feat: 接入首周运行设置并修复 limits 部分更新

README.md

@@ -147,7 +147,7 @@ docker start iris
 | `PUT /settings/models/{chat\|embedding}` | 保存模型；api_key 省略保留，空字符串清除，enabled=false 停用 |
 | `POST /settings/models/{chat\|embedding}/test` | 空 JSON 测已保存配置，模型 JSON 测草稿 |
 | `POST /settings/models/{chat\|embedding}/retry` | 对暂时不可用／账户问题请求重试 |
-| `PATCH /settings/limits` | daily_token_limit（null 不限）、learning_concurrency、learning_image_limit、native_image_learning、media_retention_days（默认 7，整数 1—36500） |
+| `PATCH /settings/limits` | daily_token_limit（null 不限）、learning_concurrency、learning_image_limit、native_image_learning、media_retention_days（默认 7，整数 1—36500）；所有字段省略即保留现值，daily_token_limit 显式 null 才表示不限；操作记录只包含实际提交的字段 |
 
 ## 中文试用与管理界面
 

frontend/src/Access.test.tsx

@@ -160,6 +160,8 @@ test.each(["high", "medium"])(
     await user.click(screen.getByRole("button", { name: "测试对话模型连接" }));
     expect(await screen.findByText(/密钥无效.*12/)).toBeInTheDocument();
     await user.click(screen.getByRole("button", { name: "保存对话模型" }));
+    await user.clear(screen.getByLabelText("学习并发"));
+    await user.type(screen.getByLabelText("学习并发"), "3");
     await user.click(screen.getByRole("button", { name: "保存用量与学习" }));
     const saved = calls.find(
       (c) =>

frontend/src/ConsolidationReport.tsx

@@ -9,6 +9,8 @@ import { Badge, Pagination, RoleTime, seconds } from "./ui";
 import { PersonaChecks } from "./PersonaDetails";
 import { personaDegrees } from "./persona-labels";
 import {
+  consolidationBudgets,
+  consolidationSteps,
   consolidationReason,
   consolidationSwitches,
   suggestionKinds,
@@ -334,6 +336,80 @@ function PersonSummaryReport({ data }: { data: MaintenanceReport }) {
   );
 }
 
+function StepBudgetReport({ data }: { data: MaintenanceReport }) {
+  const budgets = data.items.filter(
+    (item) => item.outcome === "model_step_budget",
+  );
+  const trimmed = data.items.filter(
+    (item) => item.outcome === "backlog_trimmed",
+  );
+  const expiry = data.question_expiry;
+  return (
+    <>
+      {(data.consolidation || budgets.length > 0) && (
+        <section
+          className="detail-section"
+          aria-label="各步骤模型调用上限与用量"
+        >
+          <h3>各步骤模型调用上限与用量</h3>
+          <p className="fine-print">
+            每一步独立计数，包含重试和修正；上限沿用本次运行接受时的设置。未到期、关闭或没有任务的步骤可能没有记录。
+          </p>
+          {!budgets.length && <p className="quiet">本次未记录步骤用量。</p>}
+          {budgets.map((item) => (
+            <article
+              className="maintenance-item"
+              key={item.item_key}
+              aria-label={
+                consolidationSteps[item.details.step || ""] ||
+                item.details.step ||
+                "未知步骤"
+              }
+            >
+              <h4>
+                {consolidationSteps[item.details.step || ""] ||
+                  item.details.step ||
+                  "未知步骤"}
+              </h4>
+              <p>
+                本轮上限：{item.details.max_calls ?? "未记录"} 次 · 已用：
+                {item.details.calls_used ?? "未记录"} 次
+              </p>
+              <p>
+                停止原因：
+                {item.reason ? consolidationReason(item.reason) : "未记录"}
+              </p>
+            </article>
+          ))}
+        </section>
+      )}
+      {!!trimmed.length && (
+        <section className="detail-section" aria-label="配对积压修剪">
+          <h3>配对积压修剪</h3>
+          {trimmed.map((item) => (
+            <p key={item.item_key}>
+              配对积压修剪：{item.details.count ?? "未记录"} 项 ·
+              待处理配对上限：{item.details.limit ?? "未记录"} 项
+            </p>
+          ))}
+          <p className="fine-print">
+            仅修剪待处理配对，不改变记忆；相关记忆发生变化后可重新规划。
+          </p>
+        </section>
+      )}
+      {expiry && (
+        <section className="detail-section" aria-label="询问到期结束">
+          <h3>询问到期结束</h3>
+          <p>到期自动结束的询问：{expiry.count} 条</p>
+          <p className="fine-print">
+            仅学习产生且超过保留天数仍未结束的询问；宿主或管理员创建的询问和普通目标不受影响。
+          </p>
+        </section>
+      )}
+    </>
+  );
+}
+
 export function ConsolidationReportSections({
   data,
   openMemory,
@@ -362,17 +438,18 @@ export function ConsolidationReportSections({
               </p>
               <p className="fine-print">
                 {calls.unknown_usage_calls ?? 0}{" "}
-                次调用用量未知；用量为已知合计，包含实际重试、persona
+                次调用用量未知；用量为四步骤已知合计，包含实际重试、入口画像、persona
                 及人物摘要调用。
               </p>
             </>
           )}
           <details>
             <summary>本次接受任务时的整理设置</summary>
-            <p>
-              模型调用次数预算：
-              {data.consolidation.settings.max_calls ?? "未记录"}
-            </p>
+            {consolidationBudgets.map(([key, label]) => (
+              <p key={key}>
+                {label}：{data.consolidation?.settings[key] ?? "未记录"}
+              </p>
+            ))}
             {consolidationSwitches.map(([key, label]) => (
               <p key={key}>
                 {label}：
@@ -387,6 +464,7 @@ export function ConsolidationReportSections({
           </details>
         </section>
       )}
+      <StepBudgetReport data={data} />
       {(data.consolidation ||
         data.items.some((i) => i.phase === "consolidation")) && (
         <>

frontend/src/ConsolidationSettings.test.tsx

@@ -78,7 +78,8 @@ beforeEach(() => {
           );
         if (url.endsWith("/consolidation")) {
           settings.consolidation = { ...settings.consolidation, ...body };
-          settings.lifecycle.maintenance_time = body.maintenance_time;
+          if ("maintenance_time" in body)
+            settings.lifecycle.maintenance_time = body.maintenance_time;
         }
         if (url.endsWith("/goal-dedup-judge")) settings.goal_dedup_judge = body;
         settings.operations = [
@@ -107,7 +108,7 @@ test("整理设置保存时间、调用预算与所有开关，并显示操作
     target: { value: "04:15" },
   });
   fireEvent.change(
-    within(section).getByLabelText("每次整理的模型调用次数预算"),
+    within(section).getByLabelText("记忆整理（合并／矛盾／依赖）每轮调用上限"),
     { target: { value: "0" } },
   );
   await userEvent.click(within(section).getByLabelText("矛盾检查与建议"));
@@ -122,12 +123,8 @@ test("整理设置保存时间、调用预算与所有开关，并显示操作
   expect(request.body).toEqual({
     maintenance_time: "04:15",
     max_calls: 0,
-    merge_enabled: true,
     conflict_enabled: false,
-    dependency_enabled: true,
-    persona_enabled: true,
     person_summary_enabled: false,
-    goal_review_enabled: true,
   });
   expect((request.init?.headers as any)["X-Iris-CSRF"]).toBe(
     "consolidation-settings-test",
@@ -136,19 +133,19 @@ test("整理设置保存时间、调用预算与所有开关，并显示操作
   expect(within(section).getByText(/只影响之后接受的新运行/)).toBeVisible();
 });
 
-test.each(["", "51", "1.5"])(
+test.each(["", "-1", "1001", "1.5"])(
   "整理预算 %s 被阻止，未发送请求",
   async (input) => {
     render(<SettingsPage />);
     fireEvent.change(
-      await screen.findByLabelText("每次整理的模型调用次数预算"),
+      await screen.findByLabelText("记忆整理（合并／矛盾／依赖）每轮调用上限"),
       { target: { value: input } },
     );
     await userEvent.click(
       screen.getByRole("button", { name: "保存梦境整理设置" }),
     );
     expect(
-      await screen.findByText(/模型调用次数预算须为 0—50 的整数/),
+      await screen.findByText(/每轮调用上限须为 0—1000 的整数/),
     ).toBeVisible();
     expect(requests.some((r) => r.init?.method === "PATCH")).toBe(false);
   },