989 lines
37 KiB
TypeScript
989 lines
37 KiB
TypeScript
import { Knex } from "knex";
|
||
import { v4 as uuid } from "uuid";
|
||
import { getEmbedding } from "@/utils/agent/embedding";
|
||
|
||
interface TableSchema {
|
||
name: string;
|
||
builder: (table: Knex.CreateTableBuilder) => void;
|
||
initData?: (knex: Knex) => Promise<void>;
|
||
}
|
||
|
||
export default async (knex: Knex, forceInit: boolean = false): Promise<void> => {
|
||
const tables: TableSchema[] = [
|
||
// 用户表
|
||
{
|
||
name: "o_user",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.text("name");
|
||
table.text("password");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
initData: async (knex) => {
|
||
await knex("o_user").insert([{ id: 1, name: "admin", password: "admin123" }]);
|
||
},
|
||
},
|
||
//项目表
|
||
{
|
||
name: "o_project",
|
||
builder: (table) => {
|
||
table.integer("id");
|
||
table.string("projectType");
|
||
table.string("imageModel");
|
||
table.string("imageQuality");
|
||
table.string("videoModel");
|
||
table.text("name");
|
||
table.text("intro");
|
||
table.text("type");
|
||
table.text("artStyle");
|
||
table.text("mode");
|
||
table.text("videoRatio");
|
||
table.integer("createTime");
|
||
table.integer("userId");
|
||
table.primary(["id"]);
|
||
},
|
||
},
|
||
//风格表
|
||
{
|
||
name: "o_artStyle",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.string("name");
|
||
table.text("fileUrl");
|
||
table.text("label");
|
||
table.text("prompt");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
initData: async (knex) => {},
|
||
},
|
||
//Agent配置表
|
||
{
|
||
name: "o_agentDeploy",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.string("model");
|
||
table.string("key");
|
||
table.string("modelName");
|
||
table.text("vendorId");
|
||
table.string("desc");
|
||
table.string("name");
|
||
table.boolean("disabled").defaultTo(false);
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
initData: async (knex) => {
|
||
await knex("o_agentDeploy").insert([
|
||
{
|
||
model: "",
|
||
modelName: "",
|
||
vendorId: null,
|
||
key: "scriptAgent",
|
||
name: "剧本AI",
|
||
desc: "用于读取原文生成故事骨架、改编策略，建议使用具备强大文本理解和生成能力的模型",
|
||
disabled: false,
|
||
},
|
||
{
|
||
model: "",
|
||
modelName: "",
|
||
vendorId: null,
|
||
key: "productionAgent",
|
||
name: "生产AI",
|
||
desc: "对工作流进行调度和管理，建议使用具备较强的逻辑推理和任务管理能力的模型",
|
||
disabled: false,
|
||
},
|
||
{
|
||
model: "",
|
||
modelName: "",
|
||
vendorId: null,
|
||
key: "universalAi",
|
||
name: "通用AI",
|
||
desc: "用于小说事件提取、资产提示词生成、台词提取等边缘功能，建议使用具备较强文本处理能力的模型",
|
||
disabled: false,
|
||
},
|
||
{
|
||
model: "",
|
||
modelName: "",
|
||
vendorId: null,
|
||
key: "ttsDubbing",
|
||
name: "TTS配音",
|
||
desc: "根据剧本内容生成角色配音，支持多种声音风格和情绪",
|
||
disabled: true,
|
||
},
|
||
]);
|
||
},
|
||
},
|
||
//设置表
|
||
{
|
||
name: "o_setting",
|
||
builder: (table) => {
|
||
table.text("key");
|
||
table.text("value");
|
||
table.primary(["key"]);
|
||
table.unique(["key"]);
|
||
},
|
||
initData: async (knex) => {
|
||
await knex("o_setting").insert([
|
||
{
|
||
key: "tokenKey",
|
||
value: uuid().slice(0, 8),
|
||
},
|
||
{
|
||
key: "messagesPerSummary",
|
||
value: 10,
|
||
},
|
||
{
|
||
key: "shortTermLimit",
|
||
value: 5,
|
||
},
|
||
{
|
||
key: "summaryMaxLength",
|
||
value: 500,
|
||
},
|
||
{
|
||
key: "summaryLimit",
|
||
value: 10,
|
||
},
|
||
{
|
||
key: "ragLimit",
|
||
value: 3,
|
||
},
|
||
{
|
||
key: "deepRetrieveSummaryLimit",
|
||
value: 5,
|
||
},
|
||
{
|
||
key: "modelOnnxFile",
|
||
value: '["all-MiniLM-L6-v2", "onnx", "model_fp16.onnx"]',
|
||
},
|
||
{
|
||
key: "modelDtype",
|
||
value: "fp16",
|
||
},
|
||
{
|
||
key: "switchAiDevTool",
|
||
value: "0",
|
||
},
|
||
]);
|
||
},
|
||
},
|
||
//任务中心表
|
||
{
|
||
name: "o_tasks",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.integer("projectId");
|
||
table.string("taskClass");
|
||
table.string("relatedObjects");
|
||
table.string("model");
|
||
table.text("describe");
|
||
table.string("state");
|
||
table.integer("startTime");
|
||
table.text("reason");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
initData: async (knex) => {},
|
||
},
|
||
//提示词表
|
||
{
|
||
name: "o_prompt",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.string("name");
|
||
table.string("type");
|
||
table.text("data");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
initData: async (knex) => {
|
||
await knex("o_prompt").insert([
|
||
{
|
||
name: "事件提取",
|
||
type: "eventExtraction",
|
||
data: `# 事件提取指令
|
||
|
||
你是小说文本分析助手。用户每次提供一个章节的原文，你提取该章的结构化事件信息。
|
||
|
||
## ⚠️ 输出约束（最高优先级，违反任何一条即为失败）
|
||
|
||
1. 你的**完整回复**只有一行，以 \`|\` 开头、以 \`|\` 结尾，恰好 7 个字段
|
||
2. 回复的**第一个字符**必须是 \`|\`，**最后一个字符**必须是 \`|\`
|
||
3. \`|\` 之前不许有任何字符——没有引导语、没有解释、没有"根据……"、没有"以下是……"
|
||
4. \`|\` 之后不许有任何字符——没有总结、没有提取说明、没有改编建议
|
||
5. 不输出表头行、分隔线、Markdown 标题、emoji、代码块标记
|
||
|
||
## 输出格式
|
||
|
||
\`\`\`
|
||
| 第X章 {章节标题} | {涉及角色} | {核心事件} | {主线关系} | {信息密度} | {预估集长} | {情绪强度} |
|
||
\`\`\`
|
||
|
||
### 字段规范
|
||
|
||
| 字段 | 格式要求 | 示例 |
|
||
|------|----------|------|
|
||
| 章节 | \`第X章 {章节标题}\` | \`第1章 职业危机与许愿\` |
|
||
| 涉及角色 | 有实际戏份的角色，顿号分隔 | \`林逸、白有容\` |
|
||
| 核心事件 | 30-60字，必须含动作+结果 | \`林逸因解密风潮事业崩塌，颓废中许愿触发魔法系统绑定\` |
|
||
| 主线关系 | **必须**为 \`强/中/弱（3-8字理由）\` | \`强（动机建立+系统激活）\` |
|
||
| 信息密度 | \`高\` / \`中\` / \`低\` | \`高\` |
|
||
| 预估集长 | **必须**为 \`X秒\`，禁止用分钟 | \`50秒\` |
|
||
| 情绪强度 | 文字标签，\`+\` 连接，禁止星级/数字 | \`转折+悬疑\` |
|
||
|
||
**主线关系判定**：强＝直接推动主角弧线；中＝补充世界观/人物关系/伏笔；弱＝过渡/气氛。
|
||
|
||
**预估集长参考**：高密度+高情绪→45-60秒；中→35-45秒；低→25-35秒。
|
||
|
||
**可用情绪标签**：\`冲突\`、\`恐怖\`、\`情感\`、\`转折\`、\`高潮\`、\`平铺\`、\`喜剧\`、\`悬疑\`、\`情感崩溃\`。
|
||
|
||
## 输出示例
|
||
|
||
以下两个示例展示的是**完整回复**——除这一行外没有任何其他内容：
|
||
|
||
\`\`\`
|
||
| 第1章 职业危机与许愿 | 林逸 | 职业魔术师林逸因解密打假风潮导致事业崩塌，颓废中感慨"如果会魔法就好了"，意外触发神奇魔法系统绑定 | 强（主角动机建立+系统激活） | 高 | 50秒 | 转折+悬疑 |
|
||
\`\`\`
|
||
\`\`\`
|
||
| 第12章 山间小憩 | 凌玄、苏晚卿 | 凌玄与苏晚卿在山间歇脚，苏晚卿回忆幼时往事，两人关系略有缓和但未实质推进 | 弱（气氛过渡） | 低 | 25秒 | 平铺+情感 |
|
||
\`\`\`
|
||
|
||
## 提取规则
|
||
|
||
- 忠于原文，不推测、不脑补、不加入原文未出现的情节
|
||
- 角色使用文中主要称呼，保持一致
|
||
- 多条平行事件线时，选对主角影响最大的一条，其余简要带过
|
||
- 对话密集章节，关注对话推动了什么结果，而非复述对话内容`,
|
||
},
|
||
{
|
||
name: "资产提示词生成",
|
||
type: "assetsPromptGeneration",
|
||
data: "# 资产提示词生成指令 根据提供的项目参数和资产设定，生成符合要求的提示词\n\n请根据以下参数生成提示词：\n\n**基础参数：**\n- 风格: {风格}\n- 小说类型: {小说类型}\n- 小说背景: {小说背景}\n\n**资产设定：**\n- 类型: {角色/场景/道具}\n- 名称:{名称}\n- 描述:{描述}\n\n请严格按照skill规范生成提示词。",
|
||
},
|
||
{
|
||
name: "剧本资产提取",
|
||
type: "scriptAssetExtraction",
|
||
data: `---
|
||
name: universal_agent
|
||
description: 专注于从剧本内容中提取所使用的资产（角色、场景、道具）并生成结构化资产列表的助手。
|
||
---
|
||
|
||
# Script Assets Extract
|
||
|
||
你是一个专业的剧本内容分析助手，专注于从剧本文本中识别和提取所有涉及的资产（角色、场景、道具），并为每项资产生成可供下游制作流程使用的结构化描述和提示词。
|
||
|
||
## 何时使用
|
||
|
||
用户提供剧本内容，你需要逐段阅读并提取其中涉及的所有资产（人物角色、场景地点、道具物件），输出为结构化的资产列表。产出的资产描述将用于后续 AI 图片生成和制作流程。
|
||
|
||
## 与系统的对应关系
|
||
|
||
- 资产类型：
|
||
- \`role\` — 角色（对应 \`o_assets.type = "role"\`）
|
||
- \`scene\` — 场景（对应 \`o_assets.type = "scene"\`）
|
||
- \`tool\` — 道具（对应 \`o_assets.type = "tool"\`）
|
||
- 下游用途：资产提示词生成 → AI 资产图生成 → 分镜制作
|
||
|
||
## 输出要求
|
||
|
||
**必须通过调用 \`resultTool\` 工具返回结果**，禁止以纯文本、Markdown 表格或 JSON 代码块等形式直接输出资产列表。
|
||
\`resultTool\` 的 schema 会对字段类型和枚举值做强校验，调用时请严格按照下方字段定义填写，确保数据结构正确、字段完整、类型匹配。
|
||
|
||
每个资产对象包含以下字段：
|
||
|
||
| 字段 | 类型 | 必填 | 说明 |
|
||
| ---- | ---- | ---- | ---- |
|
||
| \`name\` | string | 是 | 资产名称，使用剧本中的原始称呼,不做其他多余描述 |
|
||
| \`desc\` | string | 是 | 资产描述，30-80 字的视觉化描述 |
|
||
| \`prompt\` | string | 是 | 生成提示词，英文，用于 AI 图片生成 |
|
||
| \`type\` | enum | 是 | 资产类型：\`role\` / \`scene\` / \`tool\` |
|
||
|
||
## 提取规则
|
||
|
||
### 角色（role）
|
||
|
||
- 提取剧本中出现的所有有名字的角色
|
||
- \`desc\`：包含外貌特征、服饰风格、体态气质等视觉要素
|
||
- \`prompt\`：英文提示词，描述角色的外观特征，适用于 AI 角色图生成
|
||
- 同一角色有多个称呼时，取最常用的作为 \`name\`
|
||
- 无名龙套（如"路人甲"、"士兵"）可跳过，除非其造型对剧情有重要视觉意义
|
||
|
||
### 场景（scene）
|
||
|
||
- 提取剧本中出现的所有场景/地点
|
||
- \`desc\`：包含空间结构、光照氛围、关键陈设、色调基调等视觉要素
|
||
- \`prompt\`：英文提示词，描述场景的整体视觉风格，适用于 AI 场景图生成
|
||
- 同一场景的不同状态（如白天/夜晚）不重复提取，在 \`desc\` 中注明即可
|
||
|
||
### 道具（tool）
|
||
|
||
- 提取剧本中出现的重要道具/物品
|
||
- \`desc\`：包含外观形状、颜色材质、尺寸参考、特殊效果等视觉要素
|
||
- \`prompt\`：英文提示词，描述道具的外观细节，适用于 AI 道具图生成
|
||
- 仅提取有独立视觉意义或剧情功能的道具，通用物品可跳过
|
||
|
||
|
||
## 提示词（prompt）生成规范
|
||
|
||
- 采用逗号分隔的关键词/短语格式
|
||
- 优先描述**视觉特征**，避免抽象概念
|
||
- 包含风格关键词（如 anime style, manga style 等，根据项目风格决定）
|
||
- 角色 prompt 示例：\`a young man, sharp eyebrows, black hair, pale skin, wearing a gray Taoist robe, slender build, cold expression\`
|
||
- 场景 prompt 示例：\`dark cave interior, glowing crystals on walls, misty atmosphere, dim blue lighting, stone altar in center\`
|
||
- 道具 prompt 示例：\`ancient jade pendant, oval shape, translucent green, carved dragon pattern, glowing faintly\`
|
||
|
||
## 提取流程
|
||
|
||
1. 通读剧本全文，识别所有出现的角色、场景、道具
|
||
2. 对每个资产生成结构化的 \`name\`、\`desc\`、\`prompt\`、\`type\`
|
||
3. 去重：同一资产不重复提取
|
||
4. **必须通过调用 \`resultTool\` 工具输出完整资产列表**，不要分多次调用，一次性将所有资产放入 \`assetsList\` 数组中提交
|
||
|
||
## 提取原则
|
||
|
||
1. **忠于剧本**：所有提取基于剧本中的实际内容，不臆造未出现的资产
|
||
2. **视觉优先**：描述和提示词聚焦视觉特征，便于 AI 图片生成
|
||
3. **精简实用**：只提取对制作有实际意义的资产，避免过度提取
|
||
4. **分类准确**：严格按照 role/scene/tool 分类，不混淆
|
||
5. **提示词质量**：英文提示词应具体、可执行，能直接用于 AI 图片生成
|
||
|
||
## 注意事项
|
||
|
||
- 资产列表中**不要包含剧本内容本身**，仅提取所使用到的资产
|
||
- 角色的随身物品如果有独立剧情功能，应单独作为道具提取
|
||
- 场景中的固定陈设不需要单独提取为道具，除非该物件有独立剧情作用
|
||
`,
|
||
},
|
||
]);
|
||
},
|
||
},
|
||
//小说原文表
|
||
{
|
||
name: "o_novel",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.integer("chapterIndex");
|
||
table.text("reel");
|
||
table.text("chapter");
|
||
table.text("chapterData");
|
||
table.integer("projectId");
|
||
table.integer("eventState");
|
||
table.text("event");
|
||
table.text("errorReason");
|
||
table.integer("createTime");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
//小说事件表
|
||
{
|
||
name: "o_event",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.string("name");
|
||
table.string("detail");
|
||
table.integer("createTime");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
//事件-章节表
|
||
{
|
||
name: "o_eventChapter",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.integer("eventId").unsigned().references("id").inTable("o_event");
|
||
table.integer("novelId").unsigned().references("id").inTable("o_novel");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
//大纲表
|
||
{
|
||
name: "o_outline",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.integer("episode");
|
||
table.text("data");
|
||
table.integer("projectId");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
//大纲-原文表
|
||
{
|
||
name: "o_outlineNovel",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.integer("outlineId").unsigned().references("id").inTable("o_outline");
|
||
table.integer("novelId").unsigned().references("id").inTable("o_novel");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
//剧本
|
||
{
|
||
name: "o_script",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.text("name");
|
||
table.text("content");
|
||
table.integer("projectId");
|
||
table.integer("extractState");
|
||
table.integer("createTime");
|
||
table.text("errorReason");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
//资产表
|
||
{
|
||
name: "o_assets",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.text("name");
|
||
table.text("prompt");
|
||
table.text("remark");
|
||
table.text("type");
|
||
table.text("describe");
|
||
table.integer("scriptId"); //剧本id
|
||
table.integer("imageId").unsigned().references("id").inTable("o_image");
|
||
table.integer("assetsId");
|
||
table.integer("projectId");
|
||
table.integer("startTime");
|
||
table.string("promptState");
|
||
table.text("promptErrorReason");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
initData: async (knex) => {},
|
||
},
|
||
//生成图片表
|
||
{
|
||
name: "o_image",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.text("filePath");
|
||
table.text("type");
|
||
table.integer("assetsId");
|
||
table.text("model");
|
||
table.text("resolution");
|
||
table.text("state");
|
||
table.text("errorReason");
|
||
table.text("reason");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
//分镜
|
||
{
|
||
name: "o_storyboard",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.integer("scriptId");
|
||
table.text("prompt");
|
||
table.text("filePath");
|
||
table.text("duration");
|
||
table.text("state");
|
||
table.integer("trackId");
|
||
table.text("reason");
|
||
table.integer("projectId");
|
||
table.integer("index");
|
||
table.integer("createTime");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
//flowData-剧本
|
||
{
|
||
name: "o_agentWorkData",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.integer("projectId");
|
||
table.integer("episodesId");
|
||
table.string("key"); //用户其他方式索引
|
||
table.string("data");
|
||
table.integer("createTime");
|
||
table.integer("updateTime");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
//视频
|
||
{
|
||
name: "o_video",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.text("filePath");
|
||
table.text("errorReason");
|
||
table.integer("time");
|
||
table.text("state");
|
||
table.integer("scriptId");
|
||
table.integer("projectId");
|
||
table.integer("videoTrackId");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
// 视频轨道
|
||
{
|
||
name: "o_videoTrack",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.integer("videoId");
|
||
table.integer("projectId");
|
||
table.integer("scriptId");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
//供应商配置表
|
||
{
|
||
name: "o_vendorConfig",
|
||
builder: (table) => {
|
||
table.string("id").notNullable();
|
||
table.text("author");
|
||
table.text("description");
|
||
table.text("name");
|
||
table.text("icon");
|
||
table.text("inputs"); // 输入项配置 JSON
|
||
table.text("inputValues"); // 输入项值 JSON
|
||
table.text("models"); // 模型配置 JSON
|
||
table.text("code"); // 模型配置 JSON
|
||
table.integer("createTime");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
//图片工作流表
|
||
{
|
||
name: "o_imageFlow",
|
||
builder: (table) => {
|
||
table.integer("id").notNullable();
|
||
table.text("flowData").notNullable();
|
||
table.integer("storyboardId");
|
||
table.integer("assetsId");
|
||
table.primary(["id"]);
|
||
table.unique(["id"]);
|
||
},
|
||
},
|
||
{
|
||
name: "o_assets2Storyboard",
|
||
builder: (table) => {
|
||
table.integer("storyboardId").notNullable();
|
||
table.integer("assetId").notNullable();
|
||
table.primary(["storyboardId", "assetId"]);
|
||
table.unique(["storyboardId", "assetId"]);
|
||
},
|
||
},
|
||
{
|
||
name: "o_scriptAssets",
|
||
builder: (table) => {
|
||
table.integer("scriptId").notNullable();
|
||
table.integer("assetId").notNullable();
|
||
table.primary(["scriptId", "assetId"]);
|
||
table.unique(["scriptId", "assetId"]);
|
||
},
|
||
},
|
||
{
|
||
name: "o_skillList",
|
||
builder: (table) => {
|
||
table.text("id").notNullable();
|
||
table.text("md5").notNullable();
|
||
table.text("path").notNullable();
|
||
table.text("name").notNullable(); //文件名
|
||
table.text("description").notNullable(); //描述
|
||
table.text("embedding"); // 向量嵌入 JSON
|
||
table.text("type").notNullable(); // "main" | "references"
|
||
table.integer("createTime").notNullable();
|
||
table.integer("updateTime").notNullable();
|
||
table.integer("state").notNullable(); // 1正常，0正在生成description，-1description为空。-2归属为空,-3md5变动，-4文件不存在
|
||
table.primary(["id"]);
|
||
},
|
||
initData: async (knex) => {
|
||
const list = [
|
||
{
|
||
id: "4fb36012e56e395b425569987f5dab0e",
|
||
md5: "fca3c269c5f325a65dafa663c9bb9773",
|
||
path: "production_agent_decision.md",
|
||
name: "production_agent_decision",
|
||
description: "",
|
||
embedding: "",
|
||
type: "main",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774447310118,
|
||
state: -1,
|
||
},
|
||
{
|
||
id: "017b6338d7aa227cd614ec1fb25fd83e",
|
||
md5: "2610b80abe4bd048fe61c73adc7388ac",
|
||
path: "production_agent_execution.md",
|
||
name: "production_agent_execution",
|
||
description: "",
|
||
embedding: "",
|
||
type: "main",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774447310118,
|
||
state: -1,
|
||
},
|
||
{
|
||
id: "f03c8e67b61580de9ea5b9d166521b67",
|
||
md5: "d41d8cd98f00b204e9800998ecf8427e",
|
||
path: "production_agent_supervision.md",
|
||
name: "production_agent_supervision",
|
||
description: "",
|
||
embedding: "",
|
||
type: "main",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774447310118,
|
||
state: -1,
|
||
},
|
||
{
|
||
id: "50b49d8af5d364665b463c23f6a4d8bb",
|
||
md5: "fbba66e0df2426996277b299710c3033",
|
||
path: "script_agent_decision.md",
|
||
name: "script_agent_decision",
|
||
description: "",
|
||
embedding: "",
|
||
type: "main",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774447310118,
|
||
state: -1,
|
||
},
|
||
{
|
||
id: "427727727e1095c54b6840cd21382d82",
|
||
md5: "7e5911242af7233854d533278c6a8ccb",
|
||
path: "script_agent_execution.md",
|
||
name: "script_agent_execution",
|
||
description: "",
|
||
embedding: "",
|
||
type: "main",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774447310118,
|
||
state: -1,
|
||
},
|
||
{
|
||
id: "02848fb0dd582fd926502c77ecf9679c",
|
||
md5: "7a8b6a311b015cd47bf17cc52b935348",
|
||
path: "script_agent_supervision.md",
|
||
name: "script_agent_supervision",
|
||
description: "",
|
||
embedding: "",
|
||
type: "main",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774447310118,
|
||
state: -1,
|
||
},
|
||
{
|
||
id: "a1e818cc03a0b355b239ac1fb0512969",
|
||
md5: "1fd22029e8047aa30b0dfd703cb837ed",
|
||
path: "universal_agent.md",
|
||
name: "universal_agent",
|
||
description: "",
|
||
embedding: "",
|
||
type: "main",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774447310118,
|
||
state: -1,
|
||
},
|
||
{
|
||
id: "3e5efec258c8d8e6a39bcef12f8ee058",
|
||
md5: "efccb0464cfd472861b49ebf737d4820",
|
||
path: "references/event_extract.md",
|
||
name: "event_extract",
|
||
description:
|
||
"专为小说改编短剧设计的文本分析助手，逐章提取涉及角色、核心事件、主线关系、信息密度、预估集长及情绪强度等结构化信息，以Markdown表格形式输出，并附汇总统计，辅助短剧制作的内容规划与时长估算。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774450165911,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "52c51fa8655f899a1b7aae9b6aad7251",
|
||
md5: "783678aaab829b34e7c30a414c356bf6",
|
||
path: "references/novel_character_extract.md",
|
||
name: "novel_character_extract",
|
||
description:
|
||
"专为小说内容分析设计的角色提取助手，从原文中识别并结构化输出所有重要角色的视觉描述信息，包括外貌、服饰、体态、状态变体等字段，供美术制作和AI角色图生成使用。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774450080903,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "6d46cdca10b2f49e07e515885d1387a0",
|
||
md5: "10544d12c4ef011e6b3b63a99b8c7fa8",
|
||
path: "references/novel_props_extract.md",
|
||
name: "novel_props_extract",
|
||
description:
|
||
"专注于从小说原文中提取道具物品信息的分析助手，能识别武器、法器、药物等各类道具，生成包含外观、材质、尺寸、功能及状态变体的结构化视觉描述表格，供美术制作和AI绘图使用。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774450094771,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "1864df75d1d65f76e275046649ecaef8",
|
||
md5: "65603aa495a541f54c55b7f30e149f45",
|
||
path: "references/novel_scene_extract.md",
|
||
name: "novel_scene_extract",
|
||
description:
|
||
"专注于从小说原文中提取并结构化场景信息的分析助手，可识别各类场景地点，输出包含空间描述、光照氛围、关键陈设、色调基调等字段的标准化场景资产表，用于美术制作和AI绘图的场景概念图生成。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774450161878,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "7fbce6f90d7d85496ba9817e9622e640",
|
||
md5: "830559e8f2cd5d0fa8e6df48a164fe2d",
|
||
path: "references/video_dialogue_extract.md",
|
||
name: "video_dialogue_extract",
|
||
description:
|
||
"这是一个专门从视频分镜提示词中提取结构化台词、旁白与音效信息的AI助手配置文档，定义了完整的输出格式（含镜号、角色、台词类型、表演指导等字段）、提取规则及处理流程，用于将视频分镜描述转化为标准化台词表。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774447310118,
|
||
updateTime: 1774450180712,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "31fb5c5a1f514ec1e66b4eba9f22d4db",
|
||
md5: "43e63450efe0c9af8a3a40b036d36cb4",
|
||
path: "references/pipeline.md",
|
||
name: "pipeline",
|
||
description:
|
||
"面向短剧改编项目的四阶段流水线说明文档，涵盖事件提取、故事骨架、改编策略、剧本编写的串行执行流程，定义了决策层、执行层、监督层的协作规范及派发、审核、修复的交互格式与质量门控标准。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774451946248,
|
||
updateTime: 1774451984533,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "27dc2dfc901de2180227d0269217583a",
|
||
md5: "7d353be4bab7a794436d9abff2b9c6ee",
|
||
path: "references/adaptation_format.md",
|
||
name: "adaptation_format",
|
||
description:
|
||
"本文档规定了改编策略输出的标准格式，包括核心改编原则、删除决策和世界观呈现策略三大模块的书写规范，明确各模块所需涵盖的维度与要素，用于指导竖屏短剧等载体的文学改编工作。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774452010535,
|
||
updateTime: 1774452022083,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "d49fa09504fe784a8e6eb102756c6d56",
|
||
md5: "2ef08a7479f29d74986999ceb02092c8",
|
||
path: "references/event_format.md",
|
||
name: "event_format",
|
||
description:
|
||
"本文档规定了影视改编项目中事件表的标准输出格式，包括文件头、事件表格、各字段填写规范（章节、角色、核心事件、主线关系、情绪强度、预估时长）及汇总统计模板，用于指导从原著提取事件并评估改编集数与压缩比的第一阶段工作。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774452010535,
|
||
updateTime: 1774452030858,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "797906c2ddf0750f050bcdeae23eae3d",
|
||
md5: "f5e7fe6db7e05db69d5dc327c4c538f2",
|
||
path: "references/script_format.md",
|
||
name: "script_format",
|
||
description:
|
||
"本文档为竖屏短剧剧本的输出格式规范，定义了文件头、节拍结构、分镜脚本、画面描述、台词、转场标注等标准格式要求，并附有时长控制参数与自查清单，供AI视频生成和导演制作使用。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774452010535,
|
||
updateTime: 1774452042934,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "1abd8675c0c3e62b20c0b151d2ec0fb1",
|
||
md5: "a587532c737ce15022e1522021f099bb",
|
||
path: "references/skeleton_format.md",
|
||
name: "skeleton_format",
|
||
description:
|
||
"本文档定义了故事骨架文件（skeleton.md）的标准化输出格式，涵盖故事核、人物成长隐线、三幕结构、分集决策模板、全局删减记录、付费卡点设计及自查清单，用于指导编剧将章节事件列表转化为结构完整的剧集改编方案。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774452010535,
|
||
updateTime: 1774452057184,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "0b7828d7a6ab458a4b201122f08d6c16",
|
||
md5: "120b3c856f1b2a8a429e11319e8c95fe",
|
||
path: "references/quality_criteria.md",
|
||
name: "quality_criteria",
|
||
description:
|
||
"本文档为影视/短剧项目的质量审核标准手册，涵盖事件表、故事骨架、改编策略和剧本四大模块的详细审核规则，规定了格式规范、角色名称统一、时长合理性、画面可执行性及场景氛围一致性等审核要求，用于确保各阶段产出物的内容准确性与制作可行性。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774452068093,
|
||
updateTime: 1774452087877,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "5c1772b5f9c420d9eae9ca02914ba087",
|
||
md5: "c710ab7d237e1f0c5aa3d208e0f5b484",
|
||
path: "references/plan.md",
|
||
name: "plan",
|
||
description:
|
||
"该文档定义了AI代理生成执行计划的规范，包括任务总览、步骤列表（含编号、名称、详细内容、预期输出及依赖关系）和执行顺序标注，并提供标准回复模板，用于将用户需求拆解为可直接传入子代理工具执行的具体步骤。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774452098447,
|
||
updateTime: 1774452109574,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "75a45cf996015ca819582873887ec301",
|
||
md5: "6045d76873fd58b8b87a914a21a38439",
|
||
path: "references/derive_assets_extraction.md",
|
||
name: "derive_assets_extraction",
|
||
description:
|
||
"本文档是一份技术操作指南，说明如何根据剧本内容和已有资产列表，提取每个资产在剧情中出现的不同视觉状态变体（derive），并通过工具函数读取和写入数据，用于后续图片生成参考。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774452119499,
|
||
updateTime: 1774452129516,
|
||
state: 1,
|
||
},
|
||
{
|
||
id: "fce75f69d704c19bebcb356bc1bd6e81",
|
||
md5: "a3b3432854970f22949ba47236a6532f",
|
||
path: "references/storyboard_generation.md",
|
||
name: "storyboard_generation",
|
||
description:
|
||
"根据剧本和资产列表生成结构化分镜面板的工具指南，涵盖分镜拆分原则、字段填写规范及工具调用流程，用于将剧本转化为含画面描述、镜头语言、台词和AI绘图提示词的分镜数据。",
|
||
embedding: "",
|
||
type: "references",
|
||
createTime: 1774452119499,
|
||
updateTime: 1774452140873,
|
||
state: 1,
|
||
},
|
||
];
|
||
await Promise.all(
|
||
list.map(async (item) => {
|
||
const embedding = await getEmbedding(item.description);
|
||
item.embedding = JSON.stringify(embedding);
|
||
}),
|
||
);
|
||
await knex("o_skillList").insert(list);
|
||
},
|
||
},
|
||
{
|
||
name: "o_skillAttribution",
|
||
builder: (table) => {
|
||
table.text("skillId").notNullable().references("id").inTable("o_skillList").onDelete("CASCADE");
|
||
table.text("attribution").notNullable(); // "production_agent_decision.md" | "production_agent_execution.md" | "production_agent_supervision.md" | "script_agent_decision.md" | "script_agent_execution.md" | "script_agent_supervision.md" | "universal_agent.md"
|
||
table.primary(["skillId", "attribution"]);
|
||
table.index(["attribution"]);
|
||
},
|
||
initData: async (knex) => {
|
||
await knex("o_skillAttribution").insert([
|
||
{
|
||
skillId: "52c51fa8655f899a1b7aae9b6aad7251",
|
||
attribution: "universal_agent.md",
|
||
},
|
||
{
|
||
skillId: "6d46cdca10b2f49e07e515885d1387a0",
|
||
attribution: "universal_agent.md",
|
||
},
|
||
{
|
||
skillId: "1864df75d1d65f76e275046649ecaef8",
|
||
attribution: "universal_agent.md",
|
||
},
|
||
{
|
||
skillId: "3e5efec258c8d8e6a39bcef12f8ee058",
|
||
attribution: "universal_agent.md",
|
||
},
|
||
{
|
||
skillId: "7fbce6f90d7d85496ba9817e9622e640",
|
||
attribution: "universal_agent.md",
|
||
},
|
||
{
|
||
skillId: "31fb5c5a1f514ec1e66b4eba9f22d4db",
|
||
attribution: "script_agent_decision.md",
|
||
},
|
||
{
|
||
skillId: "27dc2dfc901de2180227d0269217583a",
|
||
attribution: "script_agent_execution.md",
|
||
},
|
||
{
|
||
skillId: "d49fa09504fe784a8e6eb102756c6d56",
|
||
attribution: "script_agent_execution.md",
|
||
},
|
||
{
|
||
skillId: "797906c2ddf0750f050bcdeae23eae3d",
|
||
attribution: "script_agent_execution.md",
|
||
},
|
||
{
|
||
skillId: "1abd8675c0c3e62b20c0b151d2ec0fb1",
|
||
attribution: "script_agent_execution.md",
|
||
},
|
||
{
|
||
skillId: "0b7828d7a6ab458a4b201122f08d6c16",
|
||
attribution: "script_agent_supervision.md",
|
||
},
|
||
{
|
||
skillId: "5c1772b5f9c420d9eae9ca02914ba087",
|
||
attribution: "production_agent_decision.md",
|
||
},
|
||
{
|
||
skillId: "75a45cf996015ca819582873887ec301",
|
||
attribution: "production_agent_execution.md",
|
||
},
|
||
{
|
||
skillId: "fce75f69d704c19bebcb356bc1bd6e81",
|
||
attribution: "production_agent_execution.md",
|
||
},
|
||
]);
|
||
},
|
||
},
|
||
//记忆表（message=原始消息, summary=压缩摘要）
|
||
{
|
||
name: "memories",
|
||
builder: (table) => {
|
||
table.text("id").notNullable();
|
||
table.text("isolationKey").notNullable(); // 记忆隔离键
|
||
table.text("type").notNullable(); // 'message' | 'summary'
|
||
table.text("role"); // 'user' | 'assistant'
|
||
table.text("name");
|
||
table.text("content").notNullable();
|
||
table.text("embedding"); // 向量嵌入 JSON
|
||
table.text("relatedMessageIds"); // summary关联的message id列表 JSON
|
||
table.integer("summarized").defaultTo(0); // message是否已被总结 0/1
|
||
table.integer("createTime").notNullable();
|
||
table.primary(["id"]);
|
||
table.index(["isolationKey", "type"]);
|
||
table.index(["isolationKey", "summarized"]);
|
||
},
|
||
},
|
||
];
|
||
|
||
for (const t of tables) {
|
||
const tableExists = await knex.schema.hasTable(t.name);
|
||
if (!tableExists || forceInit) {
|
||
if (tableExists && forceInit) {
|
||
await knex.schema.dropTable(t.name);
|
||
console.log("[初始化数据库] 已存在表删除并重建:", t.name);
|
||
} else {
|
||
console.log("[初始化数据库] 创建数据表:", t.name);
|
||
}
|
||
await knex.schema.createTable(t.name, t.builder);
|
||
if (t.initData) {
|
||
await t.initData(knex);
|
||
console.log("[初始化数据库] 表数据初始化:", t.name);
|
||
}
|
||
}
|
||
}
|
||
};
|