dsh-connect-minimaxcode 0.5.10 → 0.6.1

package/README.en.md

@@ -302,6 +302,63 @@ The card shows the parts worth acting on:
 
 This data **needs no M Plan subscription**: the endpoint is an account one, read with the OAuth bearer, unlike the limit windows, which belong to the subscription.
 
+### Image and video generation (H3)
+
+The plugin page gained a **Generate** panel, which drives MiniMax's **H3 image
+model** and **H3 video generation** directly:
+
+- Images: `MiniMax-H3-Image` (and `MiniMax-H3-Image-Highspeed`), with an
+  aspect ratio and a 1K / 2K / 4K area
+- Video: `MiniMax-H3`, with a length of 5–15 seconds, 768P / 2K and an
+  aspect ratio
+
+Both go through the endpoints the open platform's own documentation gives:
+
+```
+POST https://api.minimax.cn/v1/images/generations          # text to image, synchronous
+POST https://api.minimax.cn/v2/video_generation           # create a video task, asynchronous
+GET  https://api.minimax.cn/v2/query/video_generation/{id}  # poll the task
+```
+
+**No key has to be configured.** The generation endpoints want the open
+platform's API key (`sk-cp-…`), and that key belongs to the *account*: the
+plugin asks `token_plan_credit` with the OAuth bearer — the endpoint it
+already reads for the credit balance — and the `api_key` in the answer is
+it. Measured on 2026-11-10:
+
+| Request | Credential | Result |
+|---|---|---|
+| `/v1/images/generations` + `sk-cp-` key | 200, 8.5 s | 1184×880 JPEG, 395 KB, 6401 tokens |
+| `/v1/images/generations` + OAuth bearer | 401 `1004 login fail` | the endpoint takes only the API key |
+| `/v2/video_generation` + `sk-cp-` key | 200, task `succeeded` after 90 s | 5 s 768P clip, 162745 tokens |
+
+The generated picture is **committed to DSH's own attachment store**
+(`ctx.attachments`), so it lives in DSH's storage rather than only in this
+browser tab.
+
+#### Two differences from the desktop app, stated plainly
+
+1. **It cannot appear in the model picker like an ordinary model.** The host's
+   `ctx.llm` has exactly one registration point, `registerAdapter`, for
+   chat adapters; there is nowhere to register an image provider. The model
+   descriptor `LlmModelInfo` has no output-modality field, so the picker
+   could not tell an image model apart. pi-ai does have an image-generation
+   API (`ImagesProvider` / `generateImages`), but DSH never constructs an
+   `ImagesModels`, and one the plugin built for itself would have nothing
+   reading it. All three are host-side, and a plugin cannot change them.
+2. **Video has no carrier in DSH at all.** Across `pi-ai`, `dsh-llm`,
+   `dsh-attachment` and `dsh-llm-pi-ai` there is not one video-related
+   declaration: no video content block, no attachment media type, no
+   rendering. A video can therefore be generated, watched and downloaded in
+   the panel, but it cannot join a conversation the way a picture can.
+
+A picture *can* reach a conversation: the host publishes
+`ctx.attachments.admitPromptContent()`, so a plugin can complete "generate →
+store → durable reference" on its own. But the attachment entry point is a
+browser `File` upload and there is **no attachment library** to pick from,
+so it still has to travel as part of the next prompt. That last hop is
+done in 0.6.1 (see below).
+
 ### About token refresh
 
 The OAuth token is valid for about an hour and is refreshed automatically by the client. The plugin **only refreshes it once, with the refresh token the app already stored, when its own token has expired**, and writes the rotated new tokens back to the same file atomically, in the client's **original format** (with `generation` unchanged), so the desktop app's login state is unaffected. The plugin never starts a login, never signs out, and never calls any paid or subscription endpoint.
@@ -327,6 +384,20 @@ The OAuth token is valid for about an hour and is refreshed automatically by the
 
 ## Changelog
 
+### 0.6.1 (2026-11-11)
+
+- **A generated picture can now go straight into the conversation**: the composer tool row gained a **Generate an image** entry. Write a prompt, press generate, and the picture is sent as part of this turn — through the host's own road (`ctx.sessions.binding(id).session.prompt(content, mode, signal, requestId)`, the very call the send button makes), so the host validates it, commits it as a durable attachment and treats it as an ordinary message.
+- A host that exposes no session face gets a plain "could not send it to the conversation" instead of a silent failure, and the plugin page's panel keeps working regardless.
+- Added `tests/generate-dock.spec.ts` (4 cases) pinning: the entry is registered in the tool row, it declares the `sessions` service, the picture travels as `{type:"image", mediaType, data}` (raw base64, no data-url prefix) together with the prompt, and the message shown when there is no session to send to. All 252 cases pass.
+
+### 0.6.0 (2026-11-11)
+
+- **Added H3 image and video generation**: the plugin page gained a **Generate** panel that drives MiniMax's `MiniMax-H3-Image` (text to image) and `MiniMax-H3` (video) directly. It goes through the endpoints the open platform's documentation gives (`POST /v1/images/generations`, `POST /v2/video_generation` plus polling `/v2/query/video_generation/{id}`), and **no key has to be configured**: the generation endpoints want the account's own open-platform key, and the plugin gets it by asking `token_plan_credit` with the OAuth bearer — the endpoint it already reads for the balance.
+- **Generated pictures are committed to DSH's own attachment store** (`ctx.attachments`), so they live in DSH's storage rather than only in a browser tab.
+- The choices (aspect ratio, area, length, resolution) are published by the server, and they are the ones the upstream really serves.
+- **Measured** (2026-11-10): a picture in 8.5 s, 1184×880 JPEG / 395 KB / 6401 tokens; a 5 s 768P video whose task reached `succeeded` after 90 s, 162745 tokens. The generation endpoints take only the API key and refuse the OAuth bearer (`1004 login fail`) — every step of that chain is pinned by tests (new `tests/generation.spec.ts`, 12 cases).
+- **The differences from the desktop app, stated plainly**: H3 **cannot appear in the model picker like an ordinary model** (the host's `ctx.llm` has a single chat-adapter registration point, and the model descriptor has no output-modality field), and **video has no carrier in DSH at all** (not one video-related declaration across the four packages). A picture can reach a conversation, but the attachment entry point is a browser upload with no library to pick from, so it still has to travel as part of the next prompt — that last hop is planned for the next release.
+
 ### 0.5.10 (2026-11-10)
 
 - **Removed the `awesome-dsh-plugin-submission/` directory**: the listings are no longer maintained, the draft entry in it was merged into the official registry long ago, and the rules it described are spelled out in the upstream `contributing.md`. Keeping it only suggests it is still needed. The README's mention of it is gone as well — where the changelog mentions the directory, the text stays but the link does not, so no dead link is left behind.

package/README.md

@@ -338,6 +338,38 @@ Authorization: Bearer <OAuth 令牌>
 
 这一节读到的数据**不需要订阅 M Plan**：该接口属于账户域，用 OAuth 令牌即可读取，与限额窗口（属于订阅）不同。
 
+### 图像与视频生成（H3）
+
+插件页多了一个「生成」面板，可以直接用 MiniMax 的 **H3 生图**与 **H3 视频生成**：
+
+- 图片：`MiniMax-H3-Image`（另有 `MiniMax-H3-Image-Highspeed`），可选画幅与 1K / 2K / 4K 清晰度
+- 视频：`MiniMax-H3`，可选 5–15 秒、768P / 2K 与画幅
+
+两者走的都是**开放平台官方文档给出的接口**：
+
+```
+POST https://api.minimax.cn/v1/images/generations          # 文生图，同步返回
+POST https://api.minimax.cn/v2/video_generation           # 创建视频任务，异步
+GET  https://api.minimax.cn/v2/query/video_generation/{id}  # 轮询任务
+```
+
+**不需要用户配置任何密钥。** 生成接口要的是开放平台的 API 密钥（`sk-cp-…`），而这个密钥是**账户自己的**：插件用 OAuth 令牌问 `token_plan_credit`（它本来就在读这个接口拿积分），响应里的 `api_key` 就是它。实测（2026-11-10）：
+
+| 请求 | 鉴权 | 结果 |
+|---|---|---|
+| `/v1/images/generations` + `sk-cp-` 密钥 | 200，8.5 秒出图 | 1184×880 JPEG，395 KB，6401 tokens |
+| `/v1/images/generations` + OAuth 令牌 | 401 `1004 login fail` | 生成接口只认 API 密钥 |
+| `/v2/video_generation` + `sk-cp-` 密钥 | 200，任务 90 秒后 `succeeded` | 5 秒 768P 成片，162745 tokens |
+
+生成的图片会**存进 DSH 的附件库**（`ctx.attachments`），所以它留在 DSH 自己的存储里，而不只是这个浏览器标签页。
+
+#### 与桌面 App 的两处不同（实测所限，如实说明）
+
+1. **不能像普通模型那样出现在模型选择器里。** DSH 宿主的 `ctx.llm` 只有对话适配器这一个注册点（`registerAdapter`），没有图像 provider 的挂载处；模型描述符 `LlmModelInfo` 也没有「输出模态」字段，选择器无从区分图像模型。pi-ai 自己虽有图像生成 API（`ImagesProvider` / `generateImages`），但 DSH 从不构造 `ImagesModels`，插件自己 `new` 一个也没有人会读。这三点都是宿主层面的事，插件改不动。
+2. **视频在 DSH 侧目前没有承载。** `pi-ai`、`dsh-llm`、`dsh-attachment`、`dsh-llm-pi-ai` 四个包里关于 video 的声明各为零处 —— 没有视频内容块、没有附件媒体类型、没有渲染。视频因此只能生成后在面板里播放与下载，不能像图片那样进对话。
+
+图片进对话是可行的：宿主公开了 `ctx.attachments.admitPromptContent()`，插件可以独立完成「生成 → 落库 → 拿到持久引用」；但附件入口是浏览器 `File` 上传，**没有附件库**可供挑选，因此要进对话仍需作为下一轮 prompt 的一部分提交。这一步已由 0.6.1 完成（见下）。
+
 ### 关于续期
 
 OAuth 令牌有效期约 1 小时，由客户端自动续期。插件**只在自己的令牌已过期时**才用 App 已存的 refresh token 续一次，并把轮换后的新令牌按客户端**原格式原子写回**同一个文件（`generation` 保持不变），因此桌面 App 的登录态不受影响。插件不会发起登录、不会登出、也不会调用任何付费/订阅类接口。
@@ -363,6 +395,20 @@ OAuth 令牌有效期约 1 小时，由客户端自动续期。插件**只在自
 
 ## 更新记录
 
+### 0.6.1（2026-11-11）
+
+- **生成的图片现在能直接进对话**：输入框工具行新增一个「生成图片」入口，写下提示词点生成，图片会作为**这一轮消息的一部分**发给模型 —— 走的是宿主自己那条路（`ctx.sessions.binding(id).session.prompt(content, mode, signal, requestId)`，与输入框按发送键完全相同），因此图片会被宿主校验、落成持久附件，并按普通消息参与对话。
+- 宿主若不暴露会话句柄，该入口会明确提示「无法发送到对话」，而不是静默失败；插件页的生成面板仍可照常使用。
+- 新增 `tests/generate-dock.spec.ts`（4 个用例）钉住：入口注册在工具行、声明了 `sessions` 服务、生成的图以 `{type:"image", mediaType, data}` 的形态（原始 base64，不带 data-url 前缀）随提示词一起发出、以及没有会话句柄时的提示。全量 252 个用例通过。
+
+### 0.6.0（2026-11-11）
+
+- **新增 H3 图像与视频生成**：插件页多了一个「生成」面板，可直接调用 MiniMax 的 `MiniMax-H3-Image`（文生图）与 `MiniMax-H3`（视频生成）。走的是开放平台官方文档给出的接口（`POST /v1/images/generations`、`POST /v2/video_generation` + 轮询 `/v2/query/video_generation/{id}`），**用户无需配置任何密钥** —— 生成接口要的是账户自己的开放平台密钥，插件用 OAuth 令牌问本来就在读的 `token_plan_credit` 就能拿到。
+- **生成的图片存入 DSH 的附件库**（`ctx.attachments`），因此它留在 DSH 自己的存储里，而不只是浏览器标签页。
+- 选项（画幅、清晰度、时长、分辨率）由服务端下发，取的是上游真正提供的那一组。
+- **实测**（2026-11-10）：图片 8.5 秒出图，1184×880 JPEG / 395 KB / 6401 tokens；视频 5 秒 768P，任务 90 秒后 `succeeded`，162745 tokens。生成接口只认 API 密钥，OAuth 令牌会被拒（`1004 login fail`）—— 这条链路的每一步都有测试钉住（新增 `tests/generation.spec.ts`，12 个用例）。
+- **如实说明与桌面 App 的差异**：H3 **不能像普通模型那样出现在模型选择器里**（宿主的 `ctx.llm` 只有对话适配器一个注册点，模型描述符也没有输出模态字段），**视频在 DSH 侧目前完全没有承载**（四个包里关于 video 的声明各为零处）。图片进对话是可行的，但附件入口只有浏览器上传、没有附件库，因此仍需作为下一轮 prompt 的一部分提交 —— 这一步的宿主接线留待下一版。
+
 ### 0.5.10（2026-11-10）
 
 - **删除 `awesome-dsh-plugin-submission/` 目录**：收录已停止维护，目录里的草稿条目早已合入官方注册表、提交规则在上游 `contributing.md` 里也有完整说明，留着只会让人误以为还需要它。README 中对应的说明一并去掉（历史更新记录里提到该目录的地方保留文字、只去掉链接，不留死链）。

package/lib/client.js

@@ -19,6 +19,7 @@
     const NAMESPACE = 'dsh-connect-minimaxcode';
     const STATUS_PATH = '/plugins/dsh-connect-minimaxcode/status';
     const ACTION_PATH = '/plugins/dsh-connect-minimaxcode/action';
+    const GENERATE_PATH = '/plugins/dsh-connect-minimaxcode/generate';
     const POLL_INTERVAL_MS = 20000;
     /**
      * The provider id the host side registers with `llm.registerAdapter`.
@@ -75,6 +76,28 @@
         usageRanking: '用量排名前 {percent}%',
         usageBestDay: '最活跃的一天 {date}',
         planNoSubscription: '未订阅 M Plan（仅积分包），因此没有限额窗口',
+        generateTitle: '生成',
+        generateImage: '图片',
+        generateVideo: '视频',
+        generatePrompt: '描述你想要的画面',
+        generateAspect: '画幅',
+        generateArea: '清晰度',
+        generateDuration: '时长',
+        generateResolution: '分辨率',
+        generateRatio: '画幅',
+        generateRun: '生成',
+        generateRunning: '生成中…',
+        generateVideoRunning: '视频生成中，约需 1–2 分钟…',
+        generateImageDone: '图片已生成',
+        generateVideoDone: '视频已生成',
+        generateDownload: '下载',
+        generateSaved: '已存入 DSH 附件库',
+        generateFailed: '生成失败',
+        generateHint: '生成走 MiniMax 的开放平台，用的是你账号自己的额度。图片会并存进 DSH 的附件库。',
+        generateToChat: '生成图片',
+        generateToChatHint: '生成的图片会作为这一轮消息的一部分发给模型。',
+        generateSent: '图片已发送到对话',
+        generateSendFailed: '无法发送到对话，请到插件页生成后下载。',
     };
     const en = {
         title: 'MiniMax Code credits',
@@ -124,6 +147,28 @@
         usageRanking: 'Top {percent}% by usage',
         usageBestDay: 'Busiest day {date}',
         planNoSubscription: 'No M Plan subscription (credits only), so there are no usage windows',
+        generateTitle: 'Generate',
+        generateImage: 'Image',
+        generateVideo: 'Video',
+        generatePrompt: 'Describe what you want',
+        generateAspect: 'Aspect',
+        generateArea: 'Area',
+        generateDuration: 'Length',
+        generateResolution: 'Resolution',
+        generateRatio: 'Aspect',
+        generateRun: 'Generate',
+        generateRunning: 'Generating…',
+        generateVideoRunning: 'Generating the video, about 1–2 minutes…',
+        generateImageDone: 'Image ready',
+        generateVideoDone: 'Video ready',
+        generateDownload: 'Download',
+        generateSaved: 'Saved to DSH attachments',
+        generateFailed: 'Generation failed',
+        generateHint: 'Generation runs on MiniMax\'s open platform with your own account\'s allowance. Images are also stored in DSH\'s attachment store.',
+        generateToChat: 'Generate an image',
+        generateToChatHint: 'The generated picture is sent as part of this turn.',
+        generateSent: 'The picture is on its way to the conversation',
+        generateSendFailed: 'Could not send it to the conversation; generate on the plugin page and download it instead.',
     };
     const loader = globalThis.__ModuleLoader__;
     if (loader === undefined)
@@ -133,6 +178,13 @@
         factory(require) {
             const React = require('react');
             const h = React.createElement;
+            /**
+             * The host's client context, once the plugin is applied.
+             *
+             * The components are closures over the module, not over the context, so
+             * this is how they reach the host's services.
+             */
+            let clientContext;
             /** Format a credit amount without pretending to more precision. */
             function credits(value, digits) {
                 if (value === undefined || !Number.isFinite(value))
@@ -382,6 +434,110 @@
                     return null;
                 return h(CreditPill, props);
             }
+            /**
+             * Generate a picture and send it to the conversation.
+             *
+             * The composer already knows how to send a message with an image in it,
+             * and the picture the upstream produces is an ordinary image file, so
+             * the dock does the same thing the composer does: it asks the session
+             * to accept a turn whose content is the picture plus the prompt.
+             *
+             * Nothing here is required of the host beyond what the composer itself
+             * uses. A host that does not expose the session face simply gets the
+             * message that says so, and the plugin page still offers the picture.
+             */
+            function GenerateDock(props) {
+                const t = props.t ?? ((key, params) => {
+                    const template = zh[key] ?? key;
+                    return template.replace(/\\{(\\w+)\\}/g, (_match, name) => String(params?.[name] ?? ''));
+                });
+                const [open, setOpen] = React.useState(false);
+                const [prompt, setPrompt] = React.useState('');
+                const [busy, setBusy] = React.useState(false);
+                const [note, setNote] = React.useState('');
+                /**
+                 * The session's own send path.
+                 *
+                 * `binding(id).session.prompt(content, mode, signal, requestId)` is
+                 * what the composer calls when the person presses send, so a picture
+                 * sent this way is a normal turn.
+                 */
+                const sessionFace = () => {
+                    const sessionId = props.sessionId;
+                    if (sessionId === undefined)
+                        return undefined;
+                    const binding = clientContext?.sessions?.binding?.(sessionId);
+                    const send = binding?.session?.prompt;
+                    return typeof send === 'function' ? { prompt: send } : undefined;
+                };
+                /** Turn a data URL into the parts the session's admission accepts. */
+                const imagePart = (dataUrl, mediaType) => {
+                    const comma = dataUrl.indexOf(',');
+                    if (comma < 0)
+                        return undefined;
+                    return { type: 'image', mediaType, data: dataUrl.slice(comma + 1) };
+                };
+                async function send() {
+                    const text = prompt.trim();
+                    if (text === '')
+                        return;
+                    setBusy(true);
+                    setNote('');
+                    try {
+                        const response = await fetch(GENERATE_PATH, {
+                            method: 'POST',
+                            headers: { 'content-type': 'application/json' },
+                            body: JSON.stringify({ kind: 'image', prompt: text, area: '1K' }),
+                        });
+                        const body = await response.json();
+                        if (body.ok !== true || body.dataUrl === undefined) {
+                            setNote(body.error ?? t('generateFailed'));
+                            return;
+                        }
+                        const part = imagePart(body.dataUrl, body.mediaType ?? 'image/jpeg');
+                        const face = sessionFace();
+                        if (part === undefined || face === undefined) {
+                            setNote(t('generateSendFailed'));
+                            return;
+                        }
+                        await face.prompt([part, { type: 'text', text }], 'queue', undefined, `minimax-h3-${Date.now()}`);
+                        setPrompt('');
+                        setOpen(false);
+                        setNote(t('generateSent'));
+                    }
+                    catch {
+                        setNote(t('generateSendFailed'));
+                    }
+                    finally {
+                        setBusy(false);
+                    }
+                }
+                if (!open) {
+                    return h('button', {
+                        type: 'button',
+                        style: buttonStyle,
+                        title: t('generateToChatHint'),
+                        onClick: () => { setOpen(true); setNote(''); },
+                    }, t('generateToChat'));
+                }
+                return h('div', { style: dockStyle, 'data-minimax-generate-dock': true }, h('input', {
+                    value: prompt,
+                    placeholder: t('generatePrompt'),
+                    style: dockInputStyle,
+                    disabled: busy,
+                    onChange: (event) => setPrompt(event.target.value),
+                }), h('button', {
+                    type: 'button',
+                    style: buttonStyle,
+                    disabled: busy || prompt.trim() === '',
+                    onClick: () => { void send(); },
+                }, busy ? t('generateRunning') : t('generateRun')), h('button', {
+                    type: 'button',
+                    style: buttonStyle,
+                    disabled: busy,
+                    onClick: () => { setOpen(false); setPrompt(''); setNote(''); },
+                }, '✕'), note === '' ? null : h('span', { style: mutedStyle }, note));
+            }
             /** Ask the host to do something, and let the card read the result. */
             async function act(action, body) {
                 try {
@@ -468,6 +624,144 @@
              * The summary view is the one line the plugin list shows; the page view
              * is the full card, where the expiry detail belongs.
              */
+            /**
+             * The generation panel.
+             *
+             * A picture comes back in one answer; a video is a task that takes a
+             * minute and a half, so it is polled until the upstream says it is
+             * done. The choices come from the server rather than from this half, so
+             * they are the ones the upstream actually serves.
+             */
+            function GenerationPanel(props) {
+                const [status] = useStatus();
+                const t = props.t ?? ((key, params) => {
+                    const template = zh[key] ?? key;
+                    return template.replace(/\\{(\\w+)\\}/g, (_match, name) => String(params?.[name] ?? ''));
+                });
+                const options = status?.generation;
+                const [mode, setMode] = React.useState('image');
+                const [prompt, setPrompt] = React.useState('');
+                const [model, setModel] = React.useState('');
+                const [aspect, setAspect] = React.useState('auto');
+                const [area, setArea] = React.useState('4K');
+                const [duration, setDuration] = React.useState(5);
+                const [resolution, setResolution] = React.useState('2K');
+                const [ratio, setRatio] = React.useState('16:9');
+                const [busy, setBusy] = React.useState(false);
+                const [note, setNote] = React.useState('');
+                const [picture, setPicture] = React.useState(undefined);
+                const [video, setVideo] = React.useState(undefined);
+                const imageModels = options?.imageModels ?? ['MiniMax-H3-Image'];
+                /** Ask for a picture; the answer carries the bytes inline. */
+                async function makeImage() {
+                    setBusy(true);
+                    setNote('');
+                    setVideo(undefined);
+                    try {
+                        const response = await fetch(GENERATE_PATH, {
+                            method: 'POST',
+                            headers: { 'content-type': 'application/json' },
+                            body: JSON.stringify({
+                                kind: 'image',
+                                prompt,
+                                model: model === '' ? undefined : model,
+                                aspectRatio: aspect,
+                                area,
+                            }),
+                        });
+                        const body = await response.json();
+                        if (body.ok !== true || body.dataUrl === undefined) {
+                            setNote(body.error ?? t('generateFailed'));
+                            return;
+                        }
+                        setPicture({ dataUrl: body.dataUrl, size: body.size, saved: body.attachment !== undefined });
+                    }
+                    catch {
+                        setNote(t('generateFailed'));
+                    }
+                    finally {
+                        setBusy(false);
+                    }
+                }
+                /** Start a video task, then wait for it the way the upstream does. */
+                async function makeVideo() {
+                    setBusy(true);
+                    setNote('');
+                    setPicture(undefined);
+                    try {
+                        const response = await fetch(GENERATE_PATH, {
+                            method: 'POST',
+                            headers: { 'content-type': 'application/json' },
+                            body: JSON.stringify({
+                                kind: 'video',
+                                prompt,
+                                duration,
+                                resolution,
+                                ratio,
+                            }),
+                        });
+                        const body = await response.json();
+                        if (body.ok !== true || body.taskId === undefined) {
+                            setNote(body.error ?? t('generateFailed'));
+                            return;
+                        }
+                        // The task takes a minute and a half; the upstream is asked every
+                        // ten seconds, the way its own documentation recommends.
+                        for (let attempt = 0; attempt < 60; attempt += 1) {
+                            await new Promise(resolve => setTimeout(resolve, 10000));
+                            const polled = await fetch(`${GENERATE_PATH}?taskId=${encodeURIComponent(body.taskId)}`);
+                            const task = await polled.json();
+                            if (task.url !== undefined) {
+                                setVideo({ url: task.url, status: task.status });
+                                setNote('');
+                                return;
+                            }
+                            if (task.status !== undefined && /fail|cancel/i.test(task.status)) {
+                                setNote(task.error ?? t('generateFailed'));
+                                return;
+                            }
+                        }
+                        setNote(t('generateFailed'));
+                    }
+                    catch {
+                        setNote(t('generateFailed'));
+                    }
+                    finally {
+                        setBusy(false);
+                    }
+                }
+                const select = (value, onChange, values) => h('select', {
+                    value,
+                    style: selectStyle,
+                    onChange: (event) => onChange(event.target.value),
+                }, ...values.map(option => h('option', { key: option, value: option }, option)));
+                return h('div', { style: sectionBoxStyle, 'data-minimax-generation': true }, h('div', { style: sectionTitleStyle }, t('generateTitle')), h('div', { style: rowStyle }, h('button', {
+                    type: 'button',
+                    style: mode === 'image' ? tabActiveStyle : actionButtonStyle,
+                    onClick: () => { setMode('image'); setPicture(undefined); setVideo(undefined); },
+                }, t('generateImage')), h('button', {
+                    type: 'button',
+                    style: mode === 'video' ? tabActiveStyle : actionButtonStyle,
+                    onClick: () => { setMode('video'); setPicture(undefined); setVideo(undefined); },
+                }, t('generateVideo'))), h('textarea', {
+                    value: prompt,
+                    placeholder: t('generatePrompt'),
+                    style: promptStyle,
+                    rows: 3,
+                    onChange: (event) => setPrompt(event.target.value),
+                }), mode === 'image'
+                    ? h('div', { style: rowStyle }, select(model === '' ? imageModels[0] : model, setModel, imageModels), select(aspect, setAspect, options?.imageAspectRatios ?? ['auto', '16:9', '1:1']), select(area, setArea, options?.imageAreas ?? ['1K', '2K', '4K']))
+                    : h('div', { style: rowStyle }, select(String(duration), next => setDuration(Number(next)), (options?.videoDurations ?? [5, 10]).map(String)), select(resolution, setResolution, options?.videoResolutions ?? ['768P', '2K']), select(ratio, setRatio, options?.videoRatios ?? ['16:9', '9:16'])), h('div', { style: rowStyle }, h('button', {
+                    type: 'button',
+                    style: actionButtonStyle,
+                    disabled: busy || prompt.trim() === '',
+                    onClick: () => { void (mode === 'image' ? makeImage() : makeVideo()); },
+                }, busy ? (mode === 'video' ? t('generateVideoRunning') : t('generateRunning')) : t('generateRun'))), note === '' ? null : h('div', { style: mutedStyle }, note), picture === undefined
+                    ? null
+                    : h('div', { style: generationResultStyle, 'data-minimax-generation-image': true }, h('img', { src: picture.dataUrl, alt: prompt, style: generatedImageStyle }), h('div', { style: rowStyle }, h('span', null, picture.size === undefined ? t('generateImageDone') : `${t('generateImageDone')} · ${picture.size}`), picture.saved === true ? h('span', null, t('generateSaved')) : null, h('a', { href: picture.dataUrl, download: 'minimax-h3-image.jpg', style: linkStyle }, t('generateDownload')))), video === undefined
+                    ? null
+                    : h('div', { style: generationResultStyle, 'data-minimax-generation-video': true }, h('video', { src: video.url, controls: true, style: generatedVideoStyle }), h('div', { style: rowStyle }, h('span', null, t('generateVideoDone')), h('a', { href: video.url, download: 'minimax-h3-video.mp4', style: linkStyle }, t('generateDownload')))), h('div', { style: mutedStyle }, t('generateHint')));
+            }
             function PluginPage(props) {
                 const [status, refresh] = useStatus();
                 const t = props.t ?? ((key, params) => {
@@ -487,15 +781,19 @@
                         : '', signin?.today?.claimable === true ? ` · ${t('claimableToday')}` : '');
                 }
                 if (!available) {
-                    return h('div', { style: cardStyle, 'data-minimax-credits-card': true }, h('div', { style: cardTitleStyle }, t('title')), h('div', { style: rowStyle }, h('span', null, t('unavailable')), h('span', null, credits_?.state === 'expired' ? t('expired') : '')), loginBlock(t, login, refresh));
+                    return h('div', { style: pageStyle, 'data-minimax-page': true }, h('div', { style: cardStyle, 'data-minimax-credits-card': true }, h('div', { style: cardTitleStyle }, t('title')), h('div', { style: rowStyle }, h('span', null, t('unavailable')), h('span', null, credits_?.state === 'expired' ? t('expired') : '')), loginBlock(t, login, refresh)), 
+                    // The generation panel does not need the balance: it spends the
+                    // account's generation allowance, not its credits, so it stays
+                    // usable even when the balance cannot be read.
+                    h(GenerationPanel, props));
                 }
-                return h('div', { style: cardStyle, 'data-minimax-credits-card': true }, h('div', { style: cardTitleStyle }, t('title')), h('div', { style: bigNumberStyle }, t('balance', { total: credits(credits_.total, 2) })), row(t('free'), credits(credits_.free, 3)), row(t('purchased'), credits(credits_.purchased, 3)), expiryBlock(t, credits_), rateLimitBlock(t, status?.rateLimit), planNotice(t, status?.rateLimit), signinBlock(t, signin, refresh), h('div', { style: sectionTitleStyle }, t('lots')), (credits_.upcoming ?? []).length === 0
+                return h('div', { style: pageStyle, 'data-minimax-page': true }, h('div', { style: cardStyle, 'data-minimax-credits-card': true }, h('div', { style: cardTitleStyle }, t('title')), h('div', { style: bigNumberStyle }, t('balance', { total: credits(credits_.total, 2) })), row(t('free'), credits(credits_.free, 3)), row(t('purchased'), credits(credits_.purchased, 3)), expiryBlock(t, credits_), rateLimitBlock(t, status?.rateLimit), planNotice(t, status?.rateLimit), signinBlock(t, signin, refresh), h('div', { style: sectionTitleStyle }, t('lots')), (credits_.upcoming ?? []).length === 0
                     ? h('div', { style: mutedStyle }, t('noLots'))
                     : h('ul', { style: listStyle }, ...(credits_.upcoming ?? []).map((lot, index) => h('li', { key: `${lot.expireAtMs}-${index}`, style: rowStyle }, h('span', null, day(lot.expireAtMs)), h('span', null, credits(lot.amount, 2))))), usage !== undefined
                     ? h('div', { style: usageBoxStyle }, row(t('tokens', { count: tokens(usage.inputTokens + usage.outputTokens + usage.cacheReadTokens) }), ''), row(t('cost', { amount: money(usage.estimatedCny) }), ''), usage.requests > 0 ? row(t('requests', { count: usage.requests }), usage.lastModel ?? '') : null)
                     : null, usageProfileBlock(t, status?.usageProfile), loginBlock(t, login, refresh), credits_.updatedAtMs !== undefined
                     ? h('div', { style: mutedStyle }, t('updated', { time: new Date(credits_.updatedAtMs).toLocaleTimeString() }))
-                    : null);
+                    : null), h(GenerationPanel, props));
             }
             const wrapStyle = {
                 position: 'relative',
@@ -515,6 +813,21 @@
                 opacity: 0.8,
                 padding: '1px 8px',
             };
+            const dockStyle = {
+                display: 'flex',
+                alignItems: 'center',
+                gap: '4px',
+            };
+            const dockInputStyle = {
+                width: '160px',
+                padding: '2px 6px',
+                borderRadius: '999px',
+                border: '1px solid var(--dsh-composer-dock-inset, currentColor)',
+                background: 'transparent',
+                color: 'inherit',
+                font: 'inherit',
+                fontSize: '12px',
+            };
             const detailStyle = {
                 position: 'absolute',
                 bottom: 'calc(100% + 6px)',
@@ -550,6 +863,12 @@
                 fontSize: '13px',
                 lineHeight: 1.6,
             };
+            // The page stacks the balance card and the generation panel.
+            const pageStyle = {
+                display: 'flex',
+                flexDirection: 'column',
+                gap: '10px',
+            };
             const cardTitleStyle = { fontWeight: 600 };
             const bigNumberStyle = { fontSize: '20px', fontWeight: 600, marginBottom: '2px' };
             const sectionTitleStyle = { marginTop: '8px', fontWeight: 600 };
@@ -597,10 +916,61 @@
                 borderTop: '1px solid var(--dsh-panel-border, currentColor)',
                 opacity: 0.9,
             };
+            const tabActiveStyle = {
+                appearance: 'none',
+                padding: '2px 8px',
+                borderRadius: '6px',
+                border: '1px solid var(--dsh-panel-border, currentColor)',
+                background: 'var(--dsh-panel-bg, transparent)',
+                color: 'inherit',
+                font: 'inherit',
+                cursor: 'pointer',
+                fontWeight: 600,
+            };
+            const promptStyle = {
+                width: '100%',
+                boxSizing: 'border-box',
+                padding: '6px',
+                borderRadius: '6px',
+                border: '1px solid var(--dsh-panel-border, currentColor)',
+                background: 'transparent',
+                color: 'inherit',
+                font: 'inherit',
+                resize: 'vertical',
+            };
+            const selectStyle = {
+                padding: '2px 4px',
+                borderRadius: '6px',
+                border: '1px solid var(--dsh-panel-border, currentColor)',
+                background: 'transparent',
+                color: 'inherit',
+                font: 'inherit',
+            };
+            const generationResultStyle = {
+                marginTop: '8px',
+                display: 'flex',
+                flexDirection: 'column',
+                gap: '6px',
+            };
+            const generatedImageStyle = {
+                maxWidth: '100%',
+                borderRadius: '8px',
+                border: '1px solid var(--dsh-panel-border, currentColor)',
+            };
+            const generatedVideoStyle = {
+                maxWidth: '100%',
+                borderRadius: '8px',
+            };
             return {
-                inject: ['slots', 'locale'],
+                // The sessions service is what lets a generated picture be delivered
+                // into the conversation the composer belongs to.
+                inject: ['slots', 'locale', 'sessions'],
                 apply(ctx) {
                     const context = ctx;
+                    // The components need the host's sessions service to deliver a
+                    // generated picture into the conversation, so it is kept where the
+                    // components can reach it.
+                    clientContext = context;
                     // Visible text goes through the host's locale service.
                     context.effect(() => context.locale.register(NAMESPACE, { zh, en }), `${BUNDLE_NAME}: locale`);
                     // The composer and plugin-page slots carry no translator in their
@@ -620,6 +990,15 @@
                         locale: NAMESPACE,
                         inject: () => ({ t }),
                     }, CreditPillGate));
+                    // The image dock sits beside the balance: it is the same provider's
+                    // capability, and it needs the session the composer belongs to.
+                    context.slots.inject('conversation.input.left', () => context.slots.register({
+                        name: 'conversation.input.left',
+                        id: 'minimax-generate',
+                        order: 5,
+                        locale: NAMESPACE,
+                        inject: () => ({ t }),
+                    }, GenerateDock));
                     // The plugin's own page, keyed by the package name, where the
                     // balance and the expiry schedule are spelled out.
                     context.slots.inject('plugins.bundle.config', () => context.slots.register({

package/lib/index.d.ts

@@ -1038,6 +1038,21 @@ interface UsageProfileStatus {
   /** Why the profile could not be read. */
   error?: string;
 }
+/** The choices the generation endpoints offer, as the card needs them. */
+interface GenerationOptions {
+  /** The image models that can be asked for. */
+  imageModels: string[];
+  /** The aspect ratios an image can be made in. */
+  imageAspectRatios: string[];
+  /** The output areas an image can be made at. */
+  imageAreas: string[];
+  /** The lengths a video can be, in seconds. */
+  videoDurations: number[];
+  /** The output resolutions a video can be made at. */
+  videoResolutions: string[];
+  /** The aspect ratios a video can be made in. */
+  videoRatios: string[];
+}
 /** What the card renders. */
 interface StatusDocument {
   state: 'signed-in' | 'signed-out' | 'expired';
@@ -1070,6 +1085,8 @@ interface StatusDocument {
   login?: LoginStatus;
   /** The account's usage profile, when it could be read. */
   usageProfile?: UsageProfileStatus;
+  /** The choices the generation endpoints offer. */
+  generation?: GenerationOptions;
 }
 /** Inputs the document is derived from. */
 interface StatusDeps {
@@ -1100,6 +1117,8 @@ interface StatusDeps {
   login?: LoginStatus;
   /** The account's usage profile, when the account API could be reached. */
   usageProfile?: UsageProfileStatus;
+  /** The choices the generation endpoints offer. */
+  generation?: GenerationOptions;
 }
 /**
  * Build the card's status document.

package/lib/index.js

@@ -738,6 +738,14 @@ function imageInputHooks(ctx) {
 const MINIMAX_STATUS_PATH = "/plugins/dsh-connect-minimaxcode/status";
 /** Route the card posts to when it wants something done rather than read. */
 const MINIMAX_ACTION_PATH = "/plugins/dsh-connect-minimaxcode/action";
+/**
+* Route the card posts to when it wants a picture or a video made.
+*
+* It is a route of its own rather than another action because generation is
+* expensive and asynchronous: an action answers at once, while a video task
+* takes a minute and a half and is polled.
+*/
+const MINIMAX_GENERATE_PATH = "/plugins/dsh-connect-minimaxcode/generate";
 const HINTS = {
 	"app-missing": "未找到 MiniMax Code 的登录信息。请先安装并登录 MiniMax Code 桌面 App，然后重启 DSH。",
 	"not-signed-in": "MiniMax Code 尚未登录。请在 MiniMax Code 桌面 App 中登录后重启 DSH。",
@@ -767,7 +775,8 @@ function statusDocument(deps) {
 		...deps.usage === void 0 ? {} : { usage: deps.usage },
 		...deps.signin === void 0 ? {} : { signin: deps.signin },
 		...deps.login === void 0 ? {} : { login: deps.login },
-		...deps.usageProfile === void 0 ? {} : { usageProfile: deps.usageProfile }
+		...deps.usageProfile === void 0 ? {} : { usageProfile: deps.usageProfile },
+		...deps.generation === void 0 ? {} : { generation: deps.generation }
 	};
 	if (auth.state === "expired") return {
 		...base,
@@ -792,6 +801,305 @@ function isLoopback(host) {
 	return name === "127.0.0.1" || name === "localhost" || name === "[::1]" || name === "::1";
 }
 //#endregion
+//#region src/generation.ts
+/** Open-platform API host per region. */
+const MINIMAX_API_HOSTS = {
+	cn: "https://api.minimax.cn",
+	io: "https://api.minimaxi.com",
+	com: "https://api.minimaxi.com"
+};
+/** Image generation endpoint: the documented OpenAI-style path. */
+const IMAGE_GENERATIONS_PATH = "/v1/images/generations";
+/** The endpoint that hands out the account's open-platform key. */
+const PLAN_CREDIT_PATH = "/backend/account/token_plan_credit";
+/** The host the plan key comes from, per region. */
+const PLAN_CREDIT_HOSTS = {
+	cn: "https://www.minimaxi.com",
+	io: "https://www.minimaxi.com",
+	com: "https://www.minimaxi.com"
+};
+/** Video generation task endpoint. */
+const VIDEO_GENERATION_PATH = "/v2/video_generation";
+/** Video task query endpoint; the task id is the path segment. */
+const VIDEO_QUERY_PATH = "/v2/query/video_generation";
+/** The image models the endpoint serves. */
+const IMAGE_MODELS = ["MiniMax-H3-Image", "MiniMax-H3-Image-Highspeed"];
+/** The video model the endpoint serves. */
+const VIDEO_MODEL = "MiniMax-H3";
+/** Aspect ratios the desktop app offers for images. */
+const IMAGE_ASPECT_RATIOS = [
+	"auto",
+	"16:9",
+	"3:2",
+	"4:3",
+	"5:4",
+	"1:1",
+	"4:5",
+	"3:4",
+	"2:3",
+	"9:16",
+	"3:1",
+	"1:3"
+];
+/** Output areas the endpoint serves. */
+const IMAGE_AREAS = [
+	"1K",
+	"2K",
+	"4K"
+];
+/** Aspect ratios the desktop app offers for video. */
+const VIDEO_RATIOS = [
+	"21:9",
+	"16:9",
+	"4:3",
+	"1:1",
+	"3:4",
+	"9:16"
+];
+/** Output resolutions the endpoint serves for video. */
+const VIDEO_RESOLUTIONS = ["768P", "2K"];
+/** Durations the endpoint serves for video, in seconds. */
+const VIDEO_DURATIONS = [
+	5,
+	6,
+	7,
+	8,
+	9,
+	10,
+	11,
+	12,
+	13,
+	14,
+	15
+];
+/** A generation call that the upstream refused, with the reason it gave. */
+var GenerationError = class extends Error {
+	/** The upstream's own status code, when it gave one. */
+	code;
+	constructor(message, code) {
+		super(message);
+		this.name = "GenerationError";
+		this.code = code;
+	}
+};
+/**
+* Read the account's open-platform key.
+*
+* The credit endpoint answers the desktop app's OAuth bearer, and its
+* answer carries the key. It is not a secret the person had to configure: it
+* belongs to the account, the same way the balance does.
+*/
+async function fetchPlanApiKey(credential, region, fetchImpl = fetch) {
+	const response = await fetchImpl(`${PLAN_CREDIT_HOSTS[region]}${PLAN_CREDIT_PATH}?business=Agent&device_platform=desktop`, { headers: {
+		accept: "application/json",
+		authorization: `Bearer ${credential.token}`
+	} });
+	if (!response.ok) throw new GenerationError(`token plan credit request failed: HTTP ${response.status}`, response.status);
+	const key = (await response.json()).api_key;
+	return typeof key === "string" && key.length > 0 ? key : void 0;
+}
+/** The upstream's error message, when the body carries one. */
+function readErrorMessage(body) {
+	if (typeof body !== "object" || body === null) return void 0;
+	const record = body;
+	const error = record.error;
+	if (typeof error === "object" && error !== null) {
+		const message = error.message;
+		if (typeof message === "string") return message;
+	}
+	const base = record.base_resp;
+	if (typeof base === "object" && base !== null) {
+		const message = base.status_msg;
+		if (typeof message === "string") return message;
+	}
+}
+/**
+* The upstream's own refusal, when it answers with one.
+*
+* The generation endpoints do not always use the HTTP status to say no: a
+* task that does not exist comes back as `200` with
+* `base_resp.status_code: 2013`. Reading only the HTTP status would turn
+* that into an empty answer and leave the caller polling a task that will
+* never arrive, so the body's own code is checked too.
+*/
+function readBodyRefusal(body) {
+	const code = readErrorCode(body);
+	if (code === void 0 || code === 0) return void 0;
+	return {
+		message: readErrorMessage(body) ?? `the upstream refused the request (${code})`,
+		code
+	};
+}
+/** The upstream's own code, when the body carries one. */
+function readErrorCode(body) {
+	if (typeof body !== "object" || body === null) return void 0;
+	const record = body;
+	const error = record.error;
+	if (typeof error === "object" && error !== null) {
+		const code = error.http_code;
+		const parsed = typeof code === "string" ? Number(code) : code;
+		if (typeof parsed === "number" && Number.isFinite(parsed)) return parsed;
+	}
+	const base = record.base_resp;
+	if (typeof base === "object" && base !== null) {
+		const code = base.status_code;
+		if (typeof code === "number" && Number.isFinite(code)) return code;
+	}
+}
+/**
+* Call a generation endpoint and turn its refusal into a readable error.
+*
+* The upstream reports a bad parameter as HTTP 400 with a code in the
+* message, and a bad credential as HTTP 401; both arrive as an error body
+* whose `error.message` says what was wrong. The message is kept verbatim
+* because it names the offending field.
+*/
+async function callGeneration(apiKey, url, body, fetchImpl) {
+	const response = await fetchImpl(url, {
+		method: "POST",
+		headers: {
+			accept: "application/json",
+			authorization: `Bearer ${apiKey}`,
+			"content-type": "application/json"
+		},
+		body: JSON.stringify(body)
+	});
+	const text = await response.text();
+	let document;
+	try {
+		document = JSON.parse(text);
+	} catch {
+		if (!response.ok) throw new GenerationError(`generation request failed: HTTP ${response.status}`, response.status);
+		throw new GenerationError("generation response was not JSON");
+	}
+	if (!response.ok) throw new GenerationError(readErrorMessage(document) ?? `generation request failed: HTTP ${response.status}`, readErrorCode(document) ?? response.status);
+	const refusal = readBodyRefusal(document);
+	if (refusal !== void 0) throw new GenerationError(refusal.message, refusal.code);
+	return document ?? {};
+}
+/**
+* Generate images from a prompt.
+*
+* The bytes are fetched while the URL is still fresh: the endpoint's link
+* expires in 24 hours, and a plugin that only kept the link would leave the
+* picture behind.
+*/
+async function generateImages(options) {
+	const fetchImpl = options.fetchImpl ?? fetch;
+	const body = {
+		model: options.model ?? IMAGE_MODELS[0],
+		prompt: options.prompt,
+		response_format: "url"
+	};
+	if (options.count !== void 0) body.n = options.count;
+	if (options.aspectRatio !== void 0) body.aspect_ratio = options.aspectRatio;
+	if (options.area !== void 0) body.area = options.area;
+	const document = await callGeneration(options.apiKey, `${MINIMAX_API_HOSTS[options.region]}${IMAGE_GENERATIONS_PATH}`, body, fetchImpl);
+	const images = [];
+	for (const entry of document.data ?? []) {
+		if (typeof entry !== "object" || entry === null) continue;
+		const record = entry;
+		const image = {};
+		if (typeof record.url === "string") image.url = record.url;
+		if (typeof record.size === "string") image.size = record.size;
+		if (typeof record.revised_prompt === "string") image.revisedPrompt = record.revised_prompt;
+		if (typeof record.b64_json === "string") {
+			image.bytes = Buffer.from(record.b64_json, "base64");
+			image.mediaType = "image/jpeg";
+		}
+		images.push(image);
+	}
+	const usage = document.usage;
+	return {
+		images,
+		generatedImages: typeof usage?.generated_images === "number" ? usage.generated_images : void 0,
+		totalTokens: typeof usage?.total_tokens === "number" ? usage.total_tokens : void 0,
+		requestId: typeof document.request_id === "string" ? document.request_id : void 0
+	};
+}
+/**
+* Download a generated image's bytes.
+*
+* The link the endpoint hands out is good for a day; the picture is fetched
+* now so that nothing the plugin shows depends on that clock.
+*/
+async function downloadImage(url, fetchImpl = fetch) {
+	const response = await fetchImpl(url);
+	if (!response.ok) throw new GenerationError(`image download failed: HTTP ${response.status}`, response.status);
+	return {
+		bytes: new Uint8Array(await response.arrayBuffer()),
+		mediaType: (response.headers.get("content-type") ?? "image/jpeg").split(";")[0] ?? "image/jpeg"
+	};
+}
+/**
+* Start a video generation task.
+*
+* The task is asynchronous: the endpoint answers with an identifier and the
+* video appears later, so the caller polls {@link queryVideoTask}.
+*/
+async function startVideoTask(options) {
+	const fetchImpl = options.fetchImpl ?? fetch;
+	const content = [{
+		type: "text",
+		text: options.prompt
+	}];
+	if (options.firstFrameUrl !== void 0) content.push({
+		type: "image_url",
+		image_url: { url: options.firstFrameUrl },
+		role: "first_frame"
+	});
+	if (options.lastFrameUrl !== void 0) content.push({
+		type: "image_url",
+		image_url: { url: options.lastFrameUrl },
+		role: "last_frame"
+	});
+	const body = {
+		model: VIDEO_MODEL,
+		content
+	};
+	if (options.duration !== void 0) body.duration = options.duration;
+	if (options.resolution !== void 0) body.resolution = options.resolution;
+	if (options.ratio !== void 0) body.ratio = options.ratio;
+	const taskId = (await callGeneration(options.apiKey, `${MINIMAX_API_HOSTS[options.region]}${VIDEO_GENERATION_PATH}`, body, fetchImpl)).task_id;
+	if (typeof taskId !== "string" || taskId.length === 0) throw new GenerationError("the video task carried no task id");
+	return taskId;
+}
+/**
+* Read a video task's state.
+*
+* The finished video's address is in the task's content; the endpoint also
+* reports a plain `status` string for the simpler shape.
+*/
+async function queryVideoTask(options) {
+	const response = await (options.fetchImpl ?? fetch)(`${MINIMAX_API_HOSTS[options.region]}${VIDEO_QUERY_PATH}/${encodeURIComponent(options.taskId)}`, { headers: {
+		accept: "application/json",
+		authorization: `Bearer ${options.apiKey}`
+	} });
+	const text = await response.text();
+	let document;
+	try {
+		document = JSON.parse(text) ?? {};
+	} catch {
+		throw new GenerationError(`video task query failed: HTTP ${response.status}`, response.status);
+	}
+	if (!response.ok) throw new GenerationError(readErrorMessage(document) ?? `video task query failed: HTTP ${response.status}`, readErrorCode(document) ?? response.status);
+	const refusal = readBodyRefusal(document);
+	if (refusal !== void 0) throw new GenerationError(refusal.message, refusal.code);
+	const task = document.task ?? document;
+	const content = task.content;
+	const usage = task.usage;
+	return {
+		taskId: options.taskId,
+		status: typeof task.status === "string" ? task.status : void 0,
+		url: typeof content?.url === "string" ? content.url : void 0,
+		resolution: typeof task.resolution === "string" ? task.resolution : void 0,
+		duration: typeof task.duration === "number" ? task.duration : void 0,
+		ratio: typeof task.ratio === "string" ? task.ratio : void 0,
+		totalSeconds: typeof usage?.total_seconds === "number" ? usage.total_seconds : void 0,
+		totalTokens: typeof usage?.total_tokens === "number" ? usage.total_tokens : void 0
+	};
+}
+//#endregion
 //#region src/credits.ts
 /**
 * The MiniMax Code account's credit balance.
@@ -2783,7 +3091,15 @@ function apply(ctx) {
 			usage: buildUsageStatus(),
 			signin: buildSigninStatus(),
 			login: buildLoginStatus(),
-			usageProfile: buildUsageProfileStatus()
+			usageProfile: buildUsageProfileStatus(),
+			generation: {
+				imageModels: [...IMAGE_MODELS],
+				imageAspectRatios: [...IMAGE_ASPECT_RATIOS],
+				imageAreas: [...IMAGE_AREAS],
+				videoDurations: [...VIDEO_DURATIONS],
+				videoResolutions: [...VIDEO_RESOLUTIONS],
+				videoRatios: [...VIDEO_RATIOS]
+			}
 		});
 	};
 	(async () => {
@@ -2910,6 +3226,134 @@ function apply(ctx) {
 			};
 		}
 	};
+	/**
+	* The plan's own API key, remembered for the session.
+	*
+	* The generation endpoints want the open platform's key rather than the
+	* desktop app's OAuth bearer, and the key belongs to the account: the
+	* credit endpoint hands it out to the OAuth credential the plugin already
+	* holds. It is fetched once and kept for as long as the plugin runs.
+	*/
+	let planApiKey;
+	const resolvePlanApiKey = async () => {
+		if (planApiKey !== void 0) return planApiKey;
+		const account = await resolveAccountCredential();
+		if (account.state !== "signed-in") throw new GenerationError("尚未登录 MiniMax Code");
+		const key = await fetchPlanApiKey(account.credential, region);
+		if (key === void 0) throw new GenerationError("账户没有开放平台密钥，无法调用生成接口");
+		planApiKey = key;
+		return key;
+	};
+	/**
+	* Commit a generated picture to the host's attachment store.
+	*
+	* The store is read per request: the host may provide it after this plugin
+	* loads, and a host without one must still show the picture. The media type
+	* is narrowed to the four the store accepts, because it checks the declared
+	* type against the decoded bytes.
+	*
+	* @returns The durable reference, when the picture was stored.
+	*/
+	const saveGeneratedImage = async (bytes, mediaType) => {
+		const store = ctx.get("attachments");
+		if (store === void 0) return void 0;
+		const type = mediaType === "image/png" || mediaType === "image/webp" || mediaType === "image/gif" ? mediaType : "image/jpeg";
+		try {
+			const ref = await store.saveImage({
+				data: bytes,
+				mediaType: type,
+				name: "minimax-h3-image"
+			});
+			return {
+				attachmentId: ref.attachmentId,
+				width: ref.width,
+				height: ref.height,
+				bytes: ref.bytes
+			};
+		} catch {
+			return;
+		}
+	};
+	/**
+	* Make a picture or start a video.
+	*
+	* A picture comes back at once, as bytes: the upstream's link is good for
+	* a day, and a link that expires would leave the picture behind. A video
+	* is a task that takes a minute and a half, so this answers with the task's
+	* identifier and the card polls for the result.
+	*/
+	const runGeneration = async (body) => {
+		const kind = body.kind;
+		const prompt = typeof body.prompt === "string" ? body.prompt.trim() : "";
+		if (prompt.length === 0) return {
+			ok: false,
+			error: "请填写提示词"
+		};
+		const apiKey = await resolvePlanApiKey();
+		if (kind === "video") return {
+			ok: true,
+			kind: "video",
+			taskId: await startVideoTask({
+				apiKey,
+				region,
+				prompt,
+				duration: typeof body.duration === "number" ? body.duration : void 0,
+				resolution: typeof body.resolution === "string" ? body.resolution : void 0,
+				ratio: typeof body.ratio === "string" ? body.ratio : void 0,
+				firstFrameUrl: typeof body.firstFrameUrl === "string" ? body.firstFrameUrl : void 0,
+				lastFrameUrl: typeof body.lastFrameUrl === "string" ? body.lastFrameUrl : void 0
+			})
+		};
+		const result = await generateImages({
+			apiKey,
+			region,
+			prompt,
+			model: typeof body.model === "string" ? body.model : void 0,
+			aspectRatio: typeof body.aspectRatio === "string" ? body.aspectRatio : void 0,
+			area: typeof body.area === "string" ? body.area : void 0,
+			count: typeof body.count === "number" ? body.count : void 0
+		});
+		const first = result.images[0];
+		if (first === void 0) return {
+			ok: false,
+			error: "上游没有返回图片"
+		};
+		if (first.bytes === void 0 && first.url === void 0) return {
+			ok: false,
+			error: "上游没有返回图片"
+		};
+		const image = first.bytes !== void 0 ? {
+			bytes: first.bytes,
+			mediaType: first.mediaType ?? "image/jpeg"
+		} : await downloadImage(first.url);
+		const saved = await saveGeneratedImage(image.bytes, image.mediaType);
+		return {
+			ok: true,
+			kind: "image",
+			size: first.size,
+			revisedPrompt: first.revisedPrompt,
+			totalTokens: result.totalTokens,
+			requestId: result.requestId,
+			mediaType: image.mediaType,
+			dataUrl: `data:${image.mediaType};base64,${Buffer.from(image.bytes).toString("base64")}`,
+			...saved === void 0 ? {} : { attachment: saved }
+		};
+	};
+	/** Read a video task's state. */
+	const readVideoTask = async (taskId) => {
+		const task = await queryVideoTask({
+			apiKey: await resolvePlanApiKey(),
+			region,
+			taskId
+		});
+		return {
+			ok: true,
+			kind: "video",
+			taskId: task.taskId,
+			status: task.status,
+			url: task.url
+		};
+	};
 	ctx.inject(["webServer"], (webCtx) => {
 		webCtx.effect(() => webCtx.webServer.register({
 			kind: "exact",
@@ -2963,6 +3407,57 @@ function apply(ctx) {
 				}
 			}
 		}));
+		webCtx.effect(() => webCtx.webServer.register({
+			kind: "exact",
+			path: MINIMAX_GENERATE_PATH,
+			handler: async (req, res) => {
+				if (!isLoopback(req.headers.host)) {
+					res.writeHead(403, { "Content-Type": "application/json" });
+					res.end(JSON.stringify({ error: "loopback only" }));
+					return;
+				}
+				const url = new URL(req.url ?? "/", "http://127.0.0.1");
+				try {
+					if (req.method === "GET") {
+						const taskId = url.searchParams.get("taskId") ?? "";
+						if (taskId.length === 0) {
+							res.writeHead(400, { "Content-Type": "application/json" });
+							res.end(JSON.stringify({
+								ok: false,
+								error: "缺少 taskId"
+							}));
+							return;
+						}
+						res.writeHead(200, { "Content-Type": "application/json" });
+						res.end(JSON.stringify(await readVideoTask(taskId)));
+						return;
+					}
+					if (req.method !== "POST") {
+						res.writeHead(405, { "Content-Type": "application/json" });
+						res.end(JSON.stringify({ error: "POST only" }));
+						return;
+					}
+					let body = {};
+					try {
+						const text = await readRequestBody(req);
+						if (text.length > 0) {
+							const parsed = JSON.parse(text);
+							if (typeof parsed === "object" && parsed !== null) body = parsed;
+						}
+					} catch {}
+					res.writeHead(200, { "Content-Type": "application/json" });
+					res.end(JSON.stringify(await runGeneration(body)));
+				} catch (error) {
+					const status = error instanceof GenerationError ? 400 : 500;
+					res.writeHead(status, { "Content-Type": "application/json" });
+					res.end(JSON.stringify({
+						ok: false,
+						error: error instanceof Error ? error.message : String(error),
+						...error instanceof GenerationError && error.code !== void 0 ? { code: error.code } : {}
+					}));
+				}
+			}
+		}));
 	});
 }
 /** Read a request body, up to a size that no card needs to exceed. */

package/package.json

@@ -2,7 +2,7 @@
   "name": "dsh-connect-minimaxcode",
   "displayName": "DSH Connect MiniMax Code",
   "description": "把本机已登录的 MiniMax Code 桌面 App 模型接入 DeepSeek Harness：零配置、Anthropic Messages 兼容、动态模型目录，并显示积分余额与到期 — bring MiniMax Code desktop app models into DSH with a dynamic model catalog and a credit balance with expiry details.",
-  "version": "0.5.10",
+  "version": "0.6.1",
   "author": "JeffTsuiCUG",
   "contributors": [
     {