feat: 完善创作工作台与用户诊断支持 (#295)

* feat(canvas): 参考素材 - 增加移除按钮并删除对应连线

* feat(canvas): 完善节点面板交互与拖拽反馈

* fix(canvas): 画布交互与小地图 - 统一浮层层级并完善节点识别

* feat(prompt-optimizer): 增加 AI 提示词优化器及宿主接入

* feat(canvas): 画布 - 增加自动整理节点按钮并补充文档

* chore(web): 前端 - 修正提示词优化器格式

* chore(backend): 后端 - 修正 Go 格式检查

* feat(workspace): 完善诊断导出与本地排障支持

* chore(web): 格式化诊断导出文件

* fix(web-test): 兼容 Bun 测试运行时

* fix(web-test): 对齐模型目录协议断言

web/src/lib/plugins/builtin/eagle.ts

@@ -299,6 +299,8 @@ function fileToDataUrl(file: Blob, signal?: AbortSignal): Promise<string> {
 
 function toExternalItem(item: EagleItem, baseUrl: string, pathMap: Map<string, string[]>) {
     const kind = kindFromExtension(item.extension);
+    const folderIds = item.folderIds || [];
+    const tags = item.tags || [];
     return {
         id: item.id,
         title: item.name,
@@ -309,10 +311,10 @@ function toExternalItem(item: EagleItem, baseUrl: string, pathMap: Map<string, s
         width: item.width,
         height: item.height,
         bytes: item.size,
-        tags: item.tags,
-        folderId: item.folderIds[0],
-        folderIds: item.folderIds,
-        folderPath: item.folderIds[0] ? pathMap.get(item.folderIds[0]) : [],
+        tags,
+        folderId: folderIds[0],
+        folderIds,
+        folderPath: folderIds[0] ? pathMap.get(folderIds[0]) : [],
         description: item.annotation,
         metadata: { extension: item.extension, modificationTime: item.modificationTime, url: item.url, deleted: item.deleted },
     } satisfies ExternalAssetItem;

web/src/lib/plugins/builtin/index.ts

@@ -1 +1,2 @@
 import "./eagle";
+import "./prompt-optimizer";

web/src/lib/plugins/builtin/prompt-optimizer.ts

@@ -0,0 +1,352 @@
+import { registerPlugin } from "../plugin-registry";
+import type {
+    PluginManifest,
+    PluginTextContentPart,
+    PluginTextMessage,
+    PluginTextTool,
+    PromptOptimizationInput,
+    PromptOptimizationMode,
+    PromptOptimizationResult,
+    PromptOptimizationVariant,
+    PromptOptimizerProvider,
+    RegisteredPlugin,
+} from "../plugin-types";
+
+export const PROMPT_OPTIMIZER_PLUGIN_ID = "prompt-optimizer";
+
+const modeLabels: Record<PromptOptimizationMode, string> = {
+    expand: "扩展想法",
+    refine: "精修已有提示词",
+    style: "强化视觉风格",
+    "model-adapt": "适配当前模型",
+    reference: "结合参考素材",
+};
+
+const optimizerTool: PluginTextTool = {
+    type: "function",
+    function: {
+        name: "optimize_prompt",
+        description: "返回结构化的 AI 生图或生视频提示词优化结果。",
+        strict: true,
+        parameters: {
+            type: "object",
+            additionalProperties: false,
+            required: ["optimizedPrompt", "negativePrompt", "changes", "assumptions", "variants"],
+            properties: {
+                optimizedPrompt: { type: "string", description: "可直接用于生成的正向提示词。" },
+                negativePrompt: { type: "string", description: "需要规避的内容；没有时返回空字符串。" },
+                changes: { type: "array", items: { type: "string" }, description: "相较输入提示词做出的关键变化。" },
+                assumptions: { type: "array", items: { type: "string" }, description: "对模糊需求做出的假设；没有时返回空数组。" },
+                variants: {
+                    type: "array",
+                    items: {
+                        type: "object",
+                        additionalProperties: false,
+                        required: ["label", "prompt"],
+                        properties: {
+                            label: { type: "string" },
+                            prompt: { type: "string" },
+                        },
+                    },
+                    description: "最多两个可选的提示词版本。",
+                },
+            },
+        },
+    },
+};
+
+const manifest: PluginManifest = {
+    id: PROMPT_OPTIMIZER_PLUGIN_ID,
+    name: "AI 提示词优化器",
+    version: "0.3.0",
+    apiVersion: "1",
+    category: "ai-capability",
+    description: "把模糊的生图想法整理成可执行、可比较、适配当前模型的提示词。",
+    author: "影策团队",
+    surfaces: ["hybrid"],
+    permissions: ["canvas.read", "canvas.write", "ai.text"],
+    trusted: true,
+};
+
+const systemPrompt = [
+    "你是影策工作台里的提示词导演，负责把用户模糊的视觉想法整理成可以直接交给生成模型的提示词。",
+    "只优化表达和可执行性，不擅自改变用户明确写出的主体、身份、动作、数量、时代、地点、画幅比例或安全边界。",
+    "优先使用具体可视化语言：主体与关系、构图与景别、动作、环境、材质、光线、色彩、镜头和风格。",
+    "如果输入已经足够明确，保持原意并做克制的精修；如果信息不足，把不确定项写入 assumptions，不要假装知道。",
+    "根据 generationMode 选择表达：图片侧重静态画面、构图和材质，视频侧重动作连续性、时长感、镜头运动和首尾衔接。",
+    "当优化模式是适配当前模型时，先结合 targetProtocol 和 targetModel 判断模型族，再严格执行模型适配策略；不要编造不存在的参数、权重语法或模型能力。",
+    "最终必须调用 optimize_prompt 工具，不要输出 Markdown，不要在工具调用之外解释。",
+].join("\n");
+
+type ModelAdaptationProfile = {
+    id: string;
+    label: string;
+    promptShape: string;
+    rules: string[];
+    avoid: string[];
+};
+
+const genericImageProfile: ModelAdaptationProfile = {
+    id: "generic-image",
+    label: "通用图片模型",
+    promptShape: "自然语言描述，按主体与动作 → 环境 → 构图 → 光线与材质 → 风格排列。",
+    rules: ["先锁定主体、数量、身份和动作，再补充可见的画面细节。", "把重要的比例、文字、位置和一致性要求写明确。"],
+    avoid: ["不要臆造供应商参数或控制语法。", "关键规避项不要只放在 negativePrompt 中，正向提示词也要表达清楚。"],
+};
+
+const genericVideoProfile: ModelAdaptationProfile = {
+    id: "generic-video",
+    label: "通用视频模型",
+    promptShape: "单镜头时序描述，按起始画面 → 主体动作 → 镜头运动 → 环境变化 → 结尾状态排列。",
+    rules: ["每个动作使用可观察、连续的动词，避免一段话塞入互不相连的多个镜头。", "明确主体、镜头和环境谁在运动，并保持角色、服装和场景一致。"],
+    avoid: ["不要把静态摄影参数堆成关键词列表。", "不要新增用户未要求的转场、角色或剧情。"],
+};
+
+function containsAny(value: string, keywords: string[]) {
+    return keywords.some((keyword) => value.includes(keyword));
+}
+
+function resolveModelAdaptationProfile(input: PromptOptimizationInput): ModelAdaptationProfile {
+    const model = `${input.targetModel || ""} ${input.targetProtocol || ""}`.toLowerCase();
+    if (input.generationMode === "image") {
+        if (containsAny(model, ["gemini", "imagen", "nano-banana", "nanobanana"])) {
+            return {
+                id: "gemini-image",
+                label: "Gemini / Imagen 图片模型",
+                promptShape: "清晰的自然语言场景描述，按主体 → 背景与上下文 → 风格与摄影细节组织。",
+                rules: ["优先使用完整句子和具体视觉事实，参考图只描述确实可见的内容。", "如果画面含文字，保留原文并明确位置、层级和字数，避免无关装饰。", "控制提示词长度，先保留主体、上下文和风格这三个核心层次。"],
+                avoid: ["不要使用 SD 权重、反向提示词标签或未确认的供应商参数。"],
+            };
+        }
+        if (containsAny(model, ["gpt-image", "dall-e", "dalle", "openai-image", "openai"])) {
+            return {
+                id: "openai-image",
+                label: "OpenAI 图片模型",
+                promptShape: "简洁、完整的自然语言描述，先说要生成什么，再补充构图、光线、材质和风格。",
+                rules: ["使用模型能直接理解的画面描述，不把提示词写成控制台参数。", "对画面中的文字、位置、数量和主体关系使用明确句子。"],
+                avoid: ["不要添加 SD 权重、逗号关键词堆叠或 -- 参数。", "不要依赖单独 negativePrompt，重要限制要写进正向描述。"],
+            };
+        }
+        if (containsAny(model, ["grok-image", "grok-imagine"])) {
+            return {
+                id: "grok-image",
+                label: "Grok 图片模型",
+                promptShape: "简洁的自然语言画面描述，主体、动作、环境和视觉风格清晰分层。",
+                rules: ["保留明确的主体关系、构图和风格意图，避免把提示词写成参数清单。", "有参考图时说明要保留的主体或风格特征，不臆测看不见的细节。"],
+                avoid: ["不要添加 SD 权重、Midjourney 参数或模型未确认支持的控制语法。"],
+            };
+        }
+        if (containsAny(model, ["seedream", "jimeng", "doubao", "volcengine-ark-image", "volcengine-jimeng-image"])) {
+            return {
+                id: "seedream-image",
+                label: "Seedream / 即梦 / 火山图片模型",
+                promptShape: "中文自然语言镜头描述，按主体与动作 → 构图 → 光影 → 材质与风格组织。",
+                rules: ["优先写清人物或主体的外观、动作、空间关系和画面重点。", "需要人物一致性时重复关键外观特征，避免只写抽象的‘保持一致’。", "中文需求保持中文表达，英文专有名词只在确实有帮助时保留。"],
+                avoid: ["不要套用 SD 的权重语法、质量词串或无意义的英文标签。", "不要把多个镜头和互相冲突的动作混在一条提示词里。"],
+            };
+        }
+        if (containsAny(model, ["flux"])) {
+            return {
+                id: "flux-image",
+                label: "FLUX 图片模型",
+                promptShape: "自然语言段落 + 少量明确的视觉关键词，主体和动作放在前面。",
+                rules: ["用具体关系描述主体、位置、材质和光线，而不是重复质量形容词。", "保留用户指定的风格和构图，不主动添加模型参数。"],
+                avoid: ["不要输出很长的逗号标签清单或堆叠 negative prompt。", "不要添加未经用户要求的 LoRA、采样器或权重标记。"],
+            };
+        }
+        if (containsAny(model, ["sdxl", "stable-diffusion", "stable diffusion", "sd3", "stable3"])) {
+            return {
+                id: "stable-diffusion-image",
+                label: "Stable Diffusion / SDXL 图片模型",
+                promptShape: "紧凑的分层关键词或短句，按主体 → 构图 → 光线 → 风格与质量组织。",
+                rules: ["只保留能影响画面的关键词，主体和构图优先。", "如果原提示词已经使用括号权重，只在确有必要时保留并少量调整。", "将画面主体和反向规避项分开表达，避免互相冲突。"],
+                avoid: ["不要盲目添加过多质量词、艺术家名或模型专属标签。", "不确定权重语法是否被当前渠道支持时，不要主动新增。"],
+            };
+        }
+        if (containsAny(model, ["midjourney", "mid-journey"])) {
+            return {
+                id: "midjourney-image",
+                label: "Midjourney 图片模型",
+                promptShape: "精炼的视觉描述，主体、场景、构图和风格清晰，参数放在末尾。",
+                rules: ["优先写画面内容和视觉方向，保持短而有辨识度。", "只有用户原本提供或明确要求时才保留 --ar、--stylize 等参数。"],
+                avoid: ["不要臆造 Midjourney 参数或把所有摄影术语都堆上去。", "不要使用与当前工作台协议无关的命令前缀。"],
+            };
+        }
+        if (containsAny(model, ["ideogram", "recraft"])) {
+            return {
+                id: "design-image",
+                label: "设计与文字图片模型",
+                promptShape: "先说明画布、主体和版式，再说明文字内容、位置、层级、字体气质和配色。",
+                rules: ["画面文字保留精确原文并控制字数，明确文字与主体的空间关系。", "区分内容、版式和视觉风格，避免只给抽象审美词。"],
+                avoid: ["不要擅自改写品牌名、标题、数字或标语。"],
+            };
+        }
+        return genericImageProfile;
+    }
+
+    if (containsAny(model, ["seedance", "doubao-seedance", "volcengine-ark-video", "volcengine-jimeng-video", "jimeng-video"])) {
+        return {
+            id: "seedance-video",
+            label: "Seedance / 即梦视频模型",
+            promptShape: "单镜头时序描述，明确起始构图、动作节拍、镜头运动、环境变化和结尾状态。",
+            rules: ["用连续动作描述主体如何开始、发展和结束，保持镜头内的因果关系。", "人物、服装、场景和道具的一致性优先于堆叠风格词。", "有首帧或参考图时，明确哪些内容必须保持、哪些内容允许变化。"],
+            avoid: ["不要在一条提示词中拼接多个无关镜头。", "不要把静态图片参数当成视频动作指令。"],
+        };
+    }
+    if (containsAny(model, ["veo", "gemini-veo"])) {
+        return {
+            id: "veo-video",
+            label: "Veo 视频模型",
+            promptShape: "电影化的自然语言镜头描述，按主体、动作、镜头、环境和节奏组织。",
+            rules: ["把想法拆成可观察的动作和镜头运动，并补充时间顺序。", "用景别、机位、运镜、光线和环境声等可视化要素控制镜头。"],
+            avoid: ["不要只写情绪和抽象风格，不给动作或镜头变化。", "不要加入用户未要求的多镜头转场。"],
+        };
+    }
+    if (containsAny(model, ["kling"])) {
+        return {
+            id: "kling-video",
+            label: "Kling 视频模型",
+            promptShape: "简洁、连续的动作指令，明确主体运动和摄像机运动。",
+            rules: ["每个动作都写清主体、方向、速度或结果，避免互相冲突的动作。", "优先保证物理连续性、角色一致性和镜头稳定。"],
+            avoid: ["不要加入过多抽象叙事或无法在短片段中完成的事件。"],
+        };
+    }
+    if (containsAny(model, ["runway", "gen-3", "gen3", "minimax-video", "hailuo", "wan", "ltx-video", "hunyuan-video"])) {
+        return {
+            id: "general-video-family",
+            label: "通用视频模型",
+            promptShape: "以单个镜头为单位描述起始画面、动作过程、摄影机和结束画面。",
+            rules: ["动作使用连续动词，明确运动主体和摄像机，不把结果写成静态形容词。", "保持角色、服装、场景和光线的连续性。"],
+            avoid: ["不要堆砌静态图片关键词或一次安排过多动作。"],
+        };
+    }
+    return genericVideoProfile;
+}
+
+function buildUserMessage(input: PromptOptimizationInput, modelProfile: ModelAdaptationProfile | null = input.mode === "model-adapt" ? resolveModelAdaptationProfile(input) : null): PluginTextMessage {
+    const textParts = [
+        `优化模式：${modeLabels[input.mode]}`,
+        `目标类型：${input.generationMode === "image" ? "图片" : "视频"}`,
+        `目标生成模型：${input.targetModel?.trim() || "未指定"}`,
+        `目标接口协议：${input.targetProtocol?.trim() || "未指定"}`,
+        "",
+        "用户原始提示词：",
+        input.prompt.trim() || "（用户还没有写出具体提示词，请从参考上下文中提炼一个可执行版本）",
+    ];
+    if (modelProfile) {
+        textParts.push(
+            "",
+            `模型适配策略：${modelProfile.label}`,
+            `建议提示词结构：${modelProfile.promptShape}`,
+            "必须遵守：",
+            ...modelProfile.rules.map((rule) => `- ${rule}`),
+            "避免：",
+            ...modelProfile.avoid.map((rule) => `- ${rule}`),
+            "请在 changes 中明确说明已经采用的模型适配要点。",
+        );
+    }
+    if (input.context?.texts?.length) {
+        textParts.push("", "已连接的文本上下文：", ...input.context.texts.map((item) => `- ${item.title}：${item.text}`));
+    }
+    if (input.context?.images?.length) {
+        textParts.push("", "已连接的参考图已作为图片输入，请只在确实可见的内容上做判断：", ...input.context.images.map((item) => `- ${item.title}`));
+    }
+    textParts.push(
+        "",
+        "输出要求：",
+        "1. optimizedPrompt 是一段可以直接复制到当前生成节点的提示词。",
+        "2. negativePrompt 只填写明确有帮助的规避项，没有就留空。",
+        "3. changes 和 assumptions 使用简短中文条目。",
+        "4. variants 最多给出两个明显不同但仍忠于原意的版本，没有必要时返回空数组。",
+    );
+
+    const content: PluginTextContentPart[] = [{ type: "text", text: textParts.join("\n") }];
+    for (const image of input.context?.images || []) {
+        if (/^(data:|https?:\/\/)/i.test(image.url.trim())) content.push({ type: "image_url", image_url: { url: image.url.trim() } });
+    }
+    return { role: "user", content };
+}
+
+function parseJsonObject(value: string) {
+    const cleaned = value
+        .trim()
+        .replace(/^```(?:json)?\s*/i, "")
+        .replace(/\s*```$/, "");
+    const candidates = [cleaned];
+    const start = cleaned.indexOf("{");
+    const end = cleaned.lastIndexOf("}");
+    if (start >= 0 && end > start) candidates.push(cleaned.slice(start, end + 1));
+    for (const candidate of candidates) {
+        try {
+            const parsed: unknown = JSON.parse(candidate);
+            if (parsed && typeof parsed === "object" && !Array.isArray(parsed)) return parsed as Record<string, unknown>;
+        } catch {
+            // 某些兼容模型会在 JSON 外包一层解释，继续尝试截取对象部分。
+        }
+    }
+    return null;
+}
+
+function stringArray(value: unknown) {
+    return Array.isArray(value)
+        ? value
+              .filter((item): item is string => typeof item === "string")
+              .map((item) => item.trim())
+              .filter(Boolean)
+        : [];
+}
+
+function variants(value: unknown): PromptOptimizationVariant[] {
+    if (!Array.isArray(value)) return [];
+    return value
+        .flatMap((item): PromptOptimizationVariant[] => {
+            if (!item || typeof item !== "object") return [];
+            const record = item as Record<string, unknown>;
+            const label = typeof record.label === "string" ? record.label.trim() : "备选版本";
+            const prompt = typeof record.prompt === "string" ? record.prompt.trim() : "";
+            return prompt ? [{ label: label || "备选版本", prompt }] : [];
+        })
+        .slice(0, 2);
+}
+
+function normalizeResult(value: Record<string, unknown> | null, sourcePrompt: string, modelProfile?: ModelAdaptationProfile | null): PromptOptimizationResult {
+    const optimizedPrompt = typeof value?.optimizedPrompt === "string" ? value.optimizedPrompt.trim() : "";
+    const negativePrompt = typeof value?.negativePrompt === "string" ? value.negativePrompt.trim() : "";
+    return {
+        optimizedPrompt: optimizedPrompt || sourcePrompt.trim(),
+        negativePrompt,
+        changes: stringArray(value?.changes),
+        assumptions: stringArray(value?.assumptions),
+        variants: variants(value?.variants),
+        modelProfile: modelProfile ? { id: modelProfile.id, label: modelProfile.label } : undefined,
+    };
+}
+
+function createPromptOptimizer(context: Parameters<NonNullable<RegisteredPlugin["createPromptOptimizer"]>>[0]): PromptOptimizerProvider {
+    const textService = context.services?.ai?.text;
+    if (!textService) throw new Error("提示词优化器暂未获得文本模型服务");
+
+    return {
+        optimize: async (input, options) => {
+            const modelProfile = input.mode === "model-adapt" ? resolveModelAdaptationProfile(input) : null;
+            const response = await textService.requestToolResponse({
+                model: input.optimizerModel,
+                messages: [{ role: "system", content: systemPrompt }, buildUserMessage(input, modelProfile)],
+                tools: [optimizerTool],
+                toolChoice: { type: "function", name: "optimize_prompt" },
+                signal: options?.signal,
+                onDelta: options?.onDelta,
+            });
+            const toolCall = response.toolCalls.find((call) => call.name === "optimize_prompt");
+            const parsed = parseJsonObject(toolCall?.arguments || response.content);
+            return normalizeResult(parsed, input.prompt, modelProfile);
+        },
+    };
+}
+
+export const promptOptimizerPlugin: RegisteredPlugin = {
+    manifest,
+    createPromptOptimizer,
+};
+
+registerPlugin(promptOptimizerPlugin);

web/src/lib/plugins/plugin-types.ts

@@ -24,6 +24,7 @@ export type PluginPermission =
     | "asset.import"
     | "asset.upload"
     | "generation.run"
+    | "ai.text"
     | "external.open";
 
 export type PluginManifest = {
@@ -54,11 +55,98 @@ export type PluginStorage = {
     remove(key: string): Promise<void>;
 };
 
+export type PluginTextContentPart =
+    | { type: "text"; text: string }
+    | { type: "image_url"; image_url: { url: string } };
+
+export type PluginTextMessage = {
+    role: "system" | "user" | "assistant";
+    content: string | PluginTextContentPart[];
+};
+
+export type PluginTextTool = {
+    type: "function";
+    function: {
+        name: string;
+        description?: string;
+        parameters: Record<string, unknown>;
+        strict?: boolean;
+    };
+};
+
+export type PluginTextToolChoice = "auto" | "required" | { type: "function"; name: string };
+
+export type PluginTextToolCall = {
+    name: string;
+    arguments: string;
+};
+
+export type PluginTextResponse = {
+    content: string;
+    toolCalls: PluginTextToolCall[];
+};
+
+export type PluginTextRequest = {
+    model?: string;
+    messages: PluginTextMessage[];
+    tools?: PluginTextTool[];
+    toolChoice?: PluginTextToolChoice;
+    signal?: AbortSignal;
+    onDelta?: (text: string) => void;
+};
+
+export type PluginAiTextService = {
+    requestToolResponse: (request: PluginTextRequest) => Promise<PluginTextResponse>;
+};
+
+export type PluginHostServices = {
+    ai?: {
+        text?: PluginAiTextService;
+    };
+};
+
 export type PluginHostContext = {
     manifest: PluginManifest;
     permissions: ReadonlySet<PluginPermission>;
     storage: PluginStorage;
     config: Readonly<PluginInstallation["config"]>;
+    services?: PluginHostServices;
+};
+
+export type PromptOptimizationMode = "expand" | "refine" | "style" | "model-adapt" | "reference";
+
+export type PromptOptimizationInput = {
+    prompt: string;
+    mode: PromptOptimizationMode;
+    generationMode: "image" | "video";
+    targetModel?: string;
+    targetProtocol?: string;
+    optimizerModel?: string;
+    context?: {
+        texts?: Array<{ title: string; text: string }>;
+        images?: Array<{ title: string; url: string }>;
+    };
+};
+
+export type PromptOptimizationVariant = {
+    label: string;
+    prompt: string;
+};
+
+export type PromptOptimizationResult = {
+    optimizedPrompt: string;
+    negativePrompt: string;
+    changes: string[];
+    assumptions: string[];
+    variants: PromptOptimizationVariant[];
+    modelProfile?: { id: string; label: string };
+};
+
+export type PromptOptimizerProvider = {
+    optimize: (
+        input: PromptOptimizationInput,
+        options?: { signal?: AbortSignal; onDelta?: (text: string) => void },
+    ) => Promise<PromptOptimizationResult>;
 };
 
 export type AssetSourceQuery = {
@@ -118,6 +206,7 @@ export type RegisteredPlugin = {
     activate?: (context: PluginHostContext) => Promise<void> | void;
     deactivate?: (context: PluginHostContext) => Promise<void> | void;
     createAssetSource?: (context: PluginHostContext) => AssetSourceProvider;
+    createPromptOptimizer?: (context: PluginHostContext) => PromptOptimizerProvider;
 };
 
 export type PluginInstallation = {