min-agent 0.4.1 → 0.5.0

package/dist/tui-chat.js

@@ -9,7 +9,7 @@ import { loadImageParts } from "./agent.js";
 import { createUndoStack, clearStack } from "./tui/undo-stack.js";
 import { killActiveProcesses } from "./tools/bash.js";
 import { setTuiConfirm, setTuiQuestion } from "./confirm.js";
-import { loadConfig, getActiveProvider } from "./config.js";
+import { loadConfig, getActiveProvider, getEffectiveConfig } from "./config.js";
 import { scanProject, buildCodeSystemPrompt } from "./code-mode.js";
 import { handleSlashCommand, sysMsg } from "./tui/slash-handler.js";
 import { createAgentRunner } from "./tui/agent-runner.js";
@@ -17,6 +17,8 @@ import { listSessions, loadSession, saveSession } from "./sessions.js";
 import { hydrateMessages, lastUserText } from "./tui/hydrate.js";
 import { createPromptQueue } from "./tui/prompt-queue.js";
 import { prepareSwitch, switchFailMessage, prepareNew, newFailMessage } from "./tui/session-switch.js";
+import { getModelCatalog } from "./model-catalog.js";
+import { setThinkingEffort, setThinkingOverride, thinkingChoicesForModel, thinkingEffortLabel } from "./thinking.js";
 let confirmResolver = null;
 let questionResolver = null;
 /** Only one confirm/question overlay can be shown at a time; parallel tool calls queue up. */
@@ -26,6 +28,41 @@ function registerSessionCompletion() {
     setSlashArgProvider("sessions", provider);
     setSlashArgProvider("resume", provider);
 }
+function registerThinkCompletion() {
+    const provider = (tokens) => {
+        const flags = [
+            { value: "--project", description: "写入当前项目" },
+            { value: "--global", description: "写入全局配置" },
+        ];
+        const used = new Set(tokens);
+        const remainingFlags = flags.filter((f) => !used.has(f.value));
+        if (tokens.length === 0) {
+            const model = getActiveProvider(getEffectiveConfig())?.defaultModel;
+            return [...thinkingChoicesForModel(model).map((value) => ({ value })), ...remainingFlags];
+        }
+        return remainingFlags;
+    };
+    setSlashArgProvider("think", provider);
+    setSlashArgProvider("thinking", provider);
+}
+function registerMemoryCompletion() {
+    setSlashArgProvider("memory", (tokens) => {
+        const flags = [
+            { value: "--project", description: "写入当前项目" },
+            { value: "--global", description: "写入全局配置" },
+        ];
+        const used = new Set(tokens);
+        const remainingFlags = flags.filter((f) => !used.has(f.value));
+        if (tokens.length === 0) {
+            return [
+                { value: "on", description: "开启记忆（跨会话保留）" },
+                { value: "off", description: "关闭记忆（默认）" },
+                ...remainingFlags,
+            ];
+        }
+        return remainingFlags;
+    });
+}
 function registerPermissionCompletion() {
     setSlashArgProvider("permission", (tokens) => {
         const flags = [
@@ -106,6 +143,8 @@ export async function runTui(opts) {
     discoverSkills();
     registerSkillCompletion();
     registerPermissionCompletion();
+    registerThinkCompletion();
+    registerMemoryCompletion();
     registerSandboxCompletion();
     registerSessionCompletion();
     const instructions = await loadInstructions();
@@ -116,6 +155,8 @@ export async function runTui(opts) {
     let sessionId = resumeSessionId;
     let currentModel = modelId ?? getActiveProvider(config)?.defaultModel;
     const currentProvider = providerName;
+    if (currentModel)
+        void getModelCatalog(currentModel, getActiveProvider(config)?.baseURL);
     const tracker = new TokenTracker();
     const taskState = emptyTaskState();
     let undoStack = createUndoStack();
@@ -131,11 +172,13 @@ export async function runTui(opts) {
     }
     let runner = null;
     const tui = new TuiRenderer({
-        onSubmit: async (text) => {
-            if (text.startsWith("/")) {
+        onSubmit: async (prompt) => {
+            const text = prompt.display;
+            if (text.trim().startsWith("/")) {
                 const result = await handleSlashCommand(text, tui, messages, currentModel, tracker, (model) => {
                     currentModel = model;
                     tui.setModel(model);
+                    void refreshTokenBar();
                 }, requestExit, {
                     undoStack,
                     setUndoStack: (s) => {
@@ -171,12 +214,15 @@ export async function runTui(opts) {
             }
             lastUserMessage = text;
             tui.addMessage({ id: `user-${Date.now()}`, role: "user", content: text, timestamp: Date.now() });
-            messages.push({ role: "user", content: text });
+            messages.push({ role: "user", content: prompt.content });
             if (runner) {
                 await runner.run();
                 persistSession();
             }
         },
+        onNotice: (notice) => {
+            sysMsg(tui, notice);
+        },
         onConfirm: (accepted) => {
             tui.hideConfirm();
             if (confirmResolver) {
@@ -201,11 +247,21 @@ export async function runTui(opts) {
             }
             currentModel = model;
             tui.setModel(model);
+            void refreshTokenBar();
             sysMsg(tui, `✓ 默认模型已设为: ${model}`);
         },
         onModelCancel: () => {
             tui.hideModelPicker();
         },
+        onThinkPick: (effort, scope) => {
+            setThinkingEffort(effort, scope);
+            setThinkingOverride(effort);
+            tui.hideThinkPicker();
+            sysMsg(tui, `✓ 思考强度已设为 ${thinkingEffortLabel(effort)}（${scope === "project" ? "项目" : "全局"}）`);
+        },
+        onThinkCancel: () => {
+            tui.hideThinkPicker();
+        },
         onSessionPick: (id) => {
             tui.hideSessionPicker();
             applySession(id);
@@ -299,7 +355,8 @@ export async function runTui(opts) {
         tracker.reset();
         lastUserMessage = null;
         tui.setSessionId(undefined);
-        tui.update({ messages: [], tokenInfo: undefined });
+        tui.update({ messages: [] });
+        void refreshTokenBar();
         sysMsg(tui, "✓ 已新建会话");
     }
     function openSessionPicker() {
@@ -347,8 +404,7 @@ export async function runTui(opts) {
     tui.start();
     if (messages.length > 0)
         tui.update({ messages: hydrateMessages(messages) });
-    if (messages.length > 0)
-        await refreshTokenBar();
+    await refreshTokenBar();
     const prompt = initialPrompt?.trim() ?? "";
     const notify = (kind, imgPath) => {
         if (kind === "not_found")

package/docs/API.md

@@ -42,14 +42,18 @@ min-agent serve --host 0.0.0.0 --port 3000
 | POST | `/v1/chat/redo` | 重发会话最后一条用户消息 |
 | GET | `/v1/sessions` | 列出会话 |
 | DELETE | `/v1/sessions/:id` | 删除会话 |
-| GET | `/v1/memory` | 列出记忆（含 `project_memories`） |
+| GET | `/v1/memory` | 列出记忆（含开关状态与 `project_memories`） |
 | POST | `/v1/memory` | 添加记忆 |
 | GET | `/v1/memory/search?q=xxx` | 搜索记忆 |
 | DELETE | `/v1/memory/:index` | 删除记忆 |
+| GET | `/v1/memory/mode` | 当前记忆开关 |
+| POST | `/v1/memory/mode` | 设置记忆开关 |
 | GET | `/v1/budget` | 当前预算上限 |
 | POST | `/v1/budget` | 设置预算上限 |
 | GET | `/v1/permission` | 当前确认模式 |
 | POST | `/v1/permission` | 设置确认模式 |
+| GET | `/v1/thinking` | 当前思考强度 |
+| POST | `/v1/thinking` | 设置思考强度 |
 | GET | `/v1/sandbox` | 当前隔离策略 |
 | POST | `/v1/sandbox` | 设置隔离策略 |
 | GET | `/v1/diff` | 工作区未提交变更 |
@@ -86,7 +90,7 @@ min-agent serve --host 0.0.0.0 --port 3000
 ## `GET /v1/meta`
 
 ```json
-{ "version": "0.1.0", "cwd": "/path/to/project", "instructions_chars": 1234, "sandbox": { "mode": "off", "network": "allow", "extraWriteRoots": [], "extraReadRoots": [], "source": null, "label": "未隔离，允许联网" }, "permission": { "permission": "ask", "source": null, "label": "询问确认" } }
+{ "version": "0.1.0", "cwd": "/path/to/project", "instructions_chars": 1234, "sandbox": { "mode": "off", "network": "allow", "extraWriteRoots": [], "extraReadRoots": [], "source": null, "label": "未隔离，允许联网" }, "permission": { "permission": "ask", "source": null, "label": "询问确认" }, "thinking": { "thinking": "medium", "source": null, "label": "medium" }, "memory": { "memory": "off", "source": null, "label": "off" } }
 ```
 
 ## `GET /v1/update`
@@ -157,10 +161,12 @@ CLI 升级请使用 `min-agent update`（执行 `npm install -g min-agent`）。
 | `auto_approve` | boolean | 默认 `true`。设为 `false` 时必须 `stream: true`，并通过 `POST /v1/approvals` / `POST /v1/answers` 回调 |
 | `sandbox` | string | 本轮隔离：`off` / `workspace` / `strict`。只能比服务当前策略更严，不能放宽 |
 | `network` | string | 本轮联网：`allow` / `deny`。`deny` 只能收紧 |
+| `thinking` | string | 本轮思考强度：`off` / `low` / `medium` / `high` / `max`。不传则用配置，未配置时为 `medium` |
+| `memory` | string | 本轮是否使用记忆：`on` / `off`。不传则用配置，未配置时为 `off` |
 
 `auto_approve: false` 且未开流式时返回 `400 auto_approve_requires_stream`。确认/提问最多等待 5 分钟，超时视为拒绝或跳过。客户端断开连接时未完成的确认视为拒绝。
 
-`max_steps_reached` 为 `true` 表示本轮已用尽自动续跑次数（上下文压缩/空回复等），或显式设置了 `agent.maxSteps` 且已到达该上限。默认不按固定步数切断一轮生成。`incomplete` 为 `true` 表示模型在调用工具后没有给出完整回复，或连续多次空回复。`stopped` 为 `true` 表示因同一操作反复执行而主动结束本轮。`empty_response` 为 `true` 表示 provider 连续返回空流（HTTP 200 但没有任何内容，多为上游超时/限流），`empty_attempts` 为已尝试次数；这类空回复会以原请求重发重试（首次立即，之后指数退避），可用 `agent.maxEmptyAttempts`（默认 4）与 `agent.emptyRetryDelayMs`（默认 1000）调整。发送下一条消息即可继续。`continues` 为本轮实际自动续跑次数。`task_state` 为当前任务目标与待办列表，会写入会话并在压缩后保留。网页检索连续过久、或检索后尚未写出用户要求的结果时，代理会改为基于已有资料产出结果；本轮内再调用搜索/抓取会收到停止检索并直接产出的提示，而不是新的检索结果（即使 `autoContinue` 为 `false`）。
+`max_steps_reached` 为 `true` 表示本轮已用尽自动续跑次数（上下文压缩/空回复等），或显式设置了 `agent.maxSteps` 且已到达该上限。默认不按固定步数切断一轮生成。`incomplete` 为 `true` 表示模型在调用工具后没有给出完整回复，或连续多次空回复。`stopped` 为 `true` 表示因同一操作反复执行而主动结束本轮。`empty_response` 为 `true` 表示 provider 连续返回空流（HTTP 200 但没有任何内容，多为上游超时/限流），`empty_attempts` 为已尝试次数；这类空回复会以原请求重发重试（首次立即，之后指数退避），可用 `agent.maxEmptyAttempts`（默认 4）与 `agent.emptyRetryDelayMs`（默认 1000）调整。发送下一条消息即可继续。`continues` 为本轮实际自动续跑次数。`task_state` 为当前任务目标与待办列表，会写入会话并在压缩后保留；纯问候不会记为任务目标。网页检索连续过久、或检索后尚未写出用户要求的结果时，代理会改为基于已有资料产出结果；本轮内再调用搜索/抓取会收到停止检索并直接产出的提示，而不是新的检索结果（即使 `autoContinue` 为 `false`）。
 
 ### 非流式响应
 
@@ -331,6 +337,25 @@ CLI 升级请使用 `min-agent update`（执行 `npm install -g min-agent`）。
 
 `scope` 省略时写入全局。`permission` 必须为 `ask` / `accept-edits` / `allow-all`。
 
+## `GET /v1/thinking`
+
+```json
+{ "thinking": "max", "source": "global", "label": "max" }
+```
+
+`thinking` 为 `off` / `low` / `medium` / `high` / `max`。未配置时为 `medium`，`source` 为 `null`。已配置时 `source` 为 `cli`、`project` 或 `global`。
+
+## `POST /v1/thinking`
+
+```json
+{ "thinking": "max", "scope": "project" }
+```
+```json
+{ "ok": true, "scope": "project", "thinking": "max", "source": "project", "label": "max" }
+```
+
+`scope` 省略时写入全局，并清除项目配置里的 `thinking` 覆盖，避免旧的项目值（例如 `off`）在重启后继续生效。`thinking` 必须为 `off` / `low` / `medium` / `high` / `max`（`none` → `off`，`xhigh` / `extra-high` → `max`）。
+
 ## `GET /v1/sandbox`
 
 ```json
@@ -488,12 +513,33 @@ CLI 升级请使用 `min-agent update`（执行 `npm install -g min-agent`）。
 
 ```json
 {
+  "memory": "off",
+  "source": null,
+  "label": "off",
   "memories": [{ "content": "...", "tags": [...], "created": "..." }],
   "project_memories": [{ "content": "...", "tags": [...], "created": "..." }]
 }
 ```
 
-`memories` 始终为全局；`project_memories` 为当前目录 `.min-agent/memory.json`。
+`memories` 始终为全局；`project_memories` 为当前目录 `.min-agent/memory.json`。`memory` 为 `on` / `off`。未配置时为 `off`，`source` 为 `null`。已配置时 `source` 为 `cli`、`project` 或 `global`。关闭时仍可读写条目，但对话不会注入记忆，也不会暴露 memory 工具。
+
+## `GET /v1/memory/mode`
+
+```json
+{ "memory": "off", "source": null, "label": "off" }
+```
+
+## `POST /v1/memory/mode`
+
+```json
+{ "memory": "on", "scope": "project" }
+```
+
+```json
+{ "ok": true, "scope": "project", "memory": "on", "source": "project", "label": "on" }
+```
+
+`scope` 省略时写入全局，并清除项目配置里的 `memory` 覆盖。`memory` 必须为 `on` / `off`（`true` / `enable` → `on`，`false` / `disable` → `off`）。
 
 ## `POST /v1/memory`