@mokuyoaxis/dsh-iris 0.2.2 → 0.2.4

package/docs/releases/0.2.4.md

@@ -0,0 +1,27 @@
+# Iris 0.2.4：图片续作、统一分配与新版宿主兼容
+
+0.2.4让已有图片可以直接进入作品库并继续编辑，统一六类功能的模型选择规则，同时支持DSH rc.2与alpha.4。
+
+## 升级前检查功能分配
+
+图片、视频、TTS、Music、转写和视觉只使用各自明确分配的有序名单。名单为空时停止新调用，临时指定的模型也必须已分配，且只调用该项。以前依赖自动补入模型池的功能，需要在升级后明确分配模型。
+
+供应商、模型池和已有名单保留，不自动迁移个人配置。旧任务仍绑定原模型查询和补交付；手动模型实测独立于生产名单。DSH默认视觉改为需要明确加入名单的“DSH 当前默认模型”，独立CLI不能调用该宿主资源。见 [完整升级说明](../FUNCTION_MODEL_ASSIGNMENTS.md)。
+
+## 图片导入与原生Qwen改图
+
+工作台“导入图片”、`iris_import_image` 和CLI `artifact import` 保存静态PNG/JPEG/WebP图片，最多20 MiB，返回持久作品ID。保留原字节和源文件，不调用模型；已有Core作品无需重复导入。见 [图片导入](../IMAGE_IMPORT.md)。
+
+`iris_edit_image`、CLI和工作台改图支持DashScope原生Qwen edit/max/plus、2.0/pro系列及日期快照，继续支持聊天生图协议。模型还须具备image-gen能力并加入生图名单，具体账号与地域是否可用由服务端决定。`qwen-image-2.0-pro` 已通过真实CLI带图生成、取回与保存来源关系的实测。
+
+原图只读，发送副本按账号/模型预算处理；修改版保存为独立作品并关联来源。原生Qwen输入副本另受10 MiB上限约束。编辑精度、输出尺寸由模型决定，不提供蒙版、多参考图或像素级保真保证。见 [Artifact改图](../CLI_MANAGEMENT.md#artifact-聊天改图)。
+
+## DSH双版本兼容
+
+支持准确版本 `0.2.0-rc.2` 和 `0.2.1-alpha.4`，manifest、可选客户端peer及Doctor同步。不要求已有rc.2用户升级宿主，也不据此支持整个0.2.x。Iris最低Node.js为22.0.0；DSH若要求更高，以宿主为准。见 [兼容矩阵与验证范围](../DSH_COMPATIBILITY.md)。
+
+## 使用与验证范围
+
+Agent工具增至17项，三项Skills的共享说明包含图片导入、严格分配和Qwen改图。Music仍是可选功能，本版未更换音乐来源或增加新的音乐协议。
+
+宿主隔离验收覆盖实际官方SDK服务、工具/附件交付、HTTP与界面插槽；React/DOM和部分会话数据使用fixture，完整浏览器/移动端交互与热更新仍需实际使用验证。未增加Core数据格式迁移或生产依赖。

package/lib/actions.js

@@ -23,7 +23,7 @@ import { registerMedia, mediaLinksOf, webBase } from './media.js';
 import { RED_TEST_IMAGE } from './vision.js';
 import { runVisionRequest, readVisionImageFile, probeVisionModel, visionImageFromDataUrl } from './vision-model-routing.js';
 import { HOST_CONTRACT_VERSION, HOST_PORTS, defineHostAdapter, hasHostPort, requireHostPort } from './host-contract.js';
-import { coreSnapshotForDsh, cropForDsh, diffForDsh, framesForDsh, summaryFramesForDsh, visionImageForDsh, htmlForDsh, inspectProviderTaskForDsh, submitProviderTaskForDsh, watchProviderTaskForDsh } from './dsh-core-adapter.js';
+import { coreSnapshotForDsh, importImageForDsh, cropForDsh, diffForDsh, framesForDsh, summaryFramesForDsh, visionImageForDsh, htmlForDsh, inspectProviderTaskForDsh, submitMusicForDsh, submitProviderTaskForDsh, watchProviderTaskForDsh } from './dsh-core-adapter.js';
 import { runLocateRequest, runSummaryRequest } from './composite-vision-routing.js';
 import { normalizeSummarySource, summaryFrameRecords, summaryTimelineLabel } from './summary-input.js';
 import { normalizeVisionImageSource } from './vision-input.js';
@@ -31,8 +31,8 @@ import { formatOcrResult } from './ocr.js';
 import { runOcrRequest } from './ocr-model-routing.js';
 import { ffmpegAvailable, extractFrames, probeVideo } from './media-probe.js';
 import { ensurePrivateDir, writePrivateFile } from './private-storage.js';
-import { selectMediaProtocol, selectImageProtocol, providerForImageModel } from './provider-protocol.js';
-import { providerTaskBinding } from './provider-catalog.js';
+import { selectMediaProtocol, selectImageProtocol, providerForImageModel, supportsImageEdit } from './provider-protocol.js';
+import { musicCandidatesFromCatalog, providerTaskBinding } from './provider-catalog.js';
 import { invokeProviderOperation, providerPollTaskView, prepareProviderInput } from './provider-adapter.js';
 import { createConfiguredProviderAdapter as createProviderAdapter } from './provider-adapters.js';
 import { deriveUserState } from './task-semantics.js';
@@ -49,6 +49,11 @@ function createConfiguredProviderAdapter(provider, options = {}) {
     onRateLimit: options.allowRateLimited ? undefined : store.recordRateLimit });
 }
 
+function imageRoutingForModel(provider, modelId) {
+  const routing = selectImageProtocol(provider, modelId);
+  return { ...routing, editSupported: supportsImageEdit(routing.mediaProtocol, modelId) };
+}
+
 const EMPTY_ACTION_HOST = defineHostAdapter({
   id: 'iris-hostless',
   ports: {},
@@ -90,7 +95,7 @@ function selectProviders(capability, label, requestedModelRef) {
   let providers = store.pickAllFor(capability);
   const requested = String(requestedModelRef || '').trim();
   if (requested) {
-    const pool = models.modelPool(store.providers());
+    const pool = models.orderedModels(models.modelPool(store.providers()), capability, store.assignments()[capability], { includeRateLimited: true });
     const selected = models.resolvePoolModel(pool, requested, capability);
     if (selected?.rateLimited) throw Object.assign(new Error(store.modelHealth(selected.providerId, selected.id, capability).retryAt
       ? 'iris: 指定模型正在限流冷却，请稍后使用或手动实测'
@@ -99,7 +104,7 @@ function selectProviders(capability, label, requestedModelRef) {
     });
     const field = models.CAP_FIELD[capability];
     providers = selected ? providers.filter((p) => p.id === selected.providerId && p[field] === selected.id) : [];
-    if (!providers.length) throw new Error('iris: 指定的模型复合引用不可用或不具备' + label + '能力');
+    if (!providers.length) throw new Error('iris: 指定模型未分配给该功能、不可用或不具备' + label + '能力');
   }
   if (!providers.length) throw new Error('iris: 没有可用' + label + '供应商——先在 Iris 设置里添加、标注并分配模型');
   return providers;
@@ -462,6 +467,15 @@ export async function runAction(host, name, args, { signal } = {}) {
 
 /* ============ 动作实现 ============ */
 
+register('image_import', async (_host, args, { signal }) => {
+  if (args.image_path !== undefined && args.image_bytes !== undefined) throw new Error('图片导入只能选择一个来源');
+  const { artifact } = await importImageForDsh({ imagePath: args.image_path, imageBytes: args.image_bytes, signal });
+  const imageUrl = '/iris/api/core/artifact/' + artifact.id + '/media';
+  const text = `已导入图片（${artifact.metadata.width}×${artifact.metadata.height}），原文件保留。\n作品ID：${artifact.id}`;
+  return { ok: true, storage: 'core', artifact, artifactId: artifact.id, artifactIds: [artifact.id], imageUrl, text,
+    blocks: [{ type: 'text', text: text + `\n图片：${imageUrl}` }] };
+});
+
 /* ---------- 🖼️ 画图 ---------- */
 async function imageAction(host, args, { signal }, editing = false) {
   if (editing && !args.source_artifact_id) throw new Error('iris: 改图必须提供 source_artifact_id');
@@ -564,6 +578,33 @@ async function submitCoreTts(routes, normalizedInput, signal) {
 }
 
 /* ---------- 🔊 语音合成 ---------- */
+register('music', async (host, args, { signal }) => {
+  const normalized = normalizeGenerationInput('music', {
+    prompt: args.prompt,
+    ...(optionalActionText(args.timeout_ms) !== undefined ? { timeout_ms: args.timeout_ms } : {}),
+    ...(optionalActionText(args.retry_502) !== undefined ? { retry_502: args.retry_502 } : {}),
+    ...(optionalActionText(args.retry_delay_ms) !== undefined ? { retry_delay_ms: args.retry_delay_ms } : {})
+  });
+  const routes = musicCandidatesFromCatalog({ providers: store.allProviders(), assignments: store.assignments() }, args.model);
+  const submitted = await submitMusicForDsh({
+    candidates: routes.map(route => ({ adapter: createConfiguredProviderAdapter(route.provider), model: route.modelRef,
+      selectionReason: route.selectionReason, providerBinding: providerTaskBinding(route.provider) })),
+    providerInput: normalized.providerInput, musicOptions: normalized.musicOptions, signal
+  });
+  const task = submitted.task;
+  recordCoreTaskAttemptHealth(task, cap.CAPABILITIES.MUSIC);
+  if (submitted.localError) throw taskActionError(task.id, 'Music 结果未能可靠落盘', submitted.localError);
+  if (task.outcome !== 'succeeded' || task.deliveryState !== 'ready') {
+    throw taskActionError(task.id, task.outcome === 'succeeded'
+      ? '音乐已返回，但交付失败；可从缓存重新交付'
+      : `音乐未取得产物（已重试${submitted.retries}次）`, task.lastError);
+  }
+  return { ok: true, storage: 'core', taskId: task.id, taskIds: submitted.taskIds, retries: submitted.retries,
+    model: task.modelRef, providerId: task.providerId, artifactIds: task.artifactIds,
+    audioUrl: '/iris/api/core/artifact/' + encodeURIComponent(task.artifactIds[0]) + '/media',
+    text: `[iris] 音乐已生成（${task.modelRef}，重试${submitted.retries}次，Core task: ${task.id}）` };
+});
+
 register('tts', async (host, args, { signal }) => {
   const { providerInput } = normalizeGenerationInput('tts', {
     text: args.text, voice: optionalActionText(args.voice)
@@ -1008,9 +1049,10 @@ register('providers_list', async (host, args) => {
         id: m.id,
         capabilities: m.capabilities,
         ...(m.imageProtocol !== undefined ? { imageProtocol: m.imageProtocol } : {}),
+        ...(m.musicProtocol !== undefined ? { musicProtocol: m.musicProtocol } : {}),
         ...(m.visionInput !== undefined ? { visionInput: m.visionInput } : {}),
         ...(m.capabilities.includes(cap.CAPABILITIES.VISION) ? { visionInputEffective: visionInputLimits(p, m.id) } : {}),
-        ...(m.capabilities.includes(cap.CAPABILITIES.IMAGE) ? { imageRouting: selectImageProtocol(p, m.id) } : {}),
+        ...(m.capabilities.includes(cap.CAPABILITIES.IMAGE) ? { imageRouting: imageRoutingForModel(p, m.id) } : {}),
         verified: m.verified || null,
         health: Object.fromEntries(m.capabilities.map((capability) => [
           capability, store.modelHealth(p.id, m.id, capability)
@@ -1075,11 +1117,11 @@ register('providers_discover', async (host, args, { signal }) => {
 register('providers_test_vision', async (host, args, { signal }) => {
   const p = store.providerById(String(args.id || ''));
   if (!p) throw new Error('iris: 供应商不存在');
-  const visionProviders = store.pickAllFor(cap.CAPABILITIES.VISION, { includeRateLimited: true }).filter((x) => x.id === p.id);
-  if (!visionProviders.length) return { ok: true, tested: false, text: '该供应商未配置 vision 模型，跳过' };
-  const model = configuredModel(visionProviders[0], cap.CAPABILITIES.VISION);
+  const vision = models.providerModels(p).find(m => m.capabilities.includes(cap.CAPABILITIES.VISION));
+  if (!vision) return { ok: true, tested: false, text: '该供应商未配置 vision 模型，跳过' };
+  const model = vision.id;
   if (args.confirm_paid !== true) throw new Error('iris: 视觉实测会发起真实供应商请求，请先明确确认');
-  const r = await probeVisionModel(visionProviders[0], model, visionImageFromDataUrl(RED_TEST_IMAGE), { timeoutMs: 15000, signal });
+  const r = await probeVisionModel(p, model, visionImageFromDataUrl(RED_TEST_IMAGE), { timeoutMs: 15000, signal });
   store.setModelVerified(p.id, model, cap.CAPABILITIES.VISION, {
     ok: r.ok,
     note: r.ok ? '认出红色测试图' : (r.error || '未识别出红色'),
@@ -1100,6 +1142,9 @@ register('providers_test_vision', async (host, args, { signal }) => {
 
 /** 按能力实测单个模型；返回 {ok, note}。真实调用仅在 UI 明确确认后执行。 */
 async function probeModel(host, p, modelId, capability, signal) {
+  if (capability === cap.CAPABILITIES.MUSIC) {
+    return { skipped: true, ok: false, note: 'Music 不发起小样本探测；请通过音乐功能提交实际创作要求' };
+  }
   if (capability === cap.CAPABILITIES.VISION) {
     const r = await probeVisionModel(p, modelId, visionImageFromDataUrl(RED_TEST_IMAGE), { timeoutMs: 15000, signal });
     return {
@@ -1207,6 +1252,12 @@ register('providers_set_model_image_protocol', async (host, args) => {
   return { ok: true, text: '[iris] 模型图片协议已更新' };
 });
 
+register('providers_set_model_music_protocol', async (host, args) => {
+  const ok = store.setModelMusicProtocol(String(args.id || ''), String(args.model_id || ''), args.musicProtocol);
+  if (!ok) throw new Error('iris: 模型不在池中');
+  return { ok: true, text: '[iris] 模型音乐协议已更新' };
+});
+
 register('providers_set_model_vision_input', async (host, args) => {
   if (!Object.hasOwn(args, 'visionInput')) throw new Error('iris: 需要 visionInput 对象或 null');
   const ok = store.setModelVisionInput(String(args.id || ''), String(args.model_id || ''), args.visionInput);
@@ -1221,7 +1272,7 @@ register('providers_test_model', async (host, args, { signal }) => {
   const modelId = String(args.model_id || '').trim();
   const capability = String(args.capability || '').trim();
   if (!modelId || !cap.isKnownCapability(capability)) throw new Error('iris: 需 model_id 与合法 capability');
-  if (![cap.CAPABILITIES.VIDEO, cap.CAPABILITIES.TRANSCRIBE].includes(capability) && args.confirm_paid !== true) {
+  if (![cap.CAPABILITIES.VIDEO, cap.CAPABILITIES.TRANSCRIBE, cap.CAPABILITIES.MUSIC].includes(capability) && args.confirm_paid !== true) {
     throw new Error('iris: 模型实测会发起真实供应商请求，请先明确确认');
   }
   let result;
@@ -1250,17 +1301,23 @@ register('providers_test_model', async (host, args, { signal }) => {
 
 /** 查询：全部分配 + 每个能力可用的模型池（client 渲染下拉/排序） */
 register('assignments_get', async (host, args) => {
-  const pool = models.modelPool(store.allProviders());
+  const providers = store.allProviders();
+  const configured = new Set(store.providers().map(provider => provider.id));
+  const pool = models.modelPool(providers, { includeHostVision: true });
   const byCap = {};
   for (const m of pool) {
     for (const c of m.capabilities) {
       if (!byCap[c]) byCap[c] = [];
-      byCap[c].push({ ref: m.ref, id: m.id, providerId: m.providerId, label: m.id + ' · ' + m.providerId });
+      const provider = providers.find(provider => provider.id === m.providerId);
+      byCap[c].push({ ref: m.ref, id: m.id, providerId: m.providerId,
+        label: m.ref === models.HOST_VISION_REF ? 'DSH 当前默认模型' : m.id + ' · ' + m.providerId,
+        unavailable: m.ref !== models.HOST_VISION_REF && (!configured.has(m.providerId) || m.rateLimited === true),
+        ...(c === cap.CAPABILITIES.IMAGE && provider ? { imageRouting: imageRoutingForModel(provider, m.id) } : {}) });
     }
   }
   // order：归一化的有序分配列表（GUI 直接渲染 failover 顺序，无需处理单字符串旧格式）
   const order = {};
-  for (const c of Object.keys(byCap)) order[c] = store.assignmentOrder(c);
+  for (const c of Object.values(cap.CAPABILITIES)) order[c] = store.assignmentOrder(c);
   return { ok: true, assignments: store.assignments(), order, poolByCapability: byCap };
 });
 

package/lib/adapters.js

@@ -15,6 +15,7 @@ import { dashscopeApiBase, imageSizeViolation, isDashScopeBaseUrl, normalizeImag
 import { chmodPrivateFile, ensurePrivateDir, privateSibling } from './private-storage.js';
 import { isFreeQuotaExhausted, retryAfterDelay } from './provider-health.js';
 import { ProviderContractError } from './provider-contract.js';
+export { generateMusic } from './music-transport.js';
 
 /**
  * 网络调用超时档位（2026-09-03 健康检查）：挂死的连接绝不能冻住盯守 tick——
@@ -188,7 +189,7 @@ export function applyImageSize({ protocol, mode, size }) {
 }
 
 /** 新版多模态同步协议：响应中直接返回图片 URL。 */
-export async function generateImageMultimodal({ key, baseUrl, model, prompt, size, n = 1, timeoutMs = T_SYNC_GEN, signal }) {
+export async function generateImageMultimodal({ key, baseUrl, model, prompt, image, size, n = 1, timeoutMs = T_SYNC_GEN, signal }) {
   const apiBase = dashscopeApiBase(baseUrl);
   const requestSize = applyImageSize({ protocol: 'dashscope', mode: 'multimodal-sync', size });
   const { res, body } = await fetchImageJson(`${apiBase}/services/aigc/multimodal-generation/generation`, {
@@ -197,7 +198,10 @@ export async function generateImageMultimodal({ key, baseUrl, model, prompt, siz
     signal: timeoutSignal(timeoutMs, signal),
     body: JSON.stringify({
       model,
-      input: { messages: [{ role: 'user', content: [{ text: prompt }] }] },
+      input: { messages: [{ role: 'user', content: [
+        ...(image ? [{ image: 'data:' + image.mediaType + ';base64,' + Buffer.from(image.bytes).toString('base64') }] : []),
+        { text: prompt }
+      ] }] },
       parameters: { ...(requestSize ? { size: requestSize } : {}), n: Number(n) || 1 }
     })
   });

package/lib/api.js

@@ -33,9 +33,12 @@ import {
   importPromptOptimizerConfig,
   loadPromptOptimizerConfig,
   resetPromptOptimizerConfig,
+  resetPromptOptimizerInstructions,
   setPromptOptimizerEnabled
 } from './prompt-optimizer-config.js';
 
+const PROMPT_CONFIG_CAPABILITIES = Object.freeze({ ...PROMPT_OPTIMIZER_CAPABILITIES, resetInstructions: true });
+
 function readJsonSafe(file) {
   try {
     const raw = fs.readFileSync(file, 'utf8');
@@ -369,7 +372,7 @@ export function serveApi(req, res, host) {
       return serveSse(res);
     }
     if (urlPath === '/iris/api/prompt-optimizer/config') {
-      return sendJson(res, 200, method === 'HEAD' ? {} : { ...loadPromptOptimizerConfig(), capabilities: PROMPT_OPTIMIZER_CAPABILITIES });
+      return sendJson(res, 200, method === 'HEAD' ? {} : { ...loadPromptOptimizerConfig(), capabilities: PROMPT_CONFIG_CAPABILITIES });
     }
     if (urlPath === '/iris/api/doctor') {
       if (!host || typeof host !== 'object') return sendJson(res, 503, { error: 'host unavailable' });
@@ -506,7 +509,9 @@ function handleJsonPost(req, res, execute, respond = sendJson) {
     if (overflow) return finish(413, { error: 'request body too large (limit 1MB)' });
     try {
       const raw = Buffer.concat(chunks).toString('utf8');
-      const body = raw ? JSON.parse(raw) : {};
+      let body;
+      try { body = raw ? JSON.parse(raw) : {}; }
+      catch (_) { return finish(controller.signal.aborted ? 499 : 400, { error: controller.signal.aborted ? 'request canceled' : 'invalid JSON' }); }
       const result = await execute(body, controller.signal);
       if (controller.signal.aborted) return finish(499, { error: 'request canceled' });
       finish(200, result);
@@ -630,10 +635,14 @@ function handlePromptOptimizer(host, operation, req, res) {
     if (operation === 'optimize') return optimizePrompt(host, body, { signal });
     if (operation === 'import') {
       const imported = body && body.config !== undefined ? body.config : body;
-      return { ok: true, ...importPromptOptimizerConfig(imported), capabilities: PROMPT_OPTIMIZER_CAPABILITIES };
+      return { ok: true, ...importPromptOptimizerConfig(imported), capabilities: PROMPT_CONFIG_CAPABILITIES };
+    }
+    if (operation === 'reset') {
+      const scope = body?.scope === undefined ? 'all' : body.scope;
+      if (!['all', 'instructions'].includes(scope)) throw new Error('scope 只能是 all 或 instructions');
+      return { ok: true, ...(scope === 'instructions' ? resetPromptOptimizerInstructions() : resetPromptOptimizerConfig()), capabilities: PROMPT_CONFIG_CAPABILITIES };
     }
-    if (operation === 'reset') return { ok: true, ...resetPromptOptimizerConfig(), capabilities: PROMPT_OPTIMIZER_CAPABILITIES };
-    if (operation === 'enabled') return { ok: true, ...setPromptOptimizerEnabled(body && body.enabled), capabilities: PROMPT_OPTIMIZER_CAPABILITIES };
+    if (operation === 'enabled') return { ok: true, ...setPromptOptimizerEnabled(body && body.enabled), capabilities: PROMPT_CONFIG_CAPABILITIES };
     throw new Error('unknown prompt optimizer operation');
   });
 }

package/lib/bundled-skills.js

@@ -5,7 +5,7 @@ import path from 'node:path';
 import { fileURLToPath } from 'node:url';
 import { requireHostPort } from './host-contract.js';
 
-const SKILL_NAMES = ['iris-verify-ui', 'iris-compose-media'];
+const SKILL_NAMES = ['iris-use', 'iris-compose-media', 'iris-verify-ui'];
 const SKILL_ROOT = fileURLToPath(new URL('../.dsh/skills/', import.meta.url));
 
 function parseScalar(value, file) {

package/lib/capability.js

@@ -14,6 +14,7 @@ export const CAPABILITIES = Object.freeze({
   IMAGE: 'image-gen',
   VIDEO: 'video-gen',
   TTS: 'tts',
+  MUSIC: 'music-gen',
   TRANSCRIBE: 'transcribe',
   VISION: 'vision'
 });
@@ -26,6 +27,7 @@ const INFER_FIELD = {
   [CAPABILITIES.IMAGE]: 'imageModel',
   [CAPABILITIES.VIDEO]: 'videoModel',
   [CAPABILITIES.TTS]: 'ttsModel',
+  [CAPABILITIES.MUSIC]: 'musicModel',
   [CAPABILITIES.TRANSCRIBE]: 'transcribeModel',
   [CAPABILITIES.VISION]: 'visionModel'
 };