@aiyiran/myclaw 1.0.243 → 1.0.245

package/assets/myclaw-artifacts.js

@@ -290,17 +290,27 @@
     });
   }
 
+  function fetchArtifactsFromServerAPI(wsPrefix) {
+    var url = window.location.origin + '/cmd/api/artifacts?workspace=' + encodeURIComponent(wsPrefix) + '&t=' + Date.now();
+    return fetch(url).then(function (res) {
+      if (!res.ok) throw new Error('HTTP ' + res.status);
+      return res.json();
+    });
+  }
+
   function fetchArtifacts(contentEl) {
+    // cachedConfig 未就绪时跳过，等 initConfig 完成后的 startPolling 重试
+    if (!cachedConfig) return;
+
     var wsPrefix = getWorkspaceId();
     var fetcher;
 
     if (envInfo && envInfo.remote) {
-      // 远程环境 → 走 CDN
-      fetcher = fetchArtifactsFromCDN(wsPrefix);
+      // 远程服务器 → 走 /cmd/api（服务器直接提供 JSON）
+      fetcher = fetchArtifactsFromServerAPI(wsPrefix);
     } else {
-      // 本地环境 → 优先本地 API，失败降级 CDN
-      fetcher = fetchArtifactsFromLocalAPI(wsPrefix)
-        .catch(function () { return fetchArtifactsFromCDN(wsPrefix); });
+      // 本地环境 → 走 CDN
+      fetcher = fetchArtifactsFromCDN(wsPrefix);
     }
 
     fetcher

package/assets/myclaw-inject.js

@@ -27,6 +27,7 @@
   var committedText = "";     // 已经提交到 textarea 的文字（上一轮累积）
   var cursorOffset = 0;       // 录音开始时光标在 textarea 中的位置
   var injected = false;
+  var stopping = false;      // 正在等待最终识别结果（stopVoice 的 2 秒窗口）
 
   // ═══ 1. 右下角版本标签（点击测试麦克风） ═══
   function createVersionBar() {
@@ -323,7 +324,8 @@
       '</svg>',
     ].join("");
 
-    btn.addEventListener("click", function () {
+btn.addEventListener("click", function () {
+      console.log("[myclaw-voice] 按钮点击, recording=", recording);
       if (recording) {
         stopVoice();
       } else {
@@ -367,8 +369,8 @@
 
     voice = new window.VoiceInput({
       onResult: function (text) {
-        // 如果用户已点停止，忽略异步返回的残留结果（防止文字重复）
-        if (!recording) return;
+        // 完全停止后才忽略；stopping 期间（2秒等待窗口）仍允许写入
+        if (!recording && !stopping) return;
         // 讯飞实时返回识别文字，替换到光标位置
         pendingText = text;
         updateTextAtCursor(pendingText);
@@ -377,7 +379,7 @@
       onStatusChange: function (oldStatus, newStatus) {
         console.log("[myclaw-voice] \u72b6\u6001:", oldStatus, "->", newStatus);
 
-        if (newStatus === "idle" && recording) {
+        if (newStatus === "idle" && recording && !stopping) {
           // 讯飞 60 秒断开，但用户没有点停止 → 自动重连
           // 把当前识别的文字提交，并更新光标位置
           committedText = getTextareaValue();
@@ -435,23 +437,54 @@
     console.log("[myclaw-voice] \u5f00\u59cb\u5f55\u97f3\uff0c\u5149\u6807\u4f4d\u7f6e:", cursorOffset);
   }
 
-  function stopVoice() {
-    // 先关标志位，阻止 onResult 异步回调继续写入（核心防重复）
+  
+  /**
+   * 语音录入结束时，等待 2 秒后关闭录音资源
+   * @param {Function} [onDone] - 等待完成后执行的回调（如发送）
+   */
+  var voiceStopTimer = null;
+
+  function stopVoice(onDone) {
+    console.log("[myclaw-voice] stopVoice called, recording=", recording, "onDone=", onDone ? "yes" : "no");
+    if (!recording) {
+      console.log("[myclaw-voice] stopVoice early return — not recording");
+      return;
+    }
+
+    // 进入 stopping 态：UI 立即更新，但 onResult 仍允许在 2 秒内写入文字
+    stopping = true;
     recording = false;
     updateButtonUI();
+    console.log("[myclaw-voice] stopping=true, UI updated, starting 2s timer...");
 
-    // 立即快照当前 textarea 值作为最终文字
-    var finalText = getTextareaValue();
-
-    if (voice) {
-      voice.stop();
+    // 延迟 2 秒后关闭录音资源（等讯飞把剩余识别结果全部推过来）
+    if (voiceStopTimer) {
+      console.log("[myclaw-voice] clearing previous timer");
+      clearTimeout(voiceStopTimer);
     }
+    voiceStopTimer = setTimeout(function () {
+      voiceStopTimer = null;
+      stopping = false;
+      console.log("[myclaw-voice] 2s timer fired, closing resources...");
+
+      // 快照当前 textarea 值（2 秒内 onResult 可能已更新）
+      var finalText = getTextareaValue();
+      console.log("[myclaw-voice] finalText:", JSON.stringify(finalText.substring(0, 50)));
+      if (voice) {
+        console.log("[myclaw-voice] calling voice.stop()");
+        voice.stop();
+      }
+      committedText = finalText;
+      pendingText = "";
 
-    // 用快照覆盖，确保后续异步返回不影响
-    committedText = finalText;
-    pendingText = "";
+      // 等待完成后执行回调（如发送）
+      if (onDone) {
+        console.log("[myclaw-voice] executing onDone callback...");
+        onDone();
+      }
 
-    console.log("[myclaw-voice] \u505c\u6b62\u5f55\u97f3");
+      console.log("[myclaw-voice] 停止录音完成");
+    }, 2000);
   }
 
   // ═══ 5. DOM 注入 ═══
@@ -522,6 +555,28 @@
   }
   // ═══ 6. 拦截发送按钮 ═══
 
+  /**
+   * 拦截 Enter 键：语音态下按回车 → 等待 2 秒后发送
+   */
+  function hookVoiceEnter() {
+    document.addEventListener("keydown", function (e) {
+      if (e.key !== "Enter") return;
+      if (!recording) return;
+
+      // 语音录入中，无论焦点在哪里（textarea 或语音按钮），Enter 统一触发"停止并发送"
+      e.preventDefault();
+      e.stopPropagation();
+
+      console.log("[myclaw-voice] Enter按下, recording=", recording);
+      stopVoice(function () {
+        console.log("[myclaw-voice] Enter stopVoice callback firing...");
+        var sendBtn = document.querySelector("button.chat-send-btn, button[title=\"Send\"]");
+        if (sendBtn) sendBtn.click();
+      });
+
+    }, true);
+  }
+
   var sendHooked = false;
 
   function hookSendButton() {
@@ -536,9 +591,17 @@
       var text = getTextareaValue();
       if (!text || !text.trim()) return;  // 空文字不处理
 
-      // 1) 停止语音输入
+      // 1) 停止语音输入（等待 2 秒后关闭，关闭后触发发送）
       if (recording) {
-        stopVoice();
+        e.preventDefault();
+        e.stopPropagation();
+        console.log("[myclaw-voice] 发送按钮点击(语音态), recording=", recording);
+        stopVoice(function () {
+          console.log("[myclaw-voice] 发送按钮 stopVoice callback firing...");
+          var sendBtn = document.querySelector("button.chat-send-btn, button[title=\"Send\"]");
+          if (sendBtn) sendBtn.click();
+        });
+        return;
       }
 
       // 2) 复制到剪贴板
@@ -546,7 +609,6 @@
         navigator.clipboard.writeText(text).then(function () {
           console.log("[myclaw-send] 📋 已复制到剪贴板:", text.substring(0, 50) + (text.length > 50 ? "..." : ""));
         }).catch(function () {
-          // fallback: 老方法
           fallbackCopy(text);
         });
       } catch (ex) {
@@ -554,9 +616,8 @@
       }
 
       // 3) 让原生 click 继续走（发送消息）
-      //    不 preventDefault，不 stopPropagation
 
-      // 4) 延迟清空 textarea（等原生 handler 读完值后再清）
+      // 4) 延迟清空 textarea
       setTimeout(function () {
         setTextareaValue("");
         committedText = "";
@@ -853,7 +914,7 @@
       { label: "\uD83D\uDCAC \u6DFB\u52A0\u5BF9\u8BDD", desc: "\u6253\u5F00\u5DF2\u6709\u4F19\u4F34\u7684\u5BF9\u8BDD\u7A97\u53E3", hasInput: true, inputTitle: "\u6DFB\u52A0\u5BF9\u8BDD", placeholder: "\u8F93\u5165\u4F19\u4F34\u540D\u79F0\uFF0C\u5982 kakaxi", hint: "\u8F93\u5165\u4F60\u7684\u4F19\u4F34\u7684\u540D\u79F0\uFF08\u82F1\u6587\u5B57\u6BCD\u3001\u6570\u5B57\u3001\u8FDE\u5B57\u7B26\uFF09\uFF0C\u70B9\u51FB\u540E\u4F1A\u6253\u5F00\u5BF9\u8BDD\u7A97\u53E3", cmd: "mc tui {name}", color: "#10b981" },
       { label: "\uD83D\uDE80 \u5347\u7EA7", desc: "\u5347\u7EA7 myclaw \u5230\u6700\u65B0\u7248\u672C", hasInput: false, cmd: "mc up", color: "#8b5cf6" },
       { label: "\uD83D\uDD04 \u91CD\u542F", desc: "\u91CD\u542F\u670D\u52A1\uFF0C\u4FEE\u590D\u5927\u591A\u6570\u95EE\u9898", hasInput: false, cmd: "mc restart", color: "#ef4444" },
-      { label: "\uD83E\uDD1D \u65B0\u4F19\u4F34", desc: "\u521B\u5EFA\u4E00\u4E2A\u65B0\u7684 AI \u4F19\u4F34", hasInput: true, inputTitle: "\u65B0\u5EFA\u4F19\u4F34", placeholder: "\u8F93\u5165\u65B0\u4F19\u4F34\u540D\u79F0\uFF0C\u5982 my-cat", hint: "\u7ED9\u4F60\u7684\u65B0 AI \u4F19\u4F34\u8D77\u4E2A\u540D\u5B57\uFF08\u82F1\u6587\u5B57\u6BCD\u3001\u6570\u5B57\u3001\u8FDE\u5B57\u7B26\uFF09\uFF0C\u70B9\u51FB\u540E\u4F1A\u81EA\u52A8\u521B\u5EFA", cmd: "mc tui {name}", color: "#3b82f6" },
+      { label: "\uD83E\uDD1D \u65B0\u4F19\u4F34", desc: "\u521B\u5EFA\u4E00\u4E2A\u65B0\u7684 AI \u4F19\u4F34", hasInput: true, inputTitle: "\u65B0\u5EFA\u4F19\u4F34", placeholder: "\u8F93\u5165\u65B0\u4F19\u4F34\u540D\u79F0\uFF0C\u5982 my-cat", hint: "\u7ED9\u4F60\u7684\u65B0 AI \u4F19\u4F34\u8D77\u4E2A\u540D\u5B57\uFF08\u82F1\u6587\u5B57\u6BCD\u3001\u6570\u5B57\u3001\u8FDE\u5B57\u7B26\uFF09\uFF0C\u70B9\u51FB\u540E\u4F1A\u81EA\u52A8\u521B\u5EFA", cmd: "mc new {name}", color: "#3b82f6" },
     ];
 
     btns.forEach(function (item) {
@@ -908,6 +969,175 @@
       form.appendChild(row);
     });
 
+    // ── 删除伙伴按钮 ──
+    var delRow = document.createElement("div");
+    delRow.style.cssText = [
+      "padding:10px 14px",
+      "background:#252536",
+      "border-radius:6px",
+      "cursor:pointer",
+      "transition:background 0.15s",
+      "display:flex",
+      "align-items:center",
+      "gap:10px",
+    ].join(";");
+    delRow.onmouseenter = function () { delRow.style.background = "#2f2f4a"; };
+    delRow.onmouseleave = function () { delRow.style.background = "#252536"; };
+
+    var delBar = document.createElement("div");
+    delBar.style.cssText = "width:3px;height:28px;border-radius:2px;background:#ef4444;flex-shrink:0;";
+    delRow.appendChild(delBar);
+
+    var delInfo = document.createElement("div");
+    delInfo.style.cssText = "flex:1;display:flex;flex-direction:column;gap:2px;";
+    var delName = document.createElement("div");
+    delName.textContent = "\uD83D\uDDD1 \u5220\u9664\u4F19\u4F34";
+    delName.style.cssText = "font-size:13px;font-weight:bold;color:#ef4444;";
+    delInfo.appendChild(delName);
+    var delDesc = document.createElement("div");
+    delDesc.textContent = "\u5220\u9664\u4E00\u4E2A AI \u4F19\u4F34\uFF0C\u6B64\u64CD\u4F5C\u65E0\u6CD5\u6062\u590D";
+    delDesc.style.cssText = "font-size:11px;color:#888;";
+    delInfo.appendChild(delDesc);
+    delRow.appendChild(delInfo);
+
+    var delArrow = document.createElement("div");
+    delArrow.textContent = "\u25B6";
+    delArrow.style.cssText = "color:#555;font-size:10px;";
+    delRow.appendChild(delArrow);
+
+    delRow.onclick = function () {
+      showDeleteConfirm();
+    };
+
+    form.appendChild(delRow);
+
+    // 删除伙伴 - 双重确认弹框
+    function showDeleteConfirm() {
+      var mask = document.createElement("div");
+      mask.style.cssText = [
+        "position:fixed",
+        "top:0;left:0;width:100vw;height:100vh",
+        "background:rgba(0,0,0,0.3)",
+        "z-index:999999",
+        "display:flex",
+        "align-items:center",
+        "justify-content:center",
+        "animation:myclaw-fade-in 0.15s ease",
+      ].join(";");
+
+      var box = document.createElement("div");
+      box.style.cssText = [
+        "width:360px",
+        "background:#1e1e2e",
+        "border-radius:8px",
+        "overflow:hidden",
+        "box-shadow:0 8px 32px rgba(0,0,0,0.5)",
+      ].join(";");
+
+      // 标题
+      var h = document.createElement("div");
+      h.style.cssText = "padding:10px 14px;background:#ef4444;color:#fff;font-size:13px;display:flex;justify-content:space-between;align-items:center;";
+      h.innerHTML = '<span>\uD83D\uDDD1 \u5220\u9664\u4F19\u4F34</span>';
+      var x = document.createElement("span");
+      x.textContent = "\u2715";
+      x.style.cssText = "cursor:pointer;padding:2px 6px;border-radius:3px;";
+      x.onclick = function () { mask.remove(); };
+      h.appendChild(x);
+      box.appendChild(h);
+
+      // body
+      var body = document.createElement("div");
+      body.style.cssText = "padding:16px;display:flex;flex-direction:column;gap:12px;";
+
+      var hint1 = document.createElement("div");
+      hint1.textContent = "\u8BF7\u8F93\u5165\u8981\u5220\u9664\u7684\u4F19\u4F34 ID\uFF1A";
+      hint1.style.cssText = "font-size:12px;color:#888;";
+      body.appendChild(hint1);
+
+      var input1 = document.createElement("input");
+      input1.type = "text";
+      input1.placeholder = "\u4F19\u4F34 ID";
+      input1.style.cssText = "padding:8px 10px;background:#252536;border:1px solid #3d3d5c;border-radius:4px;color:#cdd6f4;font-size:13px;font-family:monospace;outline:none;";
+      body.appendChild(input1);
+
+      var warn = document.createElement("div");
+      warn.textContent = "\u26A0 \u6B64\u64CD\u4F5C\u65E0\u6CD5\u6062\u590D\uFF0C\u786E\u8BA4\u540E\u5C06\u6C38\u4E45\u5220\u9664\uFF01";
+      warn.style.cssText = "font-size:11px;color:#ef4444;padding:8px;background:rgba(239,68,68,0.1);border-radius:4px;";
+      body.appendChild(warn);
+
+      var confirmHint = document.createElement("div");
+      confirmHint.textContent = '\u8BF7\u8F93\u5165 "YES" \u786E\u8BA4\u5220\u9664\uFF1A';
+      confirmHint.style.cssText = "font-size:12px;color:#888;";
+      confirmHint.style.display = "none";
+      body.appendChild(confirmHint);
+
+      var input2 = document.createElement("input");
+      input2.type = "text";
+      input2.placeholder = "YES";
+      input2.style.cssText = "padding:8px 10px;background:#252536;border:1px solid #3d3d5c;border-radius:4px;color:#cdd6f4;font-size:13px;font-family:monospace;outline:none;";
+      input2.style.display = "none";
+      body.appendChild(input2);
+
+      var submitBtn = document.createElement("button");
+      submitBtn.textContent = "\u7EE7\u7EED";
+      submitBtn.style.cssText = "padding:8px 16px;background:#ef4444;border:none;border-radius:4px;color:#fff;font-size:12px;font-family:monospace;cursor:pointer;";
+      body.appendChild(submitBtn);
+
+      var cancelBtn = document.createElement("button");
+      cancelBtn.textContent = "\u53D6\u6D88";
+      cancelBtn.style.cssText = "padding:8px 16px;background:#3d3d5c;border:none;border-radius:4px;color:#cdd6f4;font-size:12px;font-family:monospace;cursor:pointer;";
+      body.appendChild(cancelBtn);
+
+      box.appendChild(body);
+      mask.appendChild(box);
+      document.body.appendChild(mask);
+
+      input1.focus();
+
+      var agentId = "";
+      submitBtn.onclick = function () {
+        if (!agentId) {
+          // 第一步：输入 agent ID
+          agentId = input1.value.trim();
+          if (!agentId) {
+            input1.style.borderColor = "#ef4444";
+            input1.focus();
+            return;
+          }
+          // 显示第二步确认
+          input1.style.display = "none";
+          hint1.style.display = "none";
+          confirmHint.style.display = "block";
+          input2.style.display = "block";
+          submitBtn.textContent = "\u786E\u8BA4\u5220\u9664";
+          input2.value = "";
+          input2.focus();
+        } else {
+          // 第二步：确认删除
+          var confirmVal = input2.value.trim();
+          if (confirmVal.toUpperCase() !== "YES") {
+            input2.style.borderColor = "#ef4444";
+            input2.focus();
+            return;
+          }
+          submitBtn.disabled = true;
+          submitBtn.textContent = "\u6267\u884C\u4E2D...";
+          runCommand("openclaw agents delete " + agentId + " --force");
+          setTimeout(function () {
+            mask.remove();
+          }, 1000);
+        }
+      };
+
+      cancelBtn.onclick = function () {
+        mask.remove();
+      };
+
+      input2.onkeydown = function (e) {
+        if (e.key === "Enter") submitBtn.click();
+      };
+    }
+
     box.appendChild(header);
     box.appendChild(form);
     overlay.appendChild(box);
@@ -939,6 +1169,9 @@
     // 初始化 VoiceInput SDK
     initVoice();
 
+    // 拦截语音态 Enter 键
+    hookVoiceEnter();
+
     // 持续监听 DOM 变化，确保按钮始终在
     new MutationObserver(function () {
       if (!document.querySelector("#myclaw-voice-btn")) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aiyiran/myclaw",
-  "version": "1.0.243",
+  "version": "1.0.245",
   "description": "",
   "main": "index.js",
   "bin": {

package/skills/yiran-skill-media/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: yiran-skill-media
-description: 统一多媒体生成技能。支持图片和音乐生成，按资源类型自动路由到最优 provider，支持主备切换。资源生成规范：所有生成的资源必须存放在当前工作目录下，调用时通过 --output-dir 传入当前工作目录的绝对路径，通过 --name 传入资源的中文名称。
+description: 统一多媒体生成技能。支持图片、音乐、文生视频和图生视频，按资源类型自动路由到最优 provider，支持多级降级。资源生成规范：所有生成的资源必须存放在当前工作目录下，调用时通过 --output-dir 传入当前工作目录的绝对路径，通过 --name 传入资源的中文名称。
 ---
 
 # 统一多媒体生成
@@ -12,10 +12,6 @@ description: 统一多媒体生成技能。支持图片和音乐生成，按资
 
 **所有生成的资源文件必须存放在当前工作目录下。** 不允许省略，不允许猜测路径。
 
-例如：
-- 你当前在 `/root/.openclaw/workspace` → 传入 `--output-dir /root/.openclaw/workspace`
-- 生成一张日落图片 → 传入 `--name 日落风景`
-
 ## 一键脚本
 
 ```bash
@@ -24,6 +20,12 @@ description: 统一多媒体生成技能。支持图片和音乐生成，按资
 
 # 音乐生成
 ./music.sh --output-dir "$(pwd)" --name 开场音乐 "epic opening" [--instrumental]
+
+# 文生视频 (Text-to-Video)
+./video.sh --output-dir "$(pwd)" --name 日落延时 "a sunset timelapse" [--duration 6] [--resolution 768P] [--aspect-ratio 16:9]
+
+# 图生视频 (Image-to-Video)
+./i2v.sh --output-dir "$(pwd)" --name 猫咪跑步 --first-frame-image "https://example.com/cat.jpg" "cat running toward camera"
 ```
 
 ## 参数说明
@@ -32,40 +34,73 @@ description: 统一多媒体生成技能。支持图片和音乐生成，按资
 
 | 参数 | 必填 | 说明 |
 |------|------|------|
-| `--output-dir` | 是 | 输出目录的绝对路径，传入当前工作目录 |
+| `--output-dir` | 是 | 输出目录的绝对路径 |
 | `--name` | 是 | 资源中文名称（如：日落风景、产品封面） |
 | `prompt` | 是 | 图片描述 |
-| `--aspect-ratio` | 否 | 比例，默认 1:1。可选：16:9, 9:16, 4:3 等 |
+| `--aspect-ratio` | 否 | 比例，默认 16:9。可选：1:1, 16:9, 9:16, 4:3, 3:4 等 |
 
 ### music.sh
 
 | 参数 | 必填 | 说明 |
 |------|------|------|
-| `--output-dir` | 是 | 输出目录的绝对路径，传入当前工作目录 |
-| `--name` | 是 | 资源中文名称（如：开场音乐、背景配乐） |
+| `--output-dir` | 是 | 输出目录的绝对路径 |
+| `--name` | 是 | 资源中文名称 |
 | `prompt` | 是 | 音乐风格/情绪描述 |
 | `--lyrics` | 否 | 歌词文本 |
 | `--instrumental` | 否 | 纯音乐模式 |
 
+### video.sh — 文生视频
+
+| 参数 | 必填 | 说明 |
+|------|------|------|
+| `--output-dir` | 是 | 输出目录的绝对路径 |
+| `--name` | 是 | 资源中文名称 |
+| `prompt` | 是 | 视频内容描述（支持运镜指令如 `[推进]`、`[左摇]`） |
+| `--duration` | 否 | 视频时长（秒），默认 6，可选 6 或 10 |
+| `--resolution` | 否 | MiniMax 分辨率：768P/1080P（默认 768P） |
+| `--aspect-ratio` | 否 | 即梦降级时使用：16:9/9:16/4:3/1:1（默认 16:9） |
+
+### i2v.sh — 图生视频
+
+| 参数 | 必填 | 说明 |
+|------|------|------|
+| `--output-dir` | 是 | 输出目录的绝对路径 |
+| `--name` | 是 | 资源中文名称 |
+| `--first-frame-image` | 是 | 首帧图片的公网 URL（JPG/PNG/WebP，<20MB） |
+| `prompt` | 否 | 基于首帧图像的动作/变化描述 |
+| `--duration` | 否 | 视频时长，默认 6 |
+| `--resolution` | 否 | 分辨率，默认 768P |
+
+**注意**：视频生成为异步任务，耗时较长（通常 1-5 分钟），脚本会自动轮询等待完成。
+
 ## 架构
 
 ```
-image.sh / music.sh    →  智能体调用的薄壳入口
-generate.py            →  统一路由调度（主备切换）
-config.json            →  provider 配置中心（key、模型、地址）
-providers/             →  各 provider 适配器
-  minimax_image.py     →  MiniMax 图片（主）
-  vapi_image.py        →  VAPI 图片（备）
-  minimax_music.py     →  MiniMax 音乐（主）
+image.sh              →  图片生成入口
+music.sh              →  音乐生成入口
+video.sh              →  文生视频入口 (Text-to-Video)
+i2v.sh                →  图生视频入口 (Image-to-Video)
+generate.py           →  统一路由调度（优先级数组，依次尝试）
+config.json           →  provider 配置中心（优先级列表）
+providers/
+  vapi_image.py       →  VAPI 图片
+  minimax_image.py    →  MiniMax 图片
+  jimeng_image.py     →  即梦 图片 4.0（异步）
+  minimax_music.py    →  MiniMax 音乐
+  minimax_video.py    →  MiniMax 视频（文生+图生）
+  jimeng_video.py     →  即梦 视频 3.0（异步）
 ```
 
-## Provider 配置
+## Provider 配置（优先级列表）
 
-编辑 `config.json` 可切换主备 provider、更换模型或 API Key。
+`config.json` 中每个资源类型是一个数组，按优先级从高到低排列。  
+第一个失败自动尝试下一个，直到成功或全部失败。
 
 当前配置：
-- **图片**：MiniMax (image-01) → fallback VAPI (nano-banana-pro)
-- **音乐**：MiniMax (music-2.6)
+- **图片**：① VAPI (nano-banana-2) → ② MiniMax (image-01) → ③ 即梦 (jimeng_t2i_v40)
+- **音乐**：① MiniMax (music-2.6)
+- **文生视频**：① MiniMax (MiniMax-Hailuo-2.3) → ② 即梦 (jimeng_t2v_v30)
+- **图生视频**：① MiniMax (MiniMax-Hailuo-2.3-Fast)
 
 ## 详细 API 参考
 

package/skills/yiran-skill-media/config.json

@@ -1,26 +1,53 @@
 {
   "output_dir": "media",
-  "image": {
-    "primary": {
+  "image": [
+    {
+      "provider": "jimeng_image",
+      "model": "jimeng_t2i_v40",
+      "access_key": "AKLTYjZkY2FiZmZkYWU5NDkxNmEwZjNlYTRjNmRlZmYwNDI",
+      "secret_key": "TjJGbU5HVTBZek14TnpFeE5HWTVOVGhsTURRNE9XRXhNR1JoTm1FeVlqaw=="
+    },
+    {
       "provider": "vapi_image",
-      "model": "nano-banana-pro",
+      "model": "nano-banana-2",
       "base_url": "https://api.v3.cm/v1",
       "api_key": "sk-PXPUzqllWKJy2oj011Df510242264219Ba21093e3d2b2335"
     },
-    "fallback": {
+    {
       "provider": "minimax_image",
       "model": "image-01",
       "base_url": "https://api.minimaxi.com/v1",
       "api_key": "sk-cp-DC5lWd2Stt9CBFzLIT2awP4K-ZEn5AkYwjl3Cdj-mIBmgjxod518F2LaVF2L9c35Wv5-Eox0F1ctJD5vXtB9p3OmxoWLd9ge9zIUIMrCVuqBYdL_s6kb8Qs"
     }
-  },
-  "music": {
-    "primary": {
+  ],
+  "music": [
+    {
       "provider": "minimax_music",
       "model": "music-2.6",
       "base_url": "https://api.minimaxi.com/v1",
       "api_key": "sk-cp-DC5lWd2Stt9CBFzLIT2awP4K-ZEn5AkYwjl3Cdj-mIBmgjxod518F2LaVF2L9c35Wv5-Eox0F1ctJD5vXtB9p3OmxoWLd9ge9zIUIMrCVuqBYdL_s6kb8Qs"
+    }
+  ],
+  "video": [
+    {
+      "provider": "minimax_video",
+      "model": "MiniMax-Hailuo-2.3",
+      "base_url": "https://api.minimaxi.com/v1",
+      "api_key": "sk-cp-DC5lWd2Stt9CBFzLIT2awP4K-ZEn5AkYwjl3Cdj-mIBmgjxod518F2LaVF2L9c35Wv5-Eox0F1ctJD5vXtB9p3OmxoWLd9ge9zIUIMrCVuqBYdL_s6kb8Qs"
     },
-    "fallback": null
-  }
+    {
+      "provider": "jimeng_video",
+      "model": "jimeng_t2v_v30",
+      "access_key": "AKLTYjZkY2FiZmZkYWU5NDkxNmEwZjNlYTRjNmRlZmYwNDI",
+      "secret_key": "TjJGbU5HVTBZek14TnpFeE5HWTVOVGhsTURRNE9XRXhNR1JoTm1FeVlqaw=="
+    }
+  ],
+  "i2v": [
+    {
+      "provider": "minimax_video",
+      "model": "MiniMax-Hailuo-2.3-Fast",
+      "base_url": "https://api.minimaxi.com/v1",
+      "api_key": "sk-cp-DC5lWd2Stt9CBFzLIT2awP4K-ZEn5AkYwjl3Cdj-mIBmgjxod518F2LaVF2L9c35Wv5-Eox0F1ctJD5vXtB9p3OmxoWLd9ge9zIUIMrCVuqBYdL_s6kb8Qs"
+    }
+  ]
 }

package/skills/yiran-skill-media/scripts/generate.py

@@ -102,9 +102,9 @@ def append_log(log_path, entry):
 
 
 def dispatch(resource_type, prompt, **kwargs):
-    """Route to primary provider, fallback on failure.
+    """Route to providers sequentially as a priority array.
 
-    Always tries primary first, then fallback if configured.
+    Always tries the first provider, then falls back to the next on failure.
     Returns (files, used_provider_cfg).
     """
     cfg = load_config()
@@ -112,9 +112,15 @@ def dispatch(resource_type, prompt, **kwargs):
     if not resource_cfg:
         raise ValueError(f"unknown resource type: {resource_type}")
 
-    providers = [resource_cfg["primary"]]
-    if resource_cfg.get("fallback"):
-        providers.append(resource_cfg["fallback"])
+    # 支持旧版 dict(primary/fallback) 过渡到新版 list 
+    if isinstance(resource_cfg, dict):
+        providers = [resource_cfg["primary"]]
+        if resource_cfg.get("fallback"):
+            providers.append(resource_cfg["fallback"])
+    elif isinstance(resource_cfg, list):
+        providers = resource_cfg
+    else:
+        raise ValueError(f"invalid config structure for {resource_type}")
 
     sys.path.insert(0, SCRIPT_DIR)
     from providers import get_adapter
@@ -136,16 +142,23 @@ def dispatch(resource_type, prompt, **kwargs):
 
 def main():
     parser = argparse.ArgumentParser(description="Unified media generation dispatcher")
-    parser.add_argument("type", choices=["image", "music"], help="Resource type")
-    parser.add_argument("prompt", help="Generation prompt")
-    parser.add_argument("--aspect-ratio", default="1:1", help="Image aspect ratio (image only)")
+    parser.add_argument("type", choices=["image", "music", "video", "i2v"], help="Resource type")
+    parser.add_argument("prompt", nargs="?", default="", help="Generation prompt")
+    parser.add_argument("--aspect-ratio", default="16:9", help="Aspect ratio (image: 1:1; video: 16:9/9:16/4:3/1:1)")
     parser.add_argument("--lyrics", default=None, help="Lyrics text (music only)")
     parser.add_argument("--instrumental", action="store_true", help="Instrumental mode (music only)")
+    parser.add_argument("--duration", type=int, default=None, help="Video duration in seconds (video/i2v, default 6)")
+    parser.add_argument("--resolution", default=None, help="Video resolution: 720P/768P/1080P (video/i2v)")
+    parser.add_argument("--first-frame-image", default=None, help="First frame image URL (i2v only, required)")
     parser.add_argument("--output", default=None, help="Output file path")
     parser.add_argument("--output-dir", required=True, help="Absolute path to output directory (required)")
     parser.add_argument("--name", required=True, help="Resource name in Chinese (required, e.g. 日落风景)")
     args = parser.parse_args()
 
+    # i2v 必须有 --first-frame-image
+    if args.type == "i2v" and not args.first_frame_image:
+        parser.error("i2v (图生视频) 模式必须提供 --first-frame-image 参数")
+
     out_dir = ensure_output_dir(args.output_dir)
 
     # Prepare kwargs
@@ -155,6 +168,28 @@ def main():
             "aspect_ratio": args.aspect_ratio,
         }
         ext = "png"
+    elif args.type == "video":
+        # 文生视频 — 纯文本，无图片
+        kwargs = {
+            "out_dir": out_dir,
+            "aspect_ratio": args.aspect_ratio,
+        }
+        if args.duration:
+            kwargs["duration"] = args.duration
+        if args.resolution:
+            kwargs["resolution"] = args.resolution
+        ext = "mp4"
+    elif args.type == "i2v":
+        # 图生视频 — 必须有首帧图片
+        kwargs = {
+            "out_dir": out_dir,
+            "first_frame_image": args.first_frame_image,
+        }
+        if args.duration:
+            kwargs["duration"] = args.duration
+        if args.resolution:
+            kwargs["resolution"] = args.resolution
+        ext = "mp4"
     else:
         kwargs = {
             "out_dir": out_dir,

package/skills/yiran-skill-media/scripts/i2v.sh

@@ -0,0 +1,47 @@
+#!/bin/bash
+# 图生视频入口 (Image-to-Video)
+# 模型: MiniMax-Hailuo-2.3-Fast
+# 用法: ./i2v.sh --output-dir /abs/path --name 中文名 --first-frame-image URL ["描述"]
+set -euo pipefail
+
+OUTPUT_DIR=""
+NAME=""
+PROMPT=""
+DURATION=""
+RESOLUTION=""
+FIRST_FRAME=""
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --output-dir) OUTPUT_DIR="$2"; shift 2 ;;
+        --name) NAME="$2"; shift 2 ;;
+        --duration) DURATION="$2"; shift 2 ;;
+        --resolution) RESOLUTION="$2"; shift 2 ;;
+        --first-frame-image) FIRST_FRAME="$2"; shift 2 ;;
+        *)
+            if [ -z "$PROMPT" ]; then
+                PROMPT="$1"
+            fi
+            shift
+            ;;
+    esac
+done
+
+if [ -z "$OUTPUT_DIR" ] || [ -z "$NAME" ] || [ -z "$FIRST_FRAME" ]; then
+    echo "用法: ./i2v.sh --output-dir <绝对路径> --name <中文名> --first-frame-image <图片URL> [\"动作描述\"]"
+    echo "示例: ./i2v.sh --output-dir \$(pwd) --name 猫咪跑步 --first-frame-image https://example.com/cat.jpg \"cat running toward camera\""
+    exit 1
+fi
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+
+ARGS=()
+ARGS+=("i2v")
+[ -n "$PROMPT" ] && ARGS+=("$PROMPT")
+ARGS+=(--output-dir "$OUTPUT_DIR")
+ARGS+=(--name "$NAME")
+ARGS+=(--first-frame-image "$FIRST_FRAME")
+[ -n "$DURATION" ] && ARGS+=(--duration "$DURATION")
+[ -n "$RESOLUTION" ] && ARGS+=(--resolution "$RESOLUTION")
+
+python3 "$SCRIPT_DIR/generate.py" "${ARGS[@]}"

package/skills/yiran-skill-media/scripts/providers/__init__.py

@@ -1,11 +1,17 @@
 from .vapi_image import VAPIImageAdapter
 from .minimax_image import MiniMaxImageAdapter
 from .minimax_music import MiniMaxMusicAdapter
+from .minimax_video import MiniMaxVideoAdapter
+from .jimeng_video import JimengVideoAdapter
+from .jimeng_image import JimengImageAdapter
 
 ADAPTERS = {
     "vapi_image": VAPIImageAdapter(),
     "minimax_image": MiniMaxImageAdapter(),
     "minimax_music": MiniMaxMusicAdapter(),
+    "minimax_video": MiniMaxVideoAdapter(),
+    "jimeng_video": JimengVideoAdapter(),
+    "jimeng_image": JimengImageAdapter(),
 }
 
 def get_adapter(name: str):
@@ -13,3 +19,5 @@ def get_adapter(name: str):
     if not adapter:
         raise ValueError(f"unknown provider: {name}")
     return adapter
+
+

package/skills/yiran-skill-media/scripts/providers/jimeng_image.py

@@ -0,0 +1,158 @@
+"""即梦(Jimeng) image adapter — Volcengine SDK, async workflow.
+Uses CVSync2AsyncSubmitTask → CVSync2AsyncGetResult polling.
+Supports text-to-image via jimeng_t2i_v40."""
+import json
+import os
+import sys
+import time
+
+import requests
+from volcengine.visual.VisualService import VisualService
+
+
+class JimengImageAdapter:
+    POLL_INTERVAL = 5    # 图片生成比视频快，5 秒轮询
+    MAX_WAIT = 120       # 最长等 2 分钟
+
+    # 比例 → 推荐的 2K 分辨率（官方文档中的推荐值）
+    RATIO_MAP = {
+        "1:1":  (2048, 2048),
+        "4:3":  (2304, 1728),
+        "3:4":  (1728, 2304),
+        "3:2":  (2496, 1664),
+        "2:3":  (1664, 2496),
+        "16:9": (2560, 1440),
+        "9:16": (1440, 2560),
+        "21:9": (3024, 1296),
+        "9:21": (1296, 3024),
+    }
+
+    def generate(self, prompt, config, **kwargs):
+        access_key = config["access_key"]
+        secret_key = config["secret_key"]
+        req_key = config.get("model", "jimeng_t2i_v40")
+        out_dir = kwargs["out_dir"]
+        aspect_ratio = kwargs.get("aspect_ratio", "1:1")
+
+        vs = VisualService()
+        vs.set_ak(access_key)
+        vs.set_sk(secret_key)
+
+        # ═══ Step 1: 提交任务 ═══
+        body = {
+            "req_key": req_key,
+            "prompt": prompt,
+            "force_single": True,  # 强制输出单图，确保智能体调用行为可预测
+        }
+
+        # 宽高映射
+        if aspect_ratio in self.RATIO_MAP:
+            w, h = self.RATIO_MAP[aspect_ratio]
+            body["width"] = w
+            body["height"] = h
+        else:
+            # 不传 width/height，让模型根据 prompt 自动判断
+            body["size"] = 2048 * 2048  # 2K 默认面积
+
+        print(f"[jimeng_image] 提交文生图任务 req_key={req_key} ratio={aspect_ratio}...", file=sys.stderr)
+        try:
+            data = vs.cv_sync2async_submit_task(body)
+        except Exception as e:
+            if hasattr(e, 'args') and len(e.args) > 0 and isinstance(e.args[0], bytes):
+                try:
+                    err_json = json.loads(e.args[0].decode('utf-8'))
+                    code = err_json.get("code", 0)
+                    msg = err_json.get("message", "")
+                    raise RuntimeError(f"即梦 API 错误: {msg} (code={code})")
+                except RuntimeError:
+                    raise
+                except Exception:
+                    pass
+            raise RuntimeError(f"即梦提交失败: {str(e)}")
+
+        if data.get("code") != 10000:
+            raise RuntimeError(f"即梦 API 错误: {data.get('message')} (code={data.get('code')})")
+
+        task_id = data["data"]["task_id"]
+        print(f"[jimeng_image] task_id={task_id}, 开始轮询...", file=sys.stderr)
+
+        # ═══ Step 2: 轮询状态 ═══
+        image_urls = None
+        binary_data = None
+        elapsed = 0
+        while elapsed < self.MAX_WAIT:
+            time.sleep(self.POLL_INTERVAL)
+            elapsed += self.POLL_INTERVAL
+
+            query_body = {
+                "req_key": req_key,
+                "task_id": task_id,
+                "req_json": json.dumps({"return_url": True}),
+            }
+            try:
+                qdata = vs.cv_sync2async_get_result(query_body)
+            except Exception as e:
+                if hasattr(e, 'args') and len(e.args) > 0 and isinstance(e.args[0], bytes):
+                    try:
+                        err_json = json.loads(e.args[0].decode('utf-8'))
+                        code = err_json.get("code", 0)
+                        if code in (50429, 50430, 50500, 50501):
+                            print(f"[jimeng_image] 轮询 {elapsed}s: 可重试错误 code={code}", file=sys.stderr)
+                            continue
+                        raise RuntimeError(f"即梦查询失败: {err_json.get('message')} (code={code})")
+                    except RuntimeError:
+                        raise
+                    except Exception:
+                        pass
+                raise RuntimeError(f"即梦查询异常: {str(e)}")
+
+            if qdata.get("code") != 10000:
+                print(f"[jimeng_image] 轮询 {elapsed}s: code={qdata.get('code')} msg={qdata.get('message')}", file=sys.stderr)
+                if qdata.get("code") in (50429, 50430, 50500, 50501):
+                    continue
+                raise RuntimeError(f"即梦状态获取失败: {qdata.get('message')} (code={qdata.get('code')})")
+
+            status = qdata.get("data", {}).get("status", "")
+            print(f"[jimeng_image] 轮询 {elapsed}s: status={status}", file=sys.stderr)
+
+            if status == "done":
+                image_urls = qdata["data"].get("image_urls", [])
+                binary_data = qdata["data"].get("binary_data_base64", [])
+                break
+            elif status in ("in_queue", "generating"):
+                continue
+            elif status in ("not_found", "expired"):
+                raise RuntimeError(f"即梦任务异常: status={status}")
+
+        if not image_urls and not binary_data:
+            raise RuntimeError(f"即梦图片生成超时 ({self.MAX_WAIT}s), task_id={task_id}")
+
+        print(f"[jimeng_image] 生成完成, 下载中...", file=sys.stderr)
+
+        # ═══ Step 3: 下载图片 ═══
+        import base64
+        saved = []
+
+        if image_urls:
+            for i, url in enumerate(image_urls):
+                fname = kwargs.get("output_path") or os.path.join(out_dir, f"image_{i}.png")
+                r = requests.get(url, timeout=60)
+                r.raise_for_status()
+                with open(fname, "wb") as f:
+                    f.write(r.content)
+                saved.append(fname)
+                print(f"[jimeng_image] 已保存: {fname}", file=sys.stderr)
+        elif binary_data:
+            for i, b64 in enumerate(binary_data):
+                if not b64:
+                    continue
+                fname = kwargs.get("output_path") or os.path.join(out_dir, f"image_{i}.png")
+                with open(fname, "wb") as f:
+                    f.write(base64.b64decode(b64))
+                saved.append(fname)
+                print(f"[jimeng_image] 已保存 (base64): {fname}", file=sys.stderr)
+
+        if not saved:
+            raise RuntimeError("即梦未返回任何图片数据")
+
+        return saved

package/skills/yiran-skill-media/scripts/providers/jimeng_video.py

@@ -0,0 +1,115 @@
+"""即梦(Jimeng) video adapter — Volcengine SDK Implementation.
+Supports text-to-video via CVSync2AsyncSubmitTask."""
+import json
+import os
+import sys
+import time
+
+import requests
+from volcengine.visual.VisualService import VisualService
+
+
+class JimengVideoAdapter:
+    POLL_INTERVAL = 10
+    MAX_WAIT = 600
+
+    def generate(self, prompt, config, **kwargs):
+        access_key = config["access_key"]
+        secret_key = config["secret_key"]
+        req_key = config.get("model", "jimeng_t2v_v30")
+        out_dir = kwargs["out_dir"]
+
+        # duration → frames: 5s=121, 10s=241 (assuming default 5s max logic for now, standard expects 121 for 5s)
+        duration = int(kwargs.get("duration", 5))
+        frames = 24 * duration + 1
+
+        aspect_ratio = kwargs.get("aspect_ratio", "16:9")
+
+        # ═══ Step 1: 提交任务 ═══
+        vs = VisualService()
+        vs.set_ak(access_key)
+        vs.set_sk(secret_key)
+
+        body = {
+            "req_key": req_key,
+            "prompt": prompt,
+            "seed": -1,
+            "frames": frames,
+            "aspect_ratio": aspect_ratio,
+        }
+
+        print(f"[jimeng_video] 提交文生视频任务 req_key={req_key}...", file=sys.stderr)
+        try:
+            data = vs.cv_sync2async_submit_task(body)
+        except Exception as e:
+            # Handle SDK exceptions which wrap the raw response
+            if hasattr(e, 'args') and len(e.args) > 0 and isinstance(e.args[0], bytes):
+                try:
+                    err_json = json.loads(e.args[0].decode('utf-8'))
+                    raise RuntimeError(f"即梦 API 错误: {err_json}")
+                except Exception:
+                    pass
+            raise RuntimeError(f"即梦提交失败: {str(e)}")
+
+        if data.get("code") != 10000:
+            raise RuntimeError(f"即梦 API 错误: {data.get('message')} (code={data.get('code')})")
+
+        task_id = data["data"]["task_id"]
+        print(f"[jimeng_video] task_id={task_id}, 开始轮询...", file=sys.stderr)
+
+        # ═══ Step 2: 轮询状态 ═══
+        video_url = None
+        elapsed = 0
+        while elapsed < self.MAX_WAIT:
+            time.sleep(self.POLL_INTERVAL)
+            elapsed += self.POLL_INTERVAL
+
+            query_body = {
+                "req_key": req_key,
+                "task_id": task_id,
+            }
+            try:
+                qdata = vs.cv_sync2async_get_result(query_body)
+            except Exception as e:
+                # Same exception handling for polling
+                if hasattr(e, 'args') and len(e.args) > 0 and isinstance(e.args[0], bytes):
+                    try:
+                        err_json = json.loads(e.args[0].decode('utf-8'))
+                        if err_json.get("code") in (50500, 50501):
+                            continue # Server error, can retry
+                        raise RuntimeError(f"即梦查询失败: {err_json}")
+                    except Exception:
+                        pass
+                raise RuntimeError(f"即梦查询异常: {str(e)}")
+
+            if qdata.get("code") != 10000:
+                print(f"[jimeng_video] 轮询 {elapsed}s: code={qdata.get('code')} msg={qdata.get('message')}", file=sys.stderr)
+                if qdata.get("code") in (50500, 50501):
+                    continue  # 可重试的内部错误
+                raise RuntimeError(f"即梦状态获取失败: {qdata.get('message')} (code={qdata.get('code')})")
+
+            status = qdata.get("data", {}).get("status", "")
+            print(f"[jimeng_video] 轮询 {elapsed}s: status={status}", file=sys.stderr)
+
+            if status == "done":
+                video_url = qdata["data"].get("video_url")
+                break
+            elif status in ("in_queue", "generating"):
+                continue
+            elif status in ("not_found", "expired"):
+                raise RuntimeError(f"即梦任务异常停机: status={status}")
+
+        if not video_url:
+            raise RuntimeError(f"即梦视频生成超时 ({self.MAX_WAIT}s), task_id={task_id}")
+
+        print(f"[jimeng_video] 生成完成, 下载中...", file=sys.stderr)
+
+        # ═══ Step 3: 下载视频 ═══
+        fname = kwargs.get("output_path") or os.path.join(out_dir, "video.mp4")
+        r = requests.get(video_url, timeout=300)
+        r.raise_for_status()
+        with open(fname, "wb") as f:
+            f.write(r.content)
+
+        print(f"[jimeng_video] 已保存: {fname}", file=sys.stderr)
+        return [fname]

package/skills/yiran-skill-media/scripts/providers/minimax_video.py

@@ -0,0 +1,115 @@
+"""MiniMax video adapter — POST /video_generation, poll status, download via file_id.
+Supports both Text-to-Video and Image-to-Video modes."""
+import json
+import os
+import sys
+import time
+
+import requests
+
+
+class MiniMaxVideoAdapter:
+    # 轮询间隔和超时
+    POLL_INTERVAL = 10   # 每 10 秒查一次
+    MAX_WAIT = 600       # 最长等 10 分钟
+
+    def generate(self, prompt, config, **kwargs):
+        base_url = config["base_url"].rstrip("/")
+        api_key = config["api_key"]
+        model = config["model"]
+        out_dir = kwargs["out_dir"]
+
+        headers = {
+            "Authorization": f"Bearer {api_key}",
+            "Content-Type": "application/json",
+        }
+
+        # ═══ Step 1: 提交生成任务 ═══
+        payload = {
+            "model": model,
+        }
+        # prompt（图生视频时可选，文生视频时必填）
+        if prompt:
+            payload["prompt"] = prompt
+        # 图生视频模式：传入首帧图片
+        if kwargs.get("first_frame_image"):
+            payload["first_frame_image"] = kwargs["first_frame_image"]
+        # 可选参数
+        if kwargs.get("duration"):
+            payload["duration"] = int(kwargs["duration"])
+        if kwargs.get("resolution"):
+            payload["resolution"] = kwargs["resolution"]
+        if kwargs.get("prompt_optimizer") is not None:
+            payload["prompt_optimizer"] = kwargs["prompt_optimizer"]
+
+        mode = "图生视频" if "first_frame_image" in payload else "文生视频"
+        print(f"[minimax_video] 提交{mode}任务 model={model}...", file=sys.stderr)
+        resp = requests.post(
+            f"{base_url}/video_generation",
+            headers=headers,
+            json=payload,
+            timeout=60,
+        )
+        resp.raise_for_status()
+        data = resp.json()
+
+        # 检查业务错误
+        base_resp = data.get("base_resp", {})
+        if base_resp.get("status_code", 0) != 0:
+            raise RuntimeError(f"MiniMax error: {base_resp.get('status_msg')}")
+
+        task_id = data.get("task_id")
+        if not task_id:
+            raise RuntimeError(f"未返回 task_id: {data}")
+
+        print(f"[minimax_video] task_id={task_id}, 开始轮询...", file=sys.stderr)
+
+        # ═══ Step 2: 轮询任务状态 ═══
+        file_id = None
+        elapsed = 0
+        while elapsed < self.MAX_WAIT:
+            time.sleep(self.POLL_INTERVAL)
+            elapsed += self.POLL_INTERVAL
+
+            query_resp = requests.get(
+                f"{base_url}/query/video_generation",
+                headers={"Authorization": f"Bearer {api_key}"},
+                params={"task_id": task_id},
+                timeout=30,
+            )
+            query_resp.raise_for_status()
+            qdata = query_resp.json()
+            status = qdata.get("status", "")
+            print(f"[minimax_video] 轮询 {elapsed}s: status={status}", file=sys.stderr)
+
+            if status == "Success":
+                file_id = qdata.get("file_id")
+                break
+            elif status == "Fail":
+                err_msg = qdata.get("error_message", "未知错误")
+                raise RuntimeError(f"视频生成失败: {err_msg}")
+            # Processing / Queueing → 继续等
+
+        if not file_id:
+            raise RuntimeError(f"视频生成超时 ({self.MAX_WAIT}s), task_id={task_id}")
+
+        print(f"[minimax_video] 生成完成 file_id={file_id}, 下载中...", file=sys.stderr)
+
+        # ═══ Step 3: 获取下载 URL 并下载 ═══
+        file_resp = requests.get(
+            f"{base_url}/files/retrieve",
+            headers={"Authorization": f"Bearer {api_key}"},
+            params={"file_id": file_id},
+            timeout=30,
+        )
+        file_resp.raise_for_status()
+        download_url = file_resp.json()["file"]["download_url"]
+
+        fname = kwargs.get("output_path") or os.path.join(out_dir, "video.mp4")
+        r = requests.get(download_url, timeout=300)
+        r.raise_for_status()
+        with open(fname, "wb") as f:
+            f.write(r.content)
+
+        print(f"[minimax_video] 已保存: {fname}", file=sys.stderr)
+        return [fname]

package/skills/yiran-skill-media/scripts/video.sh

@@ -0,0 +1,47 @@
+#!/bin/bash
+# 文生视频入口 (Text-to-Video)
+# 主力模型: MiniMax-Hailuo-2.3 | 降级: 即梦 jimeng_t2v_v30
+# 用法: ./video.sh --output-dir /abs/path --name 中文名 "描述" [--duration 6] [--resolution 768P] [--aspect-ratio 16:9]
+set -euo pipefail
+
+OUTPUT_DIR=""
+NAME=""
+PROMPT=""
+DURATION=""
+RESOLUTION=""
+ASPECT_RATIO=""
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --output-dir) OUTPUT_DIR="$2"; shift 2 ;;
+        --name) NAME="$2"; shift 2 ;;
+        --duration) DURATION="$2"; shift 2 ;;
+        --resolution) RESOLUTION="$2"; shift 2 ;;
+        --aspect-ratio) ASPECT_RATIO="$2"; shift 2 ;;
+        *)
+            if [ -z "$PROMPT" ]; then
+                PROMPT="$1"
+            fi
+            shift
+            ;;
+    esac
+done
+
+if [ -z "$OUTPUT_DIR" ] || [ -z "$NAME" ] || [ -z "$PROMPT" ]; then
+    echo "用法: ./video.sh --output-dir <绝对路径> --name <中文名> \"视频描述\""
+    echo "示例: ./video.sh --output-dir /root/.openclaw/workspace --name 日落延时 \"a sunset timelapse\""
+    exit 1
+fi
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+
+ARGS=()
+ARGS+=("video")
+ARGS+=("$PROMPT")
+ARGS+=(--output-dir "$OUTPUT_DIR")
+ARGS+=(--name "$NAME")
+[ -n "$DURATION" ] && ARGS+=(--duration "$DURATION")
+[ -n "$RESOLUTION" ] && ARGS+=(--resolution "$RESOLUTION")
+[ -n "$ASPECT_RATIO" ] && ARGS+=(--aspect-ratio "$ASPECT_RATIO")
+
+python3 "$SCRIPT_DIR/generate.py" "${ARGS[@]}"