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
import type { WechatArticleImagePlanItem, WechatArticleTopic } from '@opencreator/protocol';
import { z } from 'zod/v3';
import type { CreatorServicesConfigStore } from '../../creator-services/config-store.js';
import {
creatorServiceErrorInfo,
fetchCreatorService,
openAiCompatibleEndpoint
} from '../../creator-services/upstream-fetch.js';
import { CreatorExecutorError } from '../executor.js';
import type { ExtractedArticleSource } from './source-extractor.js';
const topicsSchema = z.object({
topics: z.array(z.object({
title: z.string().min(1).max(160),
angle: z.string().max(500).default(''),
summary: z.string().max(1_000).default('')
}).strict()).min(1).max(10)
}).strict();
const imagePlanSchema = z.object({
images: z.array(z.object({
caption: z.string().min(1).max(200),
placementHeading: z.string().min(1).max(200),
prompt: z.string().min(1).max(2_000)
}).strict()).min(1).max(10)
}).strict();
export function createWechatArticleModel(input: {
configStore: Pick<CreatorServicesConfigStore, 'read'>;
fetchImpl?: typeof fetch;
}) {
async function complete(request: {
prompt: string;
signal: AbortSignal;
json?: boolean;
}): Promise<string> {
const config = await input.configStore.read();
if (!config.llm.apiKey.trim() || !config.llm.model.trim()) {
throw new CreatorExecutorError('creator_llm_config_missing', 'LLM configuration is incomplete');
}
const response = await fetchCreatorService({
endpoint: openAiCompatibleEndpoint(config.llm.baseUrl, 'chat/completions'),
method: 'POST',
headers: {
Authorization: `Bearer ${config.llm.apiKey}`,
'Content-Type': 'application/json'
},
body: JSON.stringify({
model: config.llm.model,
...(request.json && config.llm.jsonMode ? { response_format: { type: 'json_object' } } : {}),
messages: [{ role: 'user', content: request.prompt }]
}),
proxy: config.proxy,
signal: request.signal,
maxResponseBytes: 4 * 1024 * 1024,
fetchImpl: input.fetchImpl
});
if (!response.ok) {
const failure = await creatorServiceErrorInfo(response, 'Article writing', 'llm');
throw new CreatorExecutorError(
'creator_llm_upstream_error',
failure.message,
{},
failure.publicFacts
);
}
const payload = await response.json() as {
choices?: Array<{ message?: { content?: string | Array<{ text?: string }> } }>;
};
const raw = payload.choices?.[0]?.message?.content;
const content = typeof raw === 'string'
? raw
: Array.isArray(raw)
? raw.map(item => item.text ?? '').join('')
: '';
if (!content.trim()) {
throw new CreatorExecutorError('creator_llm_empty_response', 'Article writing returned no content');
}
return content.trim();
}
return {
async generateTopics(request: {
sources: ExtractedArticleSource[];
writingPrompt: string;
templatePrompt: string;
count: number;
signal: AbortSignal;
}): Promise<WechatArticleTopic[]> {
const parsed = topicsSchema.parse(parseJson(await complete({
json: true,
signal: request.signal,
prompt: [
'你是一名资深微信公众号主编。请根据来源和用户要求提出可直接写作的选题。',
`生成 ${request.count} 个差异明显的选题,只输出严格 JSON:{"topics":[{"title":"标题","angle":"核心切入角度","summary":"文章内容摘要"}]}`,
'标题要具体、克制、有信息量,不使用虚假数据或来源没有支持的结论。',
request.templatePrompt.trim() ? `文章模板规则:\n${request.templatePrompt}` : '',
request.writingPrompt.trim() ? `用户写作要求:${request.writingPrompt}` : '用户没有额外写作要求。',
sourcePrompt(request.sources)
].filter(Boolean).join('\n\n')
})));
return parsed.topics.slice(0, request.count).map((topic, index) => ({
id: `topic-${index + 1}`,
...topic
}));
},
generateOutline(request: {
sources: ExtractedArticleSource[];
topic: WechatArticleTopic;
writingPrompt: string;
templatePrompt: string;
signal: AbortSignal;
}): Promise<string> {
return complete({
signal: request.signal,
prompt: [
'你是一名资深微信公众号主编。请为选定选题生成可编辑的文章大纲。',
'使用 Markdown。第一行是文章标题,随后列出导语、二到五个主体章节及每节要点、结尾。',
'大纲要形成完整论证或叙事推进,明确哪些事实来自素材,禁止虚构。不要直接写成全文。',
`选题:${JSON.stringify(request.topic)}`,
request.templatePrompt.trim() ? `文章模板规则:\n${request.templatePrompt}` : '',
request.writingPrompt.trim() ? `用户写作要求:${request.writingPrompt}` : '用户没有额外写作要求。',
sourcePrompt(request.sources)
].filter(Boolean).join('\n\n')
});
},
generateArticle(request: {
sources: ExtractedArticleSource[];
topic: WechatArticleTopic;
outline: string;
writingPrompt: string;
templatePrompt: string;
layoutPrompt: string;
signal: AbortSignal;
}): Promise<string> {
return complete({
signal: request.signal,
prompt: [
'你是一名资深微信公众号作者。请按选题和大纲完成一篇可直接编辑发布的中文文章。',
'只输出 Markdown 正文,从一级标题开始。保持事实边界,来源没有的信息不要编造;不要输出写作说明。',
`选题:${JSON.stringify(request.topic)}`,
`大纲:\n${request.outline}`,
request.templatePrompt.trim() ? `文章模板要求:${request.templatePrompt}` : '',
request.layoutPrompt.trim() ? `Markdown 排版要求:${request.layoutPrompt}` : '',
request.writingPrompt.trim() ? `用户写作要求:${request.writingPrompt}` : '',
sourcePrompt(request.sources)
].filter(Boolean).join('\n\n')
});
},
async generateImagePlan(request: {
articleMarkdown: string;
count: number;
stylePrompt: string;
customPrompt: string;
signal: AbortSignal;
}): Promise<WechatArticleImagePlanItem[]> {
const parsed = imagePlanSchema.parse(parseJson(await complete({
json: true,
signal: request.signal,
prompt: [
'你是一名资深视觉编辑。请为下面这篇微信公众号文章规划内容互不重复的配图。',
`生成 ${request.count} 张配图,只输出严格 JSON:{"images":[{"caption":"图片说明","placementHeading":"建议放在哪个章节标题之后","prompt":"可直接用于图像生成模型的完整英文提示词"}]}`,
'placementHeading 必须原样复制文章中真实存在的 Markdown 标题,优先选择二级或三级标题。',
'每张图必须服务于所选章节的具体内容,并覆盖不同的核心段落。prompt 要准确呈现该章节中的人物、对象、场景、动作、关系或数据含义;抽象内容可以使用贴合原意的视觉隐喻,但禁止生成与正文无关的装饰性画面。',
'主体、场景和视觉隐喻要具体。不要把文章标题、段落文字、Logo、水印或大段可见文字放进画面。',
`统一视觉风格:${request.stylePrompt}`,
request.customPrompt.trim() ? `用户补充的配图要求:${request.customPrompt}` : '',
`文章正文:\n${request.articleMarkdown.slice(0, 100_000)}`
].filter(Boolean).join('\n\n')
})));
return parsed.images.slice(0, request.count).map((item, index) => ({
id: `article-image-${index + 1}`,
...item
}));
}
};
}
function sourcePrompt(sources: ExtractedArticleSource[]): string {
if (sources.length === 0) return '本次没有内容来源,请基于用户要求写作,并避免捏造具体事实和数据。';
let remaining = 120_000;
const sections: string[] = [];
for (const [index, source] of sources.entries()) {
if (remaining <= 0) break;
const header = `[来源 ${index + 1}] ${source.title}\n出处:${source.origin}\n`;
const content = source.content.slice(0, Math.max(0, remaining - header.length));
sections.push(`${header}${content}`);
remaining -= header.length + content.length;
}
return `内容来源:\n${sections.join('\n\n')}`;
}
function parseJson(value: string): unknown {
const fenced = /^```(?:json)?\s*([\s\S]*?)\s*```$/i.exec(value.trim());
return JSON.parse(fenced?.[1] ?? value);
}