BFB\u7EAF\u6587\u672C\uFF09").argument("<id>", "\u6D4B\u8BD5\u96C6 ID").requiredOption("--prompt <text|@file>", "\u63D0\u793A\u8BCD\u5B57\u9762\u91CF\u6216 @file \u8DEF\u5F84").action(async (id, opts) => {
+    try {
+      const prompt = readPromptInput(opts.prompt);
+      const data = await setEvalPrompt(id, { prompt });
+      formatOutput({ success: true, data }, program2.opts().table);
+    } catch (err) {
+      reportCaughtError(err);
+      process.exit(toExitCode(err));
+    }
+  });
+  registerCaseCommand(program2, testset);
+  registerBatchCommand(program2, testset);
+  registerRunCommand(program2, testset);
+  registerRunCaseCommand(program2, testset);
+  registerExportCommand(program2, testset);
+}
+
 // src/commands/workspace.ts
 var DATE_RE = /^\d{4}-\d{2}-\d{2}$/;
 function toISODate(d) {
@@ -3533,6 +4172,7 @@ registerMonitorCommand(program);
 registerRepairRecordCommand(program);
 registerOperationsRecordCommand(program);
 registerChangeConsumerCommand(program);
+registerTestsetCommand(program);
 process.on("uncaughtException", (err) => {
   outputError(3, err.message);
   process.exit(3);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bty/customer-service-cli",
-  "version": "0.4.7",
+  "version": "0.4.8",
   "description": "AI Customer Service CLI - Agent friendly",
   "type": "module",
   "main": "./dist/bin.js",