Formerly KrillinAI, VideoTranslation and Dubbing Tool for 120+ Languages. Open-source AI workspace for creators, powered by Codex. Video downloader, video translation, create videos, images, audios, and edits with Agents in one place. For your Social Media like YouTube, TikTok, Instagram, etc.
TypeScript agent-skillscodex-clideskop-appdubbingimage-generationlocalizationskillsttsvideo-editingvideo-generationvideo-transcriptionvideo-translation
fix: improve creator model and runtime component settings
Leggettbda3dc1d2f40
diff --git a/.codex/skills/oc-deploy/SKILL.md b/.codex/skills/oc-deploy/SKILL.mdnew file mode 100644index 00000000..86c8b6cd--- /dev/null+++ b/.codex/skills/oc-deploy/SKILL.md@@ -0,0 +1,32 @@+---+name: oc-deploy+description: 验证、诊断和发布 OpenCreator Desktop,包括本地实际 App 打包与 E2E、master 候选构建、GitHub Actions 故障定位及明确授权后的 tag Release。用于“打包 Desktop”“验证候选包”“排查发布失败”“发布 OpenCreator”等请求;普通开发构建不使用。+---++# OpenCreator Deploy++以“同一提交先在本机证明,再消耗远端构建资源”为发布原则。遵守仓库根目录 `AGENTS.md`,尤其是 Desktop 打包和 Web/Desktop 一致性门禁。++## 先选择模式++- **diagnose**:用户问为什么失败时,只读取现有本地输出和已存在的 GitHub 日志。先给出失败阶段、原始错误和根因;未经要求不得触发新构建。+- **preflight**:在当前提交上执行本地候选包验证,不产生 GitHub 外部变更。+- **candidate**:用户明确要求远端验证时,从 `master` 对精确提交手动触发一次 `desktop-release.yml`。不得创建、移动或复用 tag,不得发布 Release。+- **release**:仅当用户明确要求正式发布或创建 tag 时使用。preflight 和精确提交的 master CI 未通过前不得创建或推送 tag。++执行任何模式前先读取 [references/runbook.md](references/runbook.md) 中对应章节。涉及失败时同时读取“故障分类”。++## 强制门禁++1. 先记录分支、HEAD、dirty 状态、远端同步状态和目标平台。不得把用户已有改动带入提交或发布。+2. 类型检查、单元测试和源码构建不能替代实际 App 验证。远端 candidate 或 release 前,当前精确提交必须完成本机原生 `desktop:package`,并让 `e2e:package` 全部通过。+3. Desktop 打包必须重建当前 `apps/web`,并通过包校验器确认 App 内 Web 资源与本次构建一致。不得手工替换 App 内资源或复用来源不明的 `dist`。+4. 任何 GitHub 构建前,确认精确提交已有成功的 `master` push CI;确认 workflow 的 `publish` 仍只允许 tag ref。+5. GitHub 下载和 `gh` 命令设置 `HTTP_PROXY` 与 `HTTPS_PROXY` 为 `http://127.0.0.1:7897`,并先用 `gh auth status` 确认认证身份。GitHub 保存的签名 secrets 会跨 workflow 运行持续存在,不要要求重复上传。+6. 不用付费 Release 构建逐次发现问题。一次远端运行首次出现 package、E2E、签名、公证或上传失败时,立即取消仍在运行的付费 sibling jobs,读取该 run 的已生成日志和 artifacts;不得为了拿日志直接 rerun。+7. candidate 只能通过 `workflow_dispatch` 从 `master` 运行。结束后确认 `publish` skipped、artifacts 已生成,并确认 tag 和 Release 均未变化。+8. release 只能对已经通过上述门禁的精确 commit 创建 tag。不得在失败后盲目移动 tag;先本地修复、重新通过门禁和 master CI,再按用户明确授权处理 tag。++## 完成标准++报告精确 commit、本地打包平台、本地 App 路径、包校验与 E2E 结果、master CI、GitHub run URL、各 job 结果及 artifacts。candidate 明确说明未创建 tag/Release;release 列出 tag 和发布附件。任何未执行门禁都必须说明原因和残余风险,不得声称可发布。diff --git a/.codex/skills/oc-deploy/agents/openai.yaml b/.codex/skills/oc-deploy/agents/openai.yamlnew file mode 100644index 00000000..4970627b--- /dev/null+++ b/.codex/skills/oc-deploy/agents/openai.yaml@@ -0,0 +1,4 @@+interface:+ display_name: "OpenCreator Deploy"+ short_description: "本地验证、候选构建与正式发布 OpenCreator"+ default_prompt: "使用 $oc-deploy 完成本地打包验证,并按授权执行候选构建或正式发布。"diff --git a/.codex/skills/oc-deploy/references/runbook.md b/.codex/skills/oc-deploy/references/runbook.mdnew file mode 100644index 00000000..25dd1435--- /dev/null+++ b/.codex/skills/oc-deploy/references/runbook.md@@ -0,0 +1,88 @@+# OpenCreator Desktop 发布运行手册++## 环境与基线++在仓库根目录先检查 `git status --short --branch`、`git rev-parse HEAD`、`origin/master` 和 `gh auth status --hostname github.com`。所有 `git fetch`、GitHub 下载和 `gh` 命令设置:++```bash+HTTP_PROXY=http://127.0.0.1:7897+HTTPS_PROXY=http://127.0.0.1:7897+```++检查 Node、pnpm 和 `runtime/krillinai/go.mod` 要求的 Go 版本。本机可能使用 `.tools/go-1.22.12/bin/go`,但不要假定它永远存在。不要提交 `.tools/` 或 `.codex/`。++## Diagnose:只读诊断++先定位现有 run,不启动新 run:++```bash+gh run list --repo krillinai/OpenCreator --workflow desktop-release.yml --limit 10+gh run view <run-id> --repo krillinai/OpenCreator --json url,headSha,event,status,conclusion,jobs+gh run view <run-id> --repo krillinai/OpenCreator --log-failed+```++判断失败属于编译、测试、资源准备、签名、公证、包校验、实际 App E2E、artifact 上传还是 publish。引用首个决定性错误,不把后续连锁错误当根因。日志不完整时下载该 run 已有 diagnostics artifact,不用 rerun 获取诊断。++## Preflight:本地候选包++先按改动影响运行定向测试。发布门禁至少包括当前机器原生架构的实际 App 打包和完整 packaged App E2E。macOS arm64 示例:++```bash+PATH="$PWD/.tools/go-1.22.12/bin:/usr/local/bin:/usr/bin:/bin:/usr/sbin:/sbin" \+HTTP_PROXY=http://127.0.0.1:7897 HTTPS_PROXY=http://127.0.0.1:7897 \+OPENCREATOR_DESKTOP_TARGET_PLATFORM=darwin OPENCREATOR_DESKTOP_TARGET_ARCH=arm64 \+pnpm desktop:package++PATH="$PWD/.tools/go-1.22.12/bin:/usr/local/bin:/usr/bin:/bin:/usr/sbin:/sbin" \+HTTP_PROXY=http://127.0.0.1:7897 HTTPS_PROXY=http://127.0.0.1:7897 \+pnpm --filter @opencreator/desktop e2e:package+```++记录实际 `.app` 路径和测试总数。当前完整套件预期 16 项;以仓库当时清单为准,但 skipped/failed 必须说明。发布实现有变化时补充:++```bash+pnpm --filter @opencreator/web test -- src/features/settings/CreatorServicesSettingsView.test.tsx+pnpm --filter @opencreator/desktop test -- test/stickman-runtime-package.test.mjs test/creator-runtime-signing.test.mjs test/mac-signing.test.mjs+./node_modules/.bin/tsc -p apps/desktop/tsconfig.test.json --noEmit+```++preflight 失败就在本地修复,不触发 GitHub package workflow。++## Candidate:master 远端候选构建++仅在用户明确要求且 preflight 已通过时执行:++1. 确认目标 commit 已推送到 `master`,且该 commit 的 `ci.yml` push run 成功。+2. 检查 `.github/workflows/desktop-release.yml`,确认 `publish.if` 仍只匹配 tag ref。+3. 记录现有 tags 和 Releases,供运行后比对。+4. 只触发一次:++```bash+HTTP_PROXY=http://127.0.0.1:7897 HTTPS_PROXY=http://127.0.0.1:7897 \+gh workflow run desktop-release.yml --repo krillinai/OpenCreator --ref master -f target=all+```++5. 任一 package job 首次失败,立即 `gh run cancel <run-id>`,再读取失败日志;不要自动重跑。+6. 成功标准:verify、目标 package jobs 和 Linux package 成功;用户附件与 KrillinAI 附件存在;publish skipped;tag 和 Release 未变化。++只验证某个已知平台问题时,可在用户同意后选单平台 target 以节省费用;完整跨平台候选验收使用 `all`。++## Release:正式 tag 发布++release 是独立授权边界。用户必须明确指定版本并要求发布:++1. 确认版本、目标 commit、工作区、`origin/master`、同 commit 本地 preflight 和 master push CI。+2. 确认 tag 不存在;已存在时由用户明确决定,不自行覆盖。+3. 创建并推送一次 tag,监控由 tag 触发的唯一 Release run。+4. 首次失败立即取消其他付费 jobs并诊断,不 rerun,不先移动 tag。+5. 成功后验证 Release 非 draft、附件完整、命名和下载说明正确。++## 故障分类与已知陷阱++- **依赖**:Daemon 精确依赖声明必须与 lockfile 解析结果一致。例如 `toml` 版本不一致会确定性失败,无需远端重跑确认。+- **macOS 签名**:`Resources/daemon`、`Resources/creator-runtime`、`Resources/stickman-runtime` 内的 dylib、ffmpeg、Chromium 等 Mach-O 必须递归签名。GitHub `release` Environment secrets 会持续保存,不需每次上传。+- **包校验**:签名会改变 Mach-O 字节。不能把签名后 Chromium 哈希直接与未签名源文件比较;应保留文件列表、版本、manifest、来源绑定检查,并验证所有 Mach-O 的 Developer ID 签名。+- **Stickman Chromium**:它是 Stickman 自动化运行时携带的浏览器,体积大且包含需要签名的原生文件,不是陌生外部应用。+- **Electron 竞态**:首次并发解压可能报 `icudtl.dat: File exists` 或安装损坏。CI 构建前用 `pnpm --filter @opencreator/desktop exec node -e "require('electron')"` 预热;Linux 不执行 `electron --version`,避免 SUID sandbox 错误。+- **E2E 平台能力**:测试从 `/creator-services/capabilities` 获取事实,不复制平台规则。已知矩阵是 macOS arm64 `WhisperKit / large-v2`、macOS x64 无本地 provider 且入口禁用、Windows x64 `Whisper.cpp / tiny`。+- **体积**:macOS arm64 App 曾约 1.35 GB,主要来自 Electron、Creator/KrillinAI 依赖与模型、Remotion FFmpeg、Stickman Chromium、libvips。优化前按 bundle 子目录统计并查重复文件,再评估按需下载、去重、压缩或拆包;不能破坏离线契约、签名、公证、manifest 和 packaged App E2E。diff --git a/apps/web/src/features/settings/CreatorServicesSettingsView.test.tsx b/apps/web/src/features/settings/CreatorServicesSettingsView.test.tsxindex 7bc565ff..cdf57c31 100644--- a/apps/web/src/features/settings/CreatorServicesSettingsView.test.tsx+++ b/apps/web/src/features/settings/CreatorServicesSettingsView.test.tsx@@ -1,5 +1,6 @@import {createDefaultCreatorServicesConfig,+ type CodexModelListResponse,type CreatorServicesCapabilitiesResponse,type CreatorServicesCredentialField,type CreatorTtsProvider@@ -69,6 +70,86 @@ describe('CreatorServicesSettingsView', () => {expect(screen.queryByDisplayValue(/secret/i)).not.toBeInTheDocument();});+ it('applies an LLM provider preset and model suggestions', async () => {+ const user = userEvent.setup();+ const modelService = createModelService();+ render(+ <CreatorServicesSettingsView+ connected+ service={createService()}+ modelService={modelService}+ />+ );++ await screen.findByRole('heading', { name: 'AI 服务' });+ await user.selectOptions(screen.getByLabelText('供应商'), 'deepseek');+ expect(screen.getByLabelText('Base URL')).toHaveValue('https://api.deepseek.com');+ expect(screen.getByLabelText('模型')).toHaveValue('deepseek-v4-pro');+ expect(screen.getByRole('option', { name: 'deepseek-v4-flash' })).toBeInTheDocument();+ expect(screen.queryByRole('option', { name: 'MiniMax-M2.7' })).not.toBeInTheDocument();++ await user.click(screen.getByRole('button', { name: '保存配置' }));+ await waitFor(() => expect(modelService.updateCodexProvider).toHaveBeenCalledWith({+ baseUrl: 'https://api.deepseek.com',+ model: 'deepseek-v4-pro'+ }));+ });++ it('defaults OpenAI to GPT-5.6 and keeps it in the Runtime model list', async () => {+ const user = userEvent.setup();+ render(+ <CreatorServicesSettingsView+ connected+ service={createService()}+ modelService={createModelService({ runtimeModels: ['gpt-5.6-sol', 'gpt-5.5'] })}+ />+ );++ await screen.findByRole('heading', { name: 'AI 服务' });+ await user.selectOptions(screen.getByLabelText('供应商'), 'openai');++ expect(screen.getByLabelText('Base URL')).toHaveValue('https://api.openai.com/v1');+ expect(screen.getByLabelText('模型')).toHaveValue('gpt-5.6-sol');+ expect(screen.getByRole('option', { name: 'gpt-5.5' })).toBeInTheDocument();+ });++ it('allows switching from a provider model to a custom model', async () => {+ const user = userEvent.setup();+ render(+ <CreatorServicesSettingsView+ connected+ service={createService()}+ modelService={createModelService()}+ />+ );++ await screen.findByRole('heading', { name: 'AI 服务' });+ await user.selectOptions(screen.getByLabelText('供应商'), 'deepseek');+ await user.selectOptions(screen.getByLabelText('模型'), '__custom__');++ const customModel = screen.getByPlaceholderText('gpt-5.6-sol');+ await user.clear(customModel);+ await user.type(customModel, 'my-custom-model');+ expect(customModel).toHaveValue('my-custom-model');+ });++ it('keeps MiniMax models separate from the OpenAI Runtime catalog', async () => {+ const user = userEvent.setup();+ render(+ <CreatorServicesSettingsView+ connected+ service={createService()}+ modelService={createModelService({ runtimeModels: ['gpt-5.5'] })}+ />+ );++ await screen.findByRole('heading', { name: 'AI 服务' });+ await user.selectOptions(screen.getByLabelText('供应商'), 'minimax');+ expect(screen.getByLabelText('模型')).toHaveValue('MiniMax-M2.7');+ expect(screen.getByRole('option', { name: 'MiniMax-M2.7-highspeed' })).toBeInTheDocument();+ expect(screen.queryByRole('option', { name: 'gpt-5.5' })).not.toBeInTheDocument();+ });+it('shows only the fields required by the selected transcription and voice providers', async () => {const user = userEvent.setup();render(@@ -234,10 +315,7 @@ describe('CreatorServicesSettingsView', () => {await user.click(screen.getByRole('tab', { name: '视频生成' }));expect(screen.getByRole('combobox', { name: '服务商' })).toHaveTextContent('Seedance');expect(screen.getByLabelText('默认模型')).toHaveValue('doubao-seedance-2-5-260628');- expect(screen.getByLabelText('默认模型')).toHaveAttribute(- 'list',- 'video-seedance-model-suggestions'- );+ expect(screen.getByRole('option', { name: 'doubao-seedance-2-5-260628' })).toBeInTheDocument();expect(screen.getByLabelText('API Key')).toHaveAttribute('type', 'password');await user.click(screen.getByRole('combobox', { name: '服务商' }));await user.click(screen.getByRole('option', { name: 'Veo' }));@@ -306,6 +384,7 @@ function createService(function createModelService(options: {authentication?: 'none' | 'chatgpt' | 'api_key';apiKeyConfigured?: boolean;+ runtimeModels?: string[];} = {}) {const provider = {baseUrl: 'https://gateway.example.test/v1',@@ -316,6 +395,18 @@ function createModelService(options: {};return {getCodexProvider: vi.fn(async () => structuredClone(provider)),+ getCodexModels: vi.fn(async (): Promise<CodexModelListResponse> => ({+ models: (options.runtimeModels ?? []).map(model => ({+ id: model,+ model,+ displayName: model,+ description: '',+ supportedReasoningEfforts: [],+ defaultReasoningEffort: null,+ inputModalities: ['text'],+ isDefault: false+ }))+ })),updateCodexProvider: vi.fn(async input => ({baseUrl: input.baseUrl,model: input.model,diff --git a/apps/web/src/features/settings/CreatorServicesSettingsView.tsx b/apps/web/src/features/settings/CreatorServicesSettingsView.tsxindex c17b34b8..2b14eb87 100644--- a/apps/web/src/features/settings/CreatorServicesSettingsView.tsx+++ b/apps/web/src/features/settings/CreatorServicesSettingsView.tsx@@ -1,5 +1,7 @@import {createDefaultCreatorServicesConfig,+ creatorProviderCatalogById,+ creatorProviderOfKind,defaultVideoGenerationModels,videoGenerationModelIds,type CodexProviderConfig,@@ -40,7 +42,9 @@ import type { ConnectionService } from '../../services/connection-service.js';import './creator-services-settings.css';export type CreatorServicesSection = 'text' | 'transcription' | 'tts' | 'image' | 'video';-type ModelSettingsService = Pick<ConnectionService, 'getCodexProvider' | 'updateCodexProvider'>;+type ModelSettingsService = Pick<ConnectionService, 'getCodexProvider' | 'updateCodexProvider'> & {+ getCodexModels?: ConnectionService['getCodexModels'];+};export function CreatorServicesSettingsView(props: {connected: boolean;@@ -56,6 +60,8 @@ export function CreatorServicesSettingsView(props: {const [config, setConfig] = useState<CreatorServicesConfig>();const [capabilities, setCapabilities] = useState<CreatorServicesCapabilitiesResponse>();const [modelProvider, setModelProvider] = useState<CodexProviderConfig>();+ const [modelProviderId, setModelProviderId] = useState('custom');+ const [runtimeModelIds, setRuntimeModelIds] = useState<string[]>([]);const [modelBaseUrl, setModelBaseUrl] = useState('');const [modelName, setModelName] = useState('');const [modelApiKey, setModelApiKey] = useState('');@@ -91,8 +97,10 @@ export function CreatorServicesSettingsView(props: {props.service.getCapabilities(),props.modelService?.getCodexProvider()?? Promise.reject(new Error('Model provider configuration is unavailable'))+ ,props.modelService?.getCodexModels?.()+ ?? Promise.reject(new Error('Model catalog is unavailable'))])- .then(([configResult, capabilitiesResult, providerResult]) => {+ .then(([configResult, capabilitiesResult, providerResult, modelsResult]) => {if (!active) return;if (configResult.status === 'rejected' || capabilitiesResult.status === 'rejected') {setConfig(undefined);@@ -101,6 +109,9 @@ export function CreatorServicesSettingsView(props: {return;}const response = configResult.value;+ setRuntimeModelIds(modelsResult.status === 'fulfilled'+ ? modelsResult.value.models.map(model => model.model)+ : []);setConfig(response.config);setCapabilities(capabilitiesResult.value);setConfiguredCredentials(new Set(response.configuredCredentials));@@ -109,10 +120,12 @@ export function CreatorServicesSettingsView(props: {if (providerResult.status === 'fulfilled') {const provider = providerResult.value;setModelProvider(provider);+ setModelProviderId(inferLlmProviderId(provider.baseUrl, provider.model));setModelBaseUrl(useLegacyTextModel ? response.config.llm.baseUrl : provider.baseUrl);setModelName(useLegacyTextModel ? response.config.llm.model : provider.model);} else {setModelProvider(undefined);+ setModelProviderId('custom');setModelBaseUrl(response.config.llm.baseUrl);setModelName(response.config.llm.model);setModelError(l(@@ -318,6 +331,17 @@ export function CreatorServicesSettingsView(props: {setModelApiKey(value.apiKey);setNotice(undefined);}}+ providerId={modelProviderId}+ runtimeModelIds={runtimeModelIds}+ onProviderIdChange={id => {+ const preset = creatorProviderCatalogById.llm[id];+ if (preset === undefined) return;+ setModelProviderId(id);+ setModelBaseUrl(preset.defaultBaseUrl ?? '');+ setModelName(preset.models[0]?.id ?? '');+ setModelApiKey('');+ setNotice(undefined);+ }}/>) : null}{activeSection === 'transcription' ? (@@ -389,6 +413,9 @@ function TextModelSettings(props: SettingsGroupProps & {apiKey: string;error?: string;onProviderChange(value: OpenAiCompatibleConfig): void;+ providerId: string;+ runtimeModelIds: readonly string[];+ onProviderIdChange(value: string): void;}) {const l = useLocalizedCopy();const legacyApiKeyConfigured = props.config.llm.source === 'custom'@@ -411,6 +438,14 @@ function TextModelSettings(props: SettingsGroupProps & {'OpenCreator Agent and text tasks such as translation and script processing share this OpenAI-compatible configuration.')}>+ <label className="creator-services-field">+ <span>{l('供应商', 'Provider')}</span>+ <select value={props.providerId} onChange={event => props.onProviderIdChange(event.target.value)}>+ {Object.values(creatorProviderCatalogById.llm).map(provider => (+ <option key={provider.id} value={provider.id}>{provider.label}</option>+ ))}+ </select>+ </label><OpenAiFieldsid="llm"credential="llm.apiKey"@@ -420,7 +455,10 @@ function TextModelSettings(props: SettingsGroupProps & {apiKey: props.apiKey,model: props.model}}- modelPlaceholder="gpt-4o-mini"+ modelPlaceholder="gpt-5.6-sol"+ modelSuggestions={props.providerId === 'openai' && props.runtimeModelIds.length > 0+ ? props.runtimeModelIds+ : creatorProviderCatalogById.llm[props.providerId]?.models.map(model => model.id)}onChange={props.onProviderChange}/><div className="creator-services-model-status is-wide" role="status">@@ -470,6 +508,15 @@ function TextModelSettings(props: SettingsGroupProps & {);}+function inferLlmProviderId(baseUrl: string, model: string): string {+ const normalizedUrl = baseUrl.toLowerCase();+ const normalizedModel = model.toLowerCase();+ if (normalizedUrl.includes('deepseek') || normalizedModel.startsWith('deepseek-')) return 'deepseek';+ if (normalizedUrl.includes('minimax') || normalizedModel.startsWith('minimax-')) return 'minimax';+ if (normalizedUrl.includes('openai.com')) return 'openai';+ return 'custom';+}+function TranscriptionSettings(props: SettingsGroupProps & {capabilities: CreatorServicesCapabilitiesResponse;}) {@@ -692,6 +739,12 @@ function TtsSettings(props: SettingsGroupProps) {]}onChange={value => props.update(config => {config.tts.provider = value as CreatorServicesConfig['tts']['provider'];+ const preset = creatorProviderOfKind('tts', value);+ if (preset?.defaultBaseUrl !== undefined && value !== 'edge-tts') {+ const target = config.tts[value as 'openai' | 'minimax' | 'aliyun'];+ target.baseUrl = target.baseUrl || preset.defaultBaseUrl;+ if (!target.model && preset.models[0] !== undefined) target.model = preset.models[0].id;+ }})}/>{provider === 'openai' ? (@@ -701,6 +754,7 @@ function TtsSettings(props: SettingsGroupProps) {configuredCredentials={props.configuredCredentials}value={props.config.tts.openai}modelPlaceholder="gpt-4o-mini-tts"+ modelSuggestions={creatorProviderOfKind('tts', 'openai-tts')?.models.map(model => model.id)}onChange={value => props.update(config => {config.tts.openai = { ...config.tts.openai, ...value };})}@@ -713,6 +767,7 @@ function TtsSettings(props: SettingsGroupProps) {configuredCredentials={props.configuredCredentials}value={props.config.tts.minimax}modelPlaceholder="speech-2.8-hd"+ modelSuggestions={creatorProviderOfKind('tts', 'minimax-tts')?.models.map(model => model.id)}baseUrlPlaceholder="https://api.minimax.io"onChange={value => props.update(config => {config.tts.minimax = { ...config.tts.minimax, ...value };@@ -726,6 +781,7 @@ function TtsSettings(props: SettingsGroupProps) {configuredCredentials={props.configuredCredentials}value={props.config.tts.aliyun}modelPlaceholder="qwen3-tts-flash"+ modelSuggestions={creatorProviderOfKind('tts', 'aliyun-tts')?.models.map(model => model.id)}baseUrlPlaceholder="https://dashscope.aliyuncs.com/api/v1"onChange={value => props.update(config => {config.tts.aliyun = { ...config.tts.aliyun, ...value };@@ -774,13 +830,18 @@ function ImageSettings(props: SettingsGroupProps) {['gemini', 'Gemini']]}onChange={value => props.update(config => {- config.image.provider = value as CreatorServicesConfig['image']['provider'];+ const nextProvider = value as CreatorServicesConfig['image']['provider'];+ config.image.provider = nextProvider;+ const preset = creatorProviderOfKind('image', nextProvider);+ const target = config.image[nextProvider];+ if (preset?.defaultBaseUrl !== undefined && !target.baseUrl) target.baseUrl = preset.defaultBaseUrl;+ if (!target.model && preset?.models[0] !== undefined) target.model = preset.models[0].id;})}/>- {provider === 'openai' ? <OpenAiFields id="image-openai" credential="image.openai.apiKey" configuredCredentials={props.configuredCredentials} value={props.config.image.openai} modelPlaceholder="gpt-image-1" onChange={value => props.update(config => { config.image.openai = value; })} /> : null}- {provider === 'jimeng' ? <OpenAiFields id="image-jimeng" credential="image.jimeng.apiKey" configuredCredentials={props.configuredCredentials} value={props.config.image.jimeng} modelPlaceholder="doubao-seedream-4-0-250828" baseUrlPlaceholder="https://ark.cn-beijing.volces.com/api/v3" onChange={value => props.update(config => { config.image.jimeng = value; })} /> : null}- {provider === 'kling' ? <KlingFields id="image-kling" accessKeyCredential="image.kling.accessKey" secretKeyCredential="image.kling.secretKey" configuredCredentials={props.configuredCredentials} value={props.config.image.kling} modelPlaceholder="kling-v2-1" onChange={value => props.update(config => { config.image.kling = value; })} /> : null}- {provider === 'gemini' ? <OpenAiFields id="image-gemini" credential="image.gemini.apiKey" configuredCredentials={props.configuredCredentials} value={props.config.image.gemini} modelPlaceholder="gemini-2.5-flash-image" baseUrlPlaceholder="https://generativelanguage.googleapis.com/v1beta" onChange={value => props.update(config => { config.image.gemini = value; })} /> : null}+ {provider === 'openai' ? <OpenAiFields id="image-openai" credential="image.openai.apiKey" configuredCredentials={props.configuredCredentials} value={props.config.image.openai} modelPlaceholder="gpt-image-1" modelSuggestions={creatorProviderOfKind('image', 'openai')?.models.map(model => model.id)} onChange={value => props.update(config => { config.image.openai = value; })} /> : null}+ {provider === 'jimeng' ? <OpenAiFields id="image-jimeng" credential="image.jimeng.apiKey" configuredCredentials={props.configuredCredentials} value={props.config.image.jimeng} modelPlaceholder="doubao-seedream-4-0-250828" modelSuggestions={creatorProviderOfKind('image', 'jimeng')?.models.map(model => model.id)} baseUrlPlaceholder="https://ark.cn-beijing.volces.com/api/v3" onChange={value => props.update(config => { config.image.jimeng = value; })} /> : null}+ {provider === 'kling' ? <KlingFields id="image-kling" accessKeyCredential="image.kling.accessKey" secretKeyCredential="image.kling.secretKey" configuredCredentials={props.configuredCredentials} value={props.config.image.kling} modelPlaceholder="kling-v2-1" modelSuggestions={creatorProviderOfKind('image', 'kling')?.models.map(model => model.id)} onChange={value => props.update(config => { config.image.kling = value; })} /> : null}+ {provider === 'gemini' ? <OpenAiFields id="image-gemini" credential="image.gemini.apiKey" configuredCredentials={props.configuredCredentials} value={props.config.image.gemini} modelPlaceholder="gemini-2.5-flash-image" modelSuggestions={creatorProviderOfKind('image', 'gemini')?.models.map(model => model.id)} baseUrlPlaceholder="https://generativelanguage.googleapis.com/v1beta" onChange={value => props.update(config => { config.image.gemini = value; })} /> : null}</SettingsFieldset>);}@@ -806,11 +867,17 @@ function VideoSettings(props: SettingsGroupProps) {['veo', 'Veo']]}onChange={value => props.update(config => {- config.video.provider = value as CreatorServicesConfig['video']['provider'];+ const nextProvider = value as CreatorServicesConfig['video']['provider'];+ config.video.provider = nextProvider;+ const catalogId = nextProvider === 'kling' ? 'kling-video' : nextProvider;+ const preset = creatorProviderOfKind('video', catalogId);+ const target = config.video[nextProvider];+ if (preset?.defaultBaseUrl !== undefined && !target.baseUrl) target.baseUrl = preset.defaultBaseUrl;+ if (!target.model && preset?.models[0] !== undefined) target.model = preset.models[0].id;})}/>{provider === 'seedance' ? <OpenAiFields id="video-seedance" credential="video.seedance.apiKey" configuredCredentials={props.configuredCredentials} value={props.config.video.seedance} modelLabel={l('默认模型', 'Default model')} modelPlaceholder={defaultVideoGenerationModels.seedance} modelSuggestions={videoGenerationModelIds.seedance} baseUrlPlaceholder="https://ark.cn-beijing.volces.com/api/v3" onChange={value => props.update(config => { config.video.seedance = value; })} /> : null}- {provider === 'kling' ? <KlingFields id="video-kling" accessKeyCredential="video.kling.accessKey" secretKeyCredential="video.kling.secretKey" configuredCredentials={props.configuredCredentials} value={props.config.video.kling} modelLabel={l('默认模型', 'Default model')} modelPlaceholder={defaultVideoGenerationModels.kling} modelSuggestions={videoGenerationModelIds.kling} onChange={value => props.update(config => { config.video.kling = value; })} /> : null}+ {provider === 'kling' ? <KlingFields id="video-kling" accessKeyCredential="video.kling.accessKey" secretKeyCredential="video.kling.secretKey" configuredCredentials={props.configuredCredentials} value={props.config.video.kling} modelLabel={l('默认模型', 'Default model')} modelPlaceholder={defaultVideoGenerationModels.kling} modelSuggestions={creatorProviderOfKind('video', 'kling-video')?.models.map(model => model.id)} onChange={value => props.update(config => { config.video.kling = value; })} /> : null}{provider === 'veo' ? <OpenAiFields id="video-veo" credential="video.veo.apiKey" configuredCredentials={props.configuredCredentials} value={props.config.video.veo} modelLabel={l('默认模型', 'Default model')} modelPlaceholder={defaultVideoGenerationModels.veo} modelSuggestions={videoGenerationModelIds.veo} baseUrlPlaceholder="https://generativelanguage.googleapis.com/v1beta" onChange={value => props.update(config => { config.video.veo = value; })} /> : null}</SettingsFieldset>);@@ -866,7 +933,7 @@ function OpenAiFields(props: {{props.modelReadonly ? (<ReadonlyModelField label={props.modelLabel ?? l('模型', 'Model')} value={props.modelPlaceholder} />) : (- <TextField+ <ModelFieldid={`${props.id}-model`}label={props.modelLabel ?? l('模型', 'Model')}value={props.value.model}@@ -895,7 +962,7 @@ function KlingFields(props: {<TextField id={`${props.id}-base-url`} label="Base URL" value={props.value.baseUrl} placeholder="https://api-beijing.klingai.com" onChange={baseUrl => props.onChange({ ...props.value, baseUrl })} wide /><PasswordField id={`${props.id}-access-key`} label="Access Key" value={props.value.accessKey} configured={props.configuredCredentials.has(props.accessKeyCredential)} onChange={accessKey => props.onChange({ ...props.value, accessKey })} /><PasswordField id={`${props.id}-secret-key`} label="Secret Key" value={props.value.secretKey} configured={props.configuredCredentials.has(props.secretKeyCredential)} onChange={secretKey => props.onChange({ ...props.value, secretKey })} />- <TextField id={`${props.id}-model`} label={props.modelLabel ?? 'Model'} value={props.value.model} placeholder={props.modelPlaceholder} suggestions={props.modelSuggestions} onChange={model => props.onChange({ ...props.value, model })} />+ <ModelField id={`${props.id}-model`} label={props.modelLabel ?? 'Model'} value={props.value.model} placeholder={props.modelPlaceholder} suggestions={props.modelSuggestions} onChange={model => props.onChange({ ...props.value, model })} /></>);}@@ -990,6 +1057,56 @@ function TextField(props: {);}+function ModelField(props: {+ id: string;+ label: string;+ value: string;+ placeholder?: string;+ suggestions?: readonly string[];+ onChange(value: string): void;+}) {+ const valueIsCustom = props.suggestions?.includes(props.value) !== true;+ const [customSelected, setCustomSelected] = useState(valueIsCustom);+ const showCustomInput = customSelected || valueIsCustom;+ const suggestionKey = props.suggestions?.join('\u0000') ?? '';+ useEffect(() => {+ if (props.suggestions?.includes(props.value) === true) setCustomSelected(false);+ }, [props.value, suggestionKey]);+ return (+ <label className="creator-services-field" htmlFor={props.id}>+ <span>{props.label}</span>+ {props.suggestions?.length ? (+ <select+ aria-label={props.label}+ value={showCustomInput ? '__custom__' : props.value}+ onChange={event => {+ if (event.target.value === '__custom__') {+ setCustomSelected(true);+ return;+ }+ setCustomSelected(false);+ props.onChange(event.target.value);+ }}+ >+ {props.suggestions.map(value => <option key={value} value={value}>{value}</option>)}+ <option value="__custom__">自定义模型 / Custom model</option>+ </select>+ ) : null}+ {(!props.suggestions?.length || showCustomInput) ? (+ <input+ id={props.id}+ type="text"+ value={props.value}+ placeholder={props.placeholder}+ spellCheck={false}+ autoComplete="off"+ onChange={event => props.onChange(event.target.value)}+ />+ ) : null}+ </label>+ );+}+function PasswordField(props: {id: string;label: string;diff --git a/apps/web/src/features/settings/RuntimeComponentsSettingsView.test.tsx b/apps/web/src/features/settings/RuntimeComponentsSettingsView.test.tsxindex cb68f8aa..5b9df4ea 100644--- a/apps/web/src/features/settings/RuntimeComponentsSettingsView.test.tsx+++ b/apps/web/src/features/settings/RuntimeComponentsSettingsView.test.tsx@@ -6,7 +6,7 @@ import { LanguageProvider } from '../../i18n/LanguageProvider.js';import { RuntimeComponentsSettingsView } from './RuntimeComponentsSettingsView.js';describe('RuntimeComponentsSettingsView', () => {- it('shows yt-dlp versions, source, schedule, and installs an available update', async () => {+ it('shows current, latest, and updated versions, and installs an available update', async () => {const updateYtDlp = vi.fn(async () => status({source: 'managed',currentVersion: '2026.08.31.120000'@@ -22,9 +22,9 @@ describe('RuntimeComponentsSettingsView', () => {expect(screen.getByRole('heading', { name: '第三方组件' })).toBeInTheDocument();expect(screen.getByRole('heading', { name: 'yt-dlp nightly' })).toBeInTheDocument();- expect(screen.getAllByText('2026.08.29.232711')).toHaveLength(2);- expect(screen.getByText('内置版本')).toBeInTheDocument();- expect(screen.getByText('每 7 天自动检查更新,不会自动安装。')).toBeInTheDocument();+ expect(screen.getAllByText('2026.08.29.232711')).toHaveLength(1);+ expect(screen.getByText('更新时间')).toBeInTheDocument();+ expect(screen.getByText('尚未检查')).toBeInTheDocument();fireEvent.click(screen.getByRole('button', {name: '更新到 2026.08.31.120000'@@ -45,8 +45,18 @@ describe('RuntimeComponentsSettingsView', () => {expect(screen.getByRole('alert')).toHaveTextContent('yt-dlp 更新校验失败,当前版本仍可继续使用。');- expect(screen.getByText('更新失败时,OpenCreator 会继续使用当前可用版本。'))- .toBeInTheDocument();+ expect(screen.queryByText('更新失败时,OpenCreator 会继续使用当前可用版本。'))+ .not.toBeInTheDocument();+ });++ it('shows a notice when a manual check finds no update', async () => {+ const checkYtDlpUpdate = vi.fn(async () => status());+ renderView({ checkYtDlpUpdate });++ fireEvent.click(screen.getByRole('button', { name: '检查更新' }));++ await waitFor(() => expect(screen.getByRole('status')).toHaveTextContent('已经是最新版本'));+ expect(checkYtDlpUpdate).toHaveBeenCalledWith(true);});});diff --git a/apps/web/src/features/settings/RuntimeComponentsSettingsView.tsx b/apps/web/src/features/settings/RuntimeComponentsSettingsView.tsxindex be4860a0..61afa388 100644--- a/apps/web/src/features/settings/RuntimeComponentsSettingsView.tsx+++ b/apps/web/src/features/settings/RuntimeComponentsSettingsView.tsx@@ -7,6 +7,7 @@ import {} from 'lucide-react';import type { RuntimeDependenciesController } from '../../app/use-runtime-dependencies.js';import { useAppLanguage } from '../../i18n/LanguageProvider.js';+import { useState } from 'react';export function RuntimeComponentsSettingsView(props: {connected: boolean;@@ -15,6 +16,7 @@ export function RuntimeComponentsSettingsView(props: {const { language, t } = useAppLanguage();const status = props.controller.ytDlpStatus;const busy = props.controller.phase !== 'idle';+ const [checkNotice, setCheckNotice] = useState<string>();const error = props.controller.error === undefined? undefined: formatRuntimeDependencyError(props.controller.error, t);@@ -58,9 +60,6 @@ export function RuntimeComponentsSettingsView(props: {);}- const source = status.source === 'managed'- ? t('settings.runtimeComponents.sourceManaged')- : t('settings.runtimeComponents.sourceBundled');const statusLabel = status.updateAvailable? t('settings.runtimeComponents.updateAvailable'): t('settings.runtimeComponents.current');@@ -91,45 +90,37 @@ export function RuntimeComponentsSettingsView(props: {<dt>{t('settings.runtimeComponents.currentVersion')}</dt><dd>{status.currentVersion}</dd></div>- <div>- <dt>{t('settings.runtimeComponents.bundledVersion')}</dt>- <dd>{status.bundledVersion}</dd>- </div>- <div>- <dt>{t('settings.runtimeComponents.source')}</dt>- <dd>{source}</dd>- </div><div><dt>{t('settings.runtimeComponents.latestVersion')}</dt><dd>{status.latestVersion ?? t('settings.runtimeComponents.notChecked')}</dd></div><div>- <dt>{t('settings.runtimeComponents.lastChecked')}</dt>- <dd>{formatDate(status.lastCheckedAt, language, t('settings.runtimeComponents.notChecked'))}</dd>+ <dt>{t('settings.runtimeComponents.installedAt')}</dt>+ <dd>{formatDate(status.installedAt, language, t('settings.runtimeComponents.notChecked'))}</dd></div>- {status.installedAt !== null ? (- <div>- <dt>{t('settings.runtimeComponents.installedAt')}</dt>- <dd>{formatDate(status.installedAt, language, t('settings.runtimeComponents.notChecked'))}</dd>- </div>- ) : null}</dl>{error === undefined ? null : (<p className="settings-error" role="alert">{error}</p>)}+ {checkNotice === undefined ? null : (+ <p className="settings-notice" role="status">{checkNotice}</p>+ )}<footer className="runtime-component-footer">- <div>- <strong>{t('settings.runtimeComponents.autoCheck')}</strong>- <small>{t('settings.runtimeComponents.fallback')}</small>- </div><div className="runtime-component-actions"><buttonclassName="settings-secondary-button"type="button"disabled={busy}- onClick={() => void props.controller.checkYtDlpUpdate(true).catch(() => undefined)}+ onClick={() => {+ setCheckNotice(undefined);+ void props.controller.checkYtDlpUpdate(true)+ .then(result => setCheckNotice(result.updateAvailable+ ? t('settings.runtimeComponents.updateAvailable')+ : t('settings.runtimeComponents.latestNotice')))+ .catch(() => undefined);+ }}>{props.controller.phase === 'checking'? <LoaderCircle className="settings-spin" size={15} aria-hidden="true" />diff --git a/apps/web/src/i18n/LanguageProvider.tsx b/apps/web/src/i18n/LanguageProvider.tsxindex b4dc1dea..54603aff 100644--- a/apps/web/src/i18n/LanguageProvider.tsx+++ b/apps/web/src/i18n/LanguageProvider.tsx@@ -118,6 +118,7 @@ const messages = {'settings.runtimeComponents.updating': '正在更新','settings.runtimeComponents.updateAvailable': '有可用更新','settings.runtimeComponents.current': '已是最新',+ 'settings.runtimeComponents.latestNotice': '已经是最新版本','settings.runtimeComponents.loading': '正在读取第三方组件状态','settings.runtimeComponents.disconnected': '连接本地运行内核后可以管理第三方组件。','settings.runtimeComponents.errorCheck': '暂时无法检查 yt-dlp 更新,请稍后重试。',@@ -364,6 +365,7 @@ const messages = {'settings.runtimeComponents.updating': 'Updating','settings.runtimeComponents.updateAvailable': 'Update available','settings.runtimeComponents.current': 'Up to date',+ 'settings.runtimeComponents.latestNotice': 'Already up to date','settings.runtimeComponents.loading': 'Loading third-party component status','settings.runtimeComponents.disconnected': 'Connect the local runtime to manage third-party components.','settings.runtimeComponents.errorCheck': 'Unable to check for yt-dlp updates right now. Try again later.',diff --git a/apps/web/src/services/connection-service.ts b/apps/web/src/services/connection-service.tsindex 323c73a2..f5059daf 100644--- a/apps/web/src/services/connection-service.ts+++ b/apps/web/src/services/connection-service.ts@@ -3,6 +3,7 @@ import type {CodexLoginCancelResponse,CodexLoginStartRequest,CodexLoginStartResponse,+ CodexModelListResponse,CodexLogoutResponse,CodexProviderConfig,CodexProviderConfigUpdateRequest,@@ -39,6 +40,9 @@ export function createConnectionService(client: ClientLike) {getCodexProvider(): Promise<CodexProviderConfig> {return client.get<CodexProviderConfig>('/codex/provider');},+ getCodexModels(): Promise<CodexModelListResponse> {+ return client.get<CodexModelListResponse>('/codex/models');+ },updateCodexProvider(request: CodexProviderConfigUpdateRequest): Promise<CodexProviderConfig> {diff --git "a/docs/plans/OpenCreator\347\273\237\344\270\200AI\346\234\215\345\212\241\344\276\233\345\272\224\345\225\206\344\270\216\351\205\215\347\275\256\344\275\223\351\252\214\345\256\236\346\226\275\350\256\241\345\210\222-2026-09-14.md" "b/docs/plans/OpenCreator\347\273\237\344\270\200AI\346\234\215\345\212\241\344\276\233\345\272\224\345\225\206\344\270\216\351\205\215\347\275\256\344\275\223\351\252\214\345\256\236\346\226\275\350\256\241\345\210\222-2026-09-14.md"new file mode 100644index 00000000..c7888516--- /dev/null+++ "b/docs/plans/OpenCreator\347\273\237\344\270\200AI\346\234\215\345\212\241\344\276\233\345\272\224\345\225\206\344\270\216\351\205\215\347\275\256\344\275\223\351\252\214\345\256\236\346\226\275\350\256\241\345\210\222-2026-09-14.md"@@ -0,0 +1,139 @@+# OpenCreator 统一 AI 服务供应商与配置体验实施计划++> 状态:草案+> 体量判断:复杂。执行跨协议、配置持久化、多个外部 API、Web/Desktop parity 和实际打包 App,保持单一 Plan 以保证共享契约一致。+> 来源方案:[OpenCreator统一AI服务供应商与配置体验方案-2026-09-14.md](../specs/OpenCreator统一AI服务供应商与配置体验方案-2026-09-14.md)+> 用户批准:已批准(2026-09-14);用户知情接受方案未经过独立 Reviewer 审核的风险。+> Reviewer 原始结论:未执行(当前会话未提供 `zhiyu-reviewer` Agent 调用能力)+> 流程结论:PASS(用户知情授权执行)+> 执行授权:已批准(2026-09-14,用户知情接受未独立审核风险)++## 契约快照++目标:统一 Provider Catalog、能力声明、默认地址/模型和配置 UI;首批支持 LLM 的 OpenAI/DeepSeek/MiniMax/Custom,图片的 OpenAI/即梦/可灵/Gemini,视频的 Seedance/可灵并核验 Veo,TTS 的 OpenAI/MiniMax/阿里云/Edge TTS;保留现有转写能力。++非目标:不代理第三方 API;不把专用协议强行改成 OpenAI-compatible;不替换 CreatorCollaborationPanel;规划 provider 不得标记为已支持。++约束:模型使用“目录下拉 + 自定义”;provider 切换自动填充默认值但允许自定义 Base URL;空密钥不覆盖已存密钥;旧配置读写归一化且不丢失;错误区分配置、鉴权、模型、能力、限流、网络和响应格式;通用进度只输出 `phase`、`percent`、`message`、`completed`、`failed` 等字段;Web/Desktop 共用前端、Daemon API 和持久化结果。++关键决策:DEC-1 Catalog 与 adapter 分离;DEC-2 默认值可编辑;DEC-3 自定义模型;DEC-4 能力控制入口;DEC-5 统一进度;DEC-6 supported/planned 分离。++## 基线与文件地图++- 协议:`packages/protocol/src/creator-services.ts`、`packages/protocol/src/media-generation.ts`、`packages/protocol/src/index.ts`。+- 配置:`apps/daemon/src/creator-services/config-store.ts`、`apps/daemon/src/api/routes.creator-services.ts`。+- 图片:`apps/daemon/src/image-generation/provider.ts`、`apps/daemon/src/image-generation/service.ts`。+- 视频:`apps/daemon/src/video-generation/service.ts`。+- Web 设置:`apps/web/src/features/settings/CreatorServicesSettingsView.tsx` 及 `creator-services-settings.css`。+- 已有测试:`apps/daemon/test/integration/creator-services-api.test.ts`、`image-generation-api.test.ts`、`creator-video-generation.test.ts`、`apps/web/src/features/settings/CreatorServicesSettingsView.test.tsx`。+- parity/App:`apps/web/e2e/web-desktop-parity.spec.ts`、`apps/desktop/e2e/desktop.spec.ts`。+- 基线:分支 `feat-optimize`,commit `14373968`。工作区另有无关未跟踪 `.codex/skills/oc-deploy/`、`.tools/`,不得修改。+- 基线命令:`pnpm --filter @opencreator/protocol typecheck`;`pnpm --filter @opencreator/daemon typecheck`;`pnpm --filter @opencreator/web typecheck`;目标包测试使用各自 `test -- <path/filter>`;最终 parity 使用现有 Playwright 配置,实际 App 使用 Desktop E2E 命令。++## 追踪矩阵++| 实施任务 | 需求/规则 | 决策 | 自动化测试 | 功能验收 |+|---|---|---|---|---|+| TASK-1 Catalog 与协议 | FR-1..FR-4, FR-6, BR-2 | DEC-1, DEC-3, DEC-4, DEC-6 | protocol Catalog/model/capability tests | AC-1, AC-2, AC-3 |+| TASK-2 配置 schema 与兼容 | FR-5, BR-1, BR-4 | DEC-2, DEC-3 | config-store normalization/credential tests | AC-3, AC-4 |+| TASK-3 LLM adapter 与设置状态 | FR-1, FR-3, FR-4 | DEC-2, DEC-3 | settings service/API and LLM provider tests | AC-1, AC-3 |+| TASK-4 图片/视频/TTS adapter | FR-2, FR-6, FR-7, BR-3 | DEC-1, DEC-4, DEC-5 | image/video/TTS integration tests | AC-2, AC-5, AC-6 |+| TASK-5 Web 设置交互 | FR-2..FR-7, BR-2 | DEC-1..DEC-4 | CreatorServicesSettingsView tests | AC-1..AC-5 |+| TASK-6 完整验收 | NFR-1, all P0/P1 | DEC-4..DEC-6 | parity and packaged App E2E | AC-7, AC-8 |++## 执行任务++### TASK-0:确认 Plan 仍然有效++核对当前分支、commit、方案批准记录、上述路径和现有符号。若 DEC、接口、数据结构或 AC 已失效,停止并退回 Brainstorm;无关工作区变化记录后继续。++### TASK-1:建立共享 Provider Catalog 与媒体能力协议 `[FR-1..FR-4, FR-6, BR-2, DEC-1, DEC-3, DEC-4, DEC-6, AC-1..AC-3]`++**交付结果**:protocol 导出 Catalog 条目、模型条目、认证字段和能力标签;首批 provider 与 planned provider 有明确状态。++**文件与符号**:修改 `packages/protocol/src/creator-services.ts`、`media-generation.ts`、`index.ts`;新增或扩展 `CreatorProviderCatalogEntry`、模型和能力类型及静态目录。++**实施步骤**:增加服务类型、协议类型、默认地址、模型 ID/label/recommended、凭据字段和能力声明;将现有 provider/model 常量迁移到单一导出;保持现有请求联合类型兼容。++**TDD**:必须。RED:新增 `provider catalog exposes DeepSeek and MiniMax defaults`,断言当前缺失的条目和能力失败。GREEN:实现共享目录并保留旧导出别名;运行 protocol typecheck/test。不得通过测试删除旧类型。++**完成门**:Catalog 测试、protocol typecheck 通过;覆盖 AC-1、AC-2、AC-3。++### TASK-2:配置 schema、凭据和旧配置归一化 `[FR-5, BR-1, BR-4, DEC-2, DEC-3, AC-3, AC-4]`++**交付结果**:配置读写由 Catalog 驱动默认补全,未知模型和自定义地址可保存,空密钥不覆盖已存值。++**文件与符号**:修改 `apps/daemon/src/creator-services/config-store.ts` 的 schema、`parseCreatorServicesConfig`、`retainCreatorServicesCredentials`、`presentCreatorServicesConfig`;同步 `packages/protocol/src/creator-services.ts` 凭据字段。++**实施步骤**:为 LLM provider 增加结构化来源而不破坏 `source` 兼容;扩展首批 provider 配置字段;把 credential retain/redact 路径同步到每个新增字段;读时补默认值,写时保留未知 model。++**TDD**:必须。RED:在 `creator-services-api.test.ts` 增加旧 OpenAI-compatible 配置、自定义模型和空 key PATCH 场景,断言当前 schema/retain 行为失败。GREEN:完成归一化并验证密钥不回显。运行 daemon 定向测试和 typecheck。++**完成门**:配置解析、兼容、脱敏和 API 定向测试通过;覆盖 AC-3、AC-4。++### TASK-3:LLM provider 选择和请求兼容 `[FR-1, FR-3, FR-4, DEC-2, DEC-3, AC-1, AC-3]`++**交付结果**:OpenAI、DeepSeek、MiniMax、Custom 选择后使用正确默认地址和模型,保留 Codex 认证兼容。++**文件与符号**:修改 LLM provider 配置服务及其测试;按现有 `ConnectionService`、`creator-services-service.ts` 和设置页 `TextModelSettings` 的实际调用边界接入 Catalog;新增/修改 Daemon 文本请求适配测试。++**实施步骤**:provider 选择产生可编辑 `baseUrl/model`;官方 provider 使用统一 OpenAI-compatible 请求;保存时空 API key 使用凭据保留;自定义 provider 保持用户 URL 和 model;错误转换为统一 API error。++**TDD**:必须。RED:新增三家 provider 的 base URL/model 选择和 fake upstream Authorization 测试,当前没有 provider 目录映射时失败。GREEN:实现选择映射和请求;运行 Web 设置定向测试、daemon 文本定向测试和两端 typecheck。++**完成门**:三家 provider 真实请求路径由 fake upstream 验证,Codex/custom 回归通过。++### TASK-4:图片、视频和 TTS 的 adapter 与能力边界 `[FR-2, FR-6, FR-7, BR-3, DEC-1, DEC-4, DEC-5, AC-2, AC-5, AC-6]`++**交付结果**:现有图片、视频、TTS provider 与 Catalog 对齐;Veo 创建/轮询/下载完整性被验证;首批新增 adapter 只在真实接口契约明确时启用。++**文件与符号**:修改 `apps/daemon/src/image-generation/provider.ts`、`image-generation/service.ts`、`video-generation/service.ts`、TTS 所在 `apps/daemon/src/creator/krillin/tts-service.ts` 及相关协议/模板 schema;测试使用现有 `image-generation-api.test.ts`、`creator-video-generation.test.ts`、TTS 集成测试。++**实施步骤**:按 Catalog 能力校验参考图、尺寸、声音和异步能力;统一错误和进度字段;对 URL 结果下载并本地保存;逐个核对 Gemini/Nano Banana、即梦、可灵、Seedance、Veo、MiniMax、阿里云的请求/响应格式;不把规划 provider 接入可用枚举。++**TDD**:必须。RED:为每个适配簇增加 fake upstream 请求路径、响应转换和不支持能力测试,当前缺少 Catalog 能力绑定时失败。GREEN:实现最小 adapter 和归一化;运行 daemon 图片/视频/TTS 定向集成测试。相同根因两次修复仍失败立即熔断该簇。++**完成门**:每个启用 provider 有请求、成功、上游错误和能力失败测试;覆盖 AC-2、AC-5、AC-6。++### TASK-5:Web 设置页统一交互 `[FR-2..FR-7, BR-2, DEC-1..DEC-4, AC-1..AC-5]`++**交付结果**:设置页按 Catalog 渲染 provider、模型下拉、自定义模型、默认地址、凭据字段和能力状态。++**文件与符号**:修改 `apps/web/src/features/settings/CreatorServicesSettingsView.tsx`、`creator-services-settings.css`、相关 settings service/types 和 `CreatorServicesSettingsView.test.tsx`。++**实施步骤**:抽取 provider/model selector;provider 切换更新默认值但不删除其他配置;未知模型选择 custom;根据 credentials/capabilities 控制字段和入口;保持密钥只显示 configured;增加配置缺失、不可用和上游错误文案。++**TDD**:必须。RED:组件测试选择 DeepSeek 后断言默认 URL/model、选择 custom 后编辑模型、已有 key 空提交仍配置,当前 UI 无 Catalog 驱动行为时失败。GREEN:接入共享目录和状态;运行 Web settings 定向测试及 typecheck。++**完成门**:组件测试通过,页面没有规划 provider 的无效入口,覆盖 AC-1..AC-5。++### TASK-6:执行完整功能验收 `[NFR-1, DEC-4, DEC-5, DEC-6, AC-7, AC-8]`++**本地实现差异自审**:检查 `git diff --stat` 和相关 diff,对照全部 TASK、契约和 AC;修复本任务相关遗漏、错误处理、安全、兼容或无关修改后重跑受影响测试。++**验收环境**:使用同一 Fake Daemon、相同项目/偏好和内容视口;fake upstream 不产生真实费用。实际 Desktop 验收使用当前打包流程,打包前重新构建 `apps/web`,并比较 `apps/web/dist` 与 App 内嵌资源文件列表和哈希。++| AC | 场景与操作 | 预期 | 验证方式 | 证据 |+|---|---|---|---|---|+| AC-1 | 设置页依次选择 OpenAI/DeepSeek/MiniMax 并保存 | 默认地址、模型、字段正确且请求可用 | Browser/Desktop parity + fake API | 记录命令、时间、退出码和输出摘要 |+| AC-2 | 切换图片/视频/TTS provider | 仅显示真实能力入口,规划项不可用 | Web E2E + adapter API | 同上 |+| AC-3 | 选择目录模型、自定义模型、未知旧模型并重开 | 值正确保存和回显 | Web + Daemon API | 同上 |+| AC-4 | 已配置 key 时提交空 key、读取旧配置 | 密钥不丢失、不明文回显 | Daemon integration | 同上 |+| AC-5 | TTS 选择 provider/model/voice 并试听 | 声音列表和错误分类正确 | Web E2E + fake upstream | 同上 |+| AC-6 | 图片/视频/TTS 执行成功和上游失败 | adapter 转换正确,进度/错误符合统一契约 | Daemon integration | 同上 |+| AC-7 | Browser/Desktop 相同视口读取并保存配置 | DOM、文案、尺寸、Runtime 请求、持久化结果一致 | `apps/web/e2e/web-desktop-parity.spec.ts` | 同上 |+| AC-8 | 启动实际打包 App | Preload、协议代理、配置流程正常,嵌入资源哈希一致 | `apps/desktop/e2e/desktop.spec.ts` + 校验脚本 | 同上 |++任一 P0 AC 为 FAIL 或 BLOCKED,不得宣布完成或可发布。单元测试不能替代 AC-7/AC-8 功能证据。++## 失败熔断与偏差++预期 RED 不计失败;进入 GREEN 后每次修复记录失败证据、根因假设和最小改动。同一测试因同一根因两次实质修复仍失败,停止该 TASK 并报告 BLOCKED。若发现需要改变 FR/BR/NFR/DEC/AC、公共接口或数据语义,停止并退回 Brainstorm。不得降低断言或删除失败测试。++## 独立审核记录++Reviewer 原始结论:未执行。当前会话没有 `zhiyu-reviewer` 调用接口,因此无法伪造 Plan 审核结论。用户已知悉方案和 Plan 阶段的审核缺失风险,并于 2026-09-14 回复“没问题,继续”,授权按本 Plan 执行。++| 问题 ID | 严重程度 | 处理决定 | 修改位置 | 关闭证据或不采纳理由 | 遗留风险 |+|---|---|---|---|---|---|+| PLAN-REVIEWER-UNAVAILABLE | 流程异常 | 用户知情接受 | 头部、独立审核记录 | 当前工具能力不支持独立 Reviewer;用户已明确授权执行 | Plan 未经独立审核;provider API 和 Desktop 门禁必须按 TASK-6 严格验证 |diff --git "a/docs/specs/OpenCreator\347\273\237\344\270\200AI\346\234\215\345\212\241\344\276\233\345\272\224\345\225\206\344\270\216\351\205\215\347\275\256\344\275\223\351\252\214\346\226\271\346\241\210-2026-09-14.md" "b/docs/specs/OpenCreator\347\273\237\344\270\200AI\346\234\215\345\212\241\344\276\233\345\272\224\345\225\206\344\270\216\351\205\215\347\275\256\344\275\223\351\252\214\346\226\271\346\241\210-2026-09-14.md"new file mode 100644index 00000000..56c58a65--- /dev/null+++ "b/docs/specs/OpenCreator\347\273\237\344\270\200AI\346\234\215\345\212\241\344\276\233\345\272\224\345\225\206\344\270\216\351\205\215\347\275\256\344\275\223\351\252\214\346\226\271\346\241\210-2026-09-14.md"@@ -0,0 +1,135 @@+# OpenCreator 统一 AI 服务供应商与配置体验方案++> 状态:已批准(2026-09-14)+> 体量判断:复杂。范围跨协议、Daemon、Web 设置页、Creator Runtime、模板参数与 Desktop 一致性,但各服务共享同一配置边界,因此保持一份内聚方案。+> 设计确认:已完成(D-1 至 D-4)+> Reviewer 原始结论:未执行(当前会话未提供 `zhiyu-reviewer` Agent 调用能力)+> 流程结论:PASS(用户知情批准)+> 用户批准:已批准(2026-09-14,用户知情接受未独立审核风险)++## 背景、目标与非目标++目标是梳理当前 AI 能力和 API 支持范围,建立统一的供应商目录、默认 Base URL、默认模型和能力声明,并让用户通过“选择供应商 -> 选择模型 -> 填写认证信息”完成配置。Web 与 Desktop 使用同一套 Web 页面、Daemon API 和持久化结果。++非目标:本轮不代理或购买第三方 API;不把非兼容服务强行抽象成 OpenAI-compatible;不替换 CreatorCollaborationPanel;不把实时联网模型市场作为首版依赖;未具备稳定公开 API 的服务不标记为已支持。++## 用户需求原文++> 比如,LLM目前支持了openai协议,但是配置不友好,可以让用户选择比如:deepseek、openai、minimax,然后自动就把base_url配好了+>+> 图片生成的支持,要加入极梦、可灵、nanobanana等等主流生图+>+> 视频生成,tts也一样,主流的都梳理一下,增加支持+>+> 配置的交互要优化,就想第1点提到的那样,要很方便的能配置,模型做到下拉可选。++## 事实基线与假设++| 证据 | 已确认事实 |+|---|---|+| `packages/protocol/src/creator-services.ts` | 统一配置已有 LLM、转写、TTS、图片、视频结构;LLM 仍是 OpenAI-compatible 配置。 |+| `packages/protocol/src/media-generation.ts` | 图片 provider 为 OpenAI、即梦、可灵、Gemini;视频 provider 为 Seedance、可灵、Veo,并已有部分模型目录。 |+| `apps/daemon/src/creator-services/config-store.ts` | 配置使用 Zod 校验,凭据分离存储并支持脱敏、旧配置归一化。 |+| `apps/daemon/src/image-generation/provider.ts`、`video-generation/service.ts` | 图片和视频已有专用请求/轮询逻辑,但能力和 provider 仍分散。 |+| `apps/web/src/features/settings/CreatorServicesSettingsView.tsx` | 设置页已按服务分区,部分 provider 有字段和模型建议,但存在硬编码和 LLM 体验不一致。 |+| `apps/web/e2e/web-desktop-parity.spec.ts`、`apps/desktop/e2e/desktop.spec.ts` | 项目已有 Web/Desktop parity 与实际打包 App E2E 门禁。 |++假设:首批 API 以公开、可授权、能够使用现有 Daemon 资产和任务模型验证的服务为准;其余服务进入规划目录,不作为已支持能力。++## 设计确认记录++| 设计部分 | 核心决定 | 用户确认原话 |+|---|---|---|+| D-1 目标与总体范围 | 采用统一 AI 服务供应商与配置体验方案;首批实现可验证 provider,其余进入规划清单。 | “没问题,继续” |+| D-2 供应商目录与适配分层 | 使用 Provider Catalog + 能力声明 + 专用 adapter;首批覆盖 LLM OpenAI/DeepSeek/MiniMax/Custom,图片 OpenAI/即梦/可灵/Gemini,视频 Seedance/可灵并核验 Veo,TTS OpenAI/MiniMax/阿里云/Edge TTS。 | “没问题,继续” |+| D-3 配置交互与兼容 | 供应商先选、模型下拉加自定义、自动填充地址和默认模型;保留旧配置、密钥和自定义地址;错误按配置、能力、上游和网络分类。 | “没问题,继续” |+| D-4 验收与交付 | 分配置体验、真实适配、Web/Desktop 一致性三阶段交付;不以未验证的供应商宣称已支持。 | “没问题,继续” |++## 需求与业务规则++| ID | 类型 | 优先级 | 描述 |+|---|---|---:|---|+| FR-1 | 功能 | P0 | LLM 支持 OpenAI、DeepSeek、MiniMax 和 Custom OpenAI-compatible 的明确选择。 |+| FR-2 | 功能 | P0 | 图片、视频、TTS provider 通过统一配置体验展示已支持能力,并清晰区分规划项。 |+| FR-3 | 功能 | P0 | Provider 选择自动填充 Base URL、推荐模型和认证字段。 |+| FR-4 | 功能 | P0 | 模型支持下拉选择和自定义模型,未知旧模型可正确回显。 |+| FR-5 | 功能 | P0 | 配置密钥继续安全存储、脱敏,并在 provider 切换和保存时保留已有密钥。 |+| FR-6 | 功能 | P1 | Provider 能力声明控制参考图、图生视频、声音列表等字段和入口。 |+| FR-7 | 功能 | P1 | TTS 声音按 provider/model 提供,支持内置声音、自定义 ID 和试听错误反馈。 |+| BR-1 | 规则 | P0 | 配置保存成功不等于上游连接成功;连接测试为用户主动操作。 |+| BR-2 | 规则 | P0 | 不可用或规划中的 provider 不得显示为点击后无效的可用入口。 |+| BR-3 | 规则 | P0 | 非 OpenAI-compatible provider 使用专用 adapter,不改变通用 Runtime 进度契约。 |+| BR-4 | 规则 | P0 | 旧配置读写归一化时保留用户地址、模型和凭据,不静默重置。 |+| NFR-1 | 约束 | P0 | Web/Desktop 共用页面、Catalog、Daemon API 和持久化结果,并在相同内容视口下验证一致。 |++## 方案比较与推荐++推荐 Provider Catalog + 能力声明 + 专用 adapter。Catalog 负责显示名、官方地址、默认模型、模型列表、认证方式、能力和文档链接;adapter 负责真实请求、异步任务、响应转换和错误映射。++仅在设置页中继续硬编码 provider 会导致 Web 与 Daemon 的支持范围漂移,排除。仅统一为 OpenAI-compatible 会无法表达可灵、视频任务和云厂商专用鉴权,排除。Catalog 不负责发请求,避免把产品元数据和上游协议耦合。++## 关键设计决策++| DEC ID | 决策 | 理由 | 约束范围 |+|---|---|---|---|+| DEC-1 | 共享包提供 Provider Catalog 和能力声明;Daemon adapter 负责上游协议。 | 统一 Web/Daemon 配置认知,同时保留专用协议差异。 | protocol、web、daemon |+| DEC-2 | Provider 切换自动填充默认值,但允许修改 Base URL、模型和自定义网关。 | 兼顾易用性、代理、中转和新模型。 | 配置 UI、schema、兼容 |+| DEC-3 | 模型选择为“推荐/可选列表 + 自定义模型”。 | 不阻塞未收录模型和旧配置。 | 所有生成与文本服务 |+| DEC-4 | 能力不足时隐藏或禁用对应入口,并提供明确原因。 | 防止伪一致和点击无效。 | 图片、视频、TTS、Desktop parity |+| DEC-5 | 统一外部进度使用 `phase`、`percent`、`message`、`completed`、`failed` 等字段。 | 通用 Panel 不依赖供应商私有事件。 | Creator Runtime、Panel |+| DEC-6 | 首批范围与规划范围分离;真实 API 未验证不得标记 supported。 | 控制交付风险和支持口径。 | Catalog、文档、验收 |++## 首批支持范围++- LLM:OpenAI、DeepSeek、MiniMax、Custom OpenAI-compatible。+- 图片:OpenAI、即梦、可灵、Gemini/Nano Banana;核对各自参考图和尺寸能力。+- 视频:Seedance、可灵;核验并补全 Veo 的创建、轮询和下载。+- TTS:OpenAI、MiniMax、阿里云、Edge TTS。+- 转写:保留 OpenAI、阿里云、Faster Whisper、WhisperKit、whisper.cpp,统一展示能力和配置。++规划目录可包含智谱、通义、Moonshot、OpenRouter、Runway、Luma、Pika、海螺、ElevenLabs、Azure Speech、火山引擎、腾讯云、Fish Speech、CosyVoice 等,但规划项不进入首批可用选择和完成统计。++## 详细设计++Catalog 条目至少包含 `id`、服务类型、显示名称、协议类型、默认 Base URL、模型条目、认证字段、能力标签、文档链接和 `supported/planned` 状态。模型条目包含 ID、显示名称、推荐标记和能力限制。++配置页采用“服务分区 -> provider 选择 -> model 选择 -> provider 字段”的顺序。API Key 只显示已配置状态;空值提交不得覆盖已存密钥。Provider 切换只影响当前表单默认值,不删除其他 provider 配置。++图片能力声明参考图数量、尺寸、输出格式和 URL/Base64 形式;视频能力声明文生/图生、时长、尺寸、进度可靠性和异步结果;TTS 能力声明声音目录来源、模型关联和试听能力。Daemon 负责将临时上游 URL 下载为本地资产,处理 URL 过期。++连接错误映射为配置缺失、认证失败、模型不存在、能力不支持、限流/配额、网络/代理/地区不可达和响应格式不兼容。异步视频若无可信进度,返回未知进度而不伪造百分比。++## 异常、兼容、迁移与回滚++保留现有配置文件和凭据文件,不新增数据库迁移。旧 provider ID 和旧 OpenAI-compatible 结构在读取时转换为当前结构;用户自定义地址、模型、密钥优先保留;未知模型回显为自定义模型。解析失败阻止保存并报告错误,不自动恢复默认。++每个 provider adapter 可独立关闭或从 Catalog 标记为 unavailable;已有配置仍可读取。若新 Catalog 或 adapter 发布后出现问题,可通过回退 provider 状态和保留旧配置结构恢复,不需要删除用户数据。++## 验收标准++| AC ID | 关联需求 | 前置条件 | 操作 | 可观察结果 | 验证层级 |+|---|---|---|---|---|---|+| AC-1 | FR-1, FR-3 | 打开模型设置 | 选择 OpenAI、DeepSeek、MiniMax | 地址、推荐模型和认证字段自动变化且可保存 | Web 组件 + Daemon |+| AC-2 | FR-2, FR-6 | 打开图片/视频/TTS 设置 | 切换不同 provider | 只显示真实支持字段,规划项不作为可用入口 | Web 组件 |+| AC-3 | FR-4 | provider 有模型目录 | 选择模型、自定义模型并重新打开 | 下拉选择和自定义值均能保存回显 | Web 组件 + Daemon |+| AC-4 | FR-5, BR-4 | 已存在旧配置和密钥 | 切换 provider、提交空 Key、重新读取 | 地址、模型和密钥均不丢失,密钥不明文回显 | Daemon 单元/集成 |+| AC-5 | FR-7 | TTS provider 可用 | 选择模型、声音并试听 | 声音列表按 provider/model 返回,错误信息可区分 | Daemon 集成 + Web |+| AC-6 | BR-3, DEC-5 | 使用 fake upstream | 执行图片、视频和 TTS 请求 | adapter 请求正确,Runtime 输出统一契约 | Daemon 集成 |+| AC-7 | NFR-1 | 同一 Fake Daemon、相同视口 | Browser/Desktop 执行配置读取和保存 | DOM、文案、关键尺寸、Runtime 请求和持久化结果一致 | parity E2E |+| AC-8 | NFR-1 | 当前 Web 构建和 Desktop 打包 | 启动实际 App 并检查资源 | App 加载本次 Web 产物,Preload、代理和配置流程可用,资源哈希一致 | Desktop E2E/打包校验 |++## 测试策略++协议层覆盖 Catalog 类型、provider/model 目录和能力声明;Daemon 覆盖 schema、旧配置归一化、凭据保留、每个 adapter 的请求/响应/错误和异步轮询;Web 覆盖切换、默认值、下拉、自定义模型、脱敏和字段可见性;parity E2E 覆盖 Browser/Desktop 相同配置流程;实际打包 App 覆盖嵌入资源、Preload、Runtime 代理和配置持久化。CI 使用 fake upstream,不依赖真实计费凭据;真实服务只做显式手动 smoke test。++## 风险与未决问题++未决问题是各规划供应商的公开 API、地区可用性和模型目录变化;不影响首批方案,但在进入 supported 前必须取得可验证的接口证据。Nano Banana 的产品名称与 Gemini API 模型 ID 需要在适配核验时明确映射,避免只按营销名称展示。++## 独立审核记录++Reviewer 原始结论:未执行。当前会话未提供 `zhiyu-reviewer` Agent 调用能力;用户已知悉该风险并明确批准进入 Plan。++| 问题 ID | 严重程度 | 处理决定 | 修改位置 | 关闭证据或不采纳理由 | 遗留风险 |+|---|---|---|---|---|---|+| REVIEWER-UNAVAILABLE | 流程异常 | 用户知情接受 | 独立审核记录、本文件头部 | 当前工具集中没有 `zhiyu-reviewer` 调用能力,无法伪造审核结论 | 方案未经过独立审核,Plan 和实现仍应重点复核 provider API、地区限制及 Web/Desktop 门禁 |diff --git a/packages/protocol/src/creator-provider-catalog.test.ts b/packages/protocol/src/creator-provider-catalog.test.tsnew file mode 100644index 00000000..b2387b9a--- /dev/null+++ b/packages/protocol/src/creator-provider-catalog.test.ts@@ -0,0 +1,38 @@+import { describe, expect, it } from 'vitest';+import {+ creatorProviderCatalog,+ creatorProviderCatalogById+} from './creator-provider-catalog.js';++describe('creator provider catalog', () => {+ it('provides selectable LLM providers with defaults', () => {+ expect(creatorProviderCatalogById.llm.openai).toMatchObject({+ models: [expect.objectContaining({ id: 'gpt-5.6-sol', recommended: true })]+ });+ expect(creatorProviderCatalogById.llm.deepseek).toMatchObject({+ label: 'DeepSeek',+ defaultBaseUrl: 'https://api.deepseek.com',+ models: expect.arrayContaining([+ expect.objectContaining({ id: 'deepseek-v4-pro' }),+ expect.objectContaining({ id: 'deepseek-v4-flash' }),+ expect.objectContaining({ id: 'deepseek-flash' })+ ])+ });+ expect(creatorProviderCatalogById.llm.minimax).toMatchObject({+ label: 'MiniMax',+ defaultBaseUrl: 'https://api.minimax.io',+ models: expect.arrayContaining([+ expect.objectContaining({ id: 'MiniMax-M2.7' }),+ expect.objectContaining({ id: 'MiniMax-M2.7-highspeed' })+ ])+ });+ expect(creatorProviderCatalogById.llm.minimax?.models).not.toEqual(+ expect.arrayContaining([expect.objectContaining({ id: 'MiniMax-Text-01' })])+ );+ });++ it('marks only implemented providers as supported', () => {+ expect(creatorProviderCatalog.every(entry => entry.status === 'supported')).toBe(true);+ expect(creatorProviderCatalogById.image.gemini?.status).toBe('supported');+ });+});diff --git a/packages/protocol/src/creator-provider-catalog.ts b/packages/protocol/src/creator-provider-catalog.tsnew file mode 100644index 00000000..17ccd862--- /dev/null+++ b/packages/protocol/src/creator-provider-catalog.ts@@ -0,0 +1,59 @@+export type CreatorProviderKind = 'llm' | 'image' | 'video' | 'tts' | 'transcription';+export type CreatorProviderProtocol = 'openai-compatible' | 'gemini' | 'kling' | 'custom';+export type CreatorProviderStatus = 'supported' | 'planned';+export type CreatorProviderCredential = 'apiKey' | 'accessKey' | 'secretKey' | 'region';++export type CreatorProviderModel = {+ id: string;+ label: string;+ recommended?: boolean;+ capabilities?: readonly string[];+};++export type CreatorProviderCatalogEntry = {+ id: string;+ kind: CreatorProviderKind;+ label: string;+ protocol: CreatorProviderProtocol;+ defaultBaseUrl?: string;+ models: readonly CreatorProviderModel[];+ credentials: readonly CreatorProviderCredential[];+ capabilities?: readonly string[];+ status: CreatorProviderStatus;+};++const entry = (value: CreatorProviderCatalogEntry) => value;++export const creatorProviderCatalog = [+ entry({ id: 'openai', kind: 'llm', label: 'OpenAI', protocol: 'openai-compatible', defaultBaseUrl: 'https://api.openai.com/v1', models: [{ id: 'gpt-5.6-sol', label: 'GPT-5.6 Sol', recommended: true }], credentials: ['apiKey'], status: 'supported' }),+ entry({ id: 'deepseek', kind: 'llm', label: 'DeepSeek', protocol: 'openai-compatible', defaultBaseUrl: 'https://api.deepseek.com', models: [{ id: 'deepseek-v4-pro', label: 'DeepSeek V4 Pro', recommended: true }, { id: 'deepseek-v4-flash', label: 'DeepSeek V4 Flash' }, { id: 'deepseek-flash', label: 'DeepSeek Flash' }], credentials: ['apiKey'], status: 'supported' }),+ entry({ id: 'minimax', kind: 'llm', label: 'MiniMax', protocol: 'openai-compatible', defaultBaseUrl: 'https://api.minimax.io', models: [{ id: 'MiniMax-M2.7', label: 'MiniMax M2.7', recommended: true }, { id: 'MiniMax-M2.7-highspeed', label: 'MiniMax M2.7 Highspeed' }, { id: 'MiniMax-M2.5', label: 'MiniMax M2.5' }, { id: 'MiniMax-M2.5-highspeed', label: 'MiniMax M2.5 Highspeed' }], credentials: ['apiKey'], status: 'supported' }),+ entry({ id: 'custom', kind: 'llm', label: 'Custom OpenAI-compatible', protocol: 'custom', models: [], credentials: ['apiKey'], status: 'supported' }),+ entry({ id: 'openai', kind: 'image', label: 'OpenAI', protocol: 'openai-compatible', defaultBaseUrl: 'https://api.openai.com/v1', models: [{ id: 'gpt-image-1', label: 'GPT Image', recommended: true }], credentials: ['apiKey'], capabilities: ['text-to-image', 'image-edit', 'reference-image'], status: 'supported' }),+ entry({ id: 'jimeng', kind: 'image', label: '即梦 / Seedream', protocol: 'openai-compatible', defaultBaseUrl: 'https://ark.cn-beijing.volces.com/api/v3', models: [{ id: 'doubao-seedream-4-0-250828', label: 'Seedream 4.0', recommended: true }], credentials: ['apiKey'], capabilities: ['text-to-image'], status: 'supported' }),+ entry({ id: 'kling', kind: 'image', label: '可灵 / Kling', protocol: 'kling', defaultBaseUrl: 'https://api-beijing.klingai.com', models: [{ id: 'kling-v2-1', label: 'Kling v2.1', recommended: true }], credentials: ['accessKey', 'secretKey'], capabilities: ['text-to-image'], status: 'supported' }),+ entry({ id: 'gemini', kind: 'image', label: 'Nano Banana / Gemini', protocol: 'gemini', defaultBaseUrl: 'https://generativelanguage.googleapis.com/v1beta', models: [{ id: 'gemini-2.5-flash-image', label: 'Nano Banana', recommended: true }], credentials: ['apiKey'], capabilities: ['text-to-image', 'reference-image'], status: 'supported' }),+ entry({ id: 'seedance', kind: 'video', label: 'Seedance', protocol: 'openai-compatible', defaultBaseUrl: 'https://ark.cn-beijing.volces.com/api/v3', models: [{ id: 'doubao-seedance-2-5-260628', label: 'Seedance 2.5', recommended: true }], credentials: ['apiKey'], capabilities: ['text-to-video', 'image-to-video'], status: 'supported' }),+ entry({ id: 'kling-video', kind: 'video', label: '可灵 / Kling', protocol: 'kling', defaultBaseUrl: 'https://api-beijing.klingai.com', models: [{ id: 'kling-v2-1-master', label: 'Kling v2.1 Master', recommended: true }], credentials: ['accessKey', 'secretKey'], capabilities: ['text-to-video', 'image-to-video'], status: 'supported' }),+ entry({ id: 'veo', kind: 'video', label: 'Veo', protocol: 'gemini', defaultBaseUrl: 'https://generativelanguage.googleapis.com/v1beta', models: [{ id: 'veo-3.1-generate-preview', label: 'Veo 3.1', recommended: true }], credentials: ['apiKey'], capabilities: ['text-to-video', 'image-to-video'], status: 'supported' }),+ entry({ id: 'openai-tts', kind: 'tts', label: 'OpenAI Voice', protocol: 'openai-compatible', defaultBaseUrl: 'https://api.openai.com/v1', models: [{ id: 'gpt-4o-mini-tts', label: 'GPT-4o mini TTS', recommended: true }], credentials: ['apiKey'], capabilities: ['speech-generation', 'voice-preview'], status: 'supported' }),+ entry({ id: 'minimax-tts', kind: 'tts', label: 'MiniMax Voice', protocol: 'openai-compatible', defaultBaseUrl: 'https://api.minimax.io', models: [{ id: 'speech-2.8-hd', label: 'Speech 2.8 HD', recommended: true }], credentials: ['apiKey'], capabilities: ['speech-generation', 'voice-preview'], status: 'supported' }),+ entry({ id: 'aliyun-tts', kind: 'tts', label: '阿里云 Voice', protocol: 'openai-compatible', defaultBaseUrl: 'https://dashscope.aliyuncs.com/api/v1', models: [{ id: 'qwen3-tts-flash', label: 'Qwen3 TTS Flash', recommended: true }], credentials: ['apiKey'], capabilities: ['speech-generation', 'voice-preview'], status: 'supported' }),+ entry({ id: 'edge-tts', kind: 'tts', label: 'Edge TTS', protocol: 'custom', models: [], credentials: [], capabilities: ['speech-generation'], status: 'supported' })+] as const;++export const creatorProviderCatalogById = {+ llm: Object.fromEntries(creatorProviderCatalog.filter(item => item.kind === 'llm').map(item => [item.id, item])),+ image: Object.fromEntries(creatorProviderCatalog.filter(item => item.kind === 'image').map(item => [item.id, item]))+} as Record<'llm' | 'image', Record<string, CreatorProviderCatalogEntry>>;++export function creatorProvidersOfKind(kind: CreatorProviderKind): readonly CreatorProviderCatalogEntry[] {+ return creatorProviderCatalog.filter(provider => provider.kind === kind);+}++export function creatorProviderOfKind(+ kind: CreatorProviderKind,+ id: string+): CreatorProviderCatalogEntry | undefined {+ return creatorProviderCatalog.find(provider => provider.kind === kind && provider.id === id);+}diff --git a/packages/protocol/src/index.ts b/packages/protocol/src/index.tsindex a8f80c97..613eb54f 100644--- a/packages/protocol/src/index.ts+++ b/packages/protocol/src/index.ts@@ -5,6 +5,7 @@ export * from './creator-visual-assets.js';export * from './creator-download.js';export * from './api.js';export * from './creator-services.js';+export * from './creator-provider-catalog.js';export * from './errors.js';export * from './events.js';export * from './media-generation.js';