| 1 | /** |
| 2 | * 消息内容解析工具 |
| 3 | * 支持多种格式的用户消息内容解析： |
| 4 | * 1. Markdown 引用格式：[image]: url, [video]: url, [audio]: url, [document]: url |
| 5 | * 2. Claude Prompt 格式：{ type, source, text, cache_control } |
| 6 | */ |
| 7 | |
| 8 | import type { IParsedUserContent, IPromptContentItem, IUploadedMedia } from '@/store/agent' |
| 9 | |
| 10 | /** |
| 11 | * Markdown 引用格式正则 |
| 12 | * 匹配：[image]: url, [video]: url, [audio]: url, [document]: url |
| 13 | */ |
| 14 | const MARKDOWN_REFERENCE_REGEX = /^\[(image|video|audio|document)\]: *(\S+)$/gim |
| 15 | |
| 16 | /** |
| 17 | * 判断文本是否为 JSON 格式的 Claude Prompt |
| 18 | */ |
| 19 | function isClaudePromptFormat(text: string): boolean { |
| 20 | const trimmed = text.trim() |
| 21 | return trimmed.startsWith('[') && trimmed.includes('"type"') && trimmed.includes('"source"') |
| 22 | } |
| 23 | |
| 24 | /** |
| 25 | * 解析 Markdown 引用格式 |
| 26 | * @example |
| 27 | * [image]: https://example.com/image.jpg |
| 28 | * [video]: https://example.com/video.mp4 |
| 29 | * [audio]: https://example.com/audio.mp3 |
| 30 | */ |
| 31 | function parseMarkdownReferences(text: string): IParsedUserContent { |
| 32 | const medias: IUploadedMedia[] = [] |
| 33 | let cleanedText = text |
| 34 | |
| 35 | // 提取所有引用 |
| 36 | const matches = text.matchAll(MARKDOWN_REFERENCE_REGEX) |
| 37 | |
| 38 | for (const match of matches) { |
| 39 | const [fullMatch, type, url] = match |
| 40 | const trimmedUrl = url.trim() |
| 41 | |
| 42 | if (trimmedUrl) { |
| 43 | medias.push({ |
| 44 | url: trimmedUrl, |
| 45 | type: type as IUploadedMedia['type'], |
| 46 | }) |
| 47 | // 从文本中移除这条引用 |
| 48 | cleanedText = cleanedText.replace(fullMatch, '') |
| 49 | } |
| 50 | } |
| 51 | |
| 52 | return { |
| 53 | text: cleanedText.trim(), |
| 54 | medias, |
| 55 | hasSpecialFormat: medias.length > 0, |
| 56 | } |
| 57 | } |
| 58 | |
| 59 | /** |
| 60 | * 解析 Claude Prompt 格式 |
| 61 | * @example |
| 62 | * [ |
| 63 | * { "type": "image", "source": { "type": "url", "url": "..." } }, |
| 64 | * { "type": "text", "text": "..." } |
| 65 | * ] |
| 66 | */ |
| 67 | function parseClaudePrompt(text: string): IParsedUserContent { |
| 68 | try { |
| 69 | const items: IPromptContentItem[] = JSON.parse(text) |
| 70 | |
| 71 | if (!Array.isArray(items)) { |
| 72 | throw new TypeError('Invalid format: not an array') |
| 73 | } |
| 74 | |
| 75 | const medias: IUploadedMedia[] = [] |
| 76 | const textParts: string[] = [] |
| 77 | |
| 78 | items.forEach((item) => { |
| 79 | if (item.type === 'text' && item.text) { |
| 80 | textParts.push(item.text) |
| 81 | } |
| 82 | else if (['image', 'video', 'audio', 'document'].includes(item.type)) { |
| 83 | if (item.source?.url) { |
| 84 | const media: IUploadedMedia = { |
| 85 | url: item.source.url, |
| 86 | type: item.type as IUploadedMedia['type'], |
| 87 | } |
| 88 | |
| 89 | // 添加缓存控制（如果存在） |
| 90 | if (item.cache_control) { |
| 91 | media.cache_control = item.cache_control |
| 92 | } |
| 93 | |
| 94 | medias.push(media) |
| 95 | } |
| 96 | } |
| 97 | }) |
| 98 | |
| 99 | return { |
| 100 | text: textParts.join('\n\n'), |
| 101 | medias, |
| 102 | hasSpecialFormat: true, |
| 103 | } |
| 104 | } |
| 105 | catch (error) { |
| 106 | console.error('Failed to parse Claude prompt format:', error) |
| 107 | // 解析失败，返回原始文本 |
| 108 | return { |
| 109 | text, |
| 110 | medias: [], |
| 111 | hasSpecialFormat: false, |
| 112 | } |
| 113 | } |
| 114 | } |
| 115 | |
| 116 | /** |
| 117 | * 解析用户消息内容 |
| 118 | * 自动识别并解析不同格式 |
| 119 | */ |
| 120 | export function parseUserMessageContent(content: string | any[]): IParsedUserContent { |
| 121 | // 如果已经是数组格式（Claude format），直接解析 |
| 122 | if (Array.isArray(content)) { |
| 123 | try { |
| 124 | const jsonStr = JSON.stringify(content) |
| 125 | return parseClaudePrompt(jsonStr) |
| 126 | } |
| 127 | catch { |
| 128 | return { |
| 129 | text: '', |
| 130 | medias: [], |
| 131 | hasSpecialFormat: false, |
| 132 | } |
| 133 | } |
| 134 | } |
| 135 | |
| 136 | // 字符串格式 |
| 137 | if (typeof content !== 'string') { |
| 138 | return { |
| 139 | text: '', |
| 140 | medias: [], |
| 141 | hasSpecialFormat: false, |
| 142 | } |
| 143 | } |
| 144 | |
| 145 | const trimmedContent = content.trim() |
| 146 | |
| 147 | // 判断是否为 JSON 格式的 Claude Prompt |
| 148 | if (isClaudePromptFormat(trimmedContent)) { |
| 149 | return parseClaudePrompt(trimmedContent) |
| 150 | } |
| 151 | |
| 152 | // 判断是否包含 Markdown 引用 |
| 153 | if (MARKDOWN_REFERENCE_REGEX.test(trimmedContent)) { |
| 154 | // 重置正则的 lastIndex |
| 155 | MARKDOWN_REFERENCE_REGEX.lastIndex = 0 |
| 156 | return parseMarkdownReferences(trimmedContent) |
| 157 | } |
| 158 | |
| 159 | // 普通文本 |
| 160 | return { |
| 161 | text: trimmedContent, |
| 162 | medias: [], |
| 163 | hasSpecialFormat: false, |
| 164 | } |
| 165 | } |
| 166 | |
| 167 | /** |
| 168 | * 格式化媒体类型显示名称 |
| 169 | */ |
| 170 | export function formatMediaTypeName(type: 'image' | 'video' | 'audio' | 'document'): string { |
| 171 | const names: Record<string, string> = { |
| 172 | image: 'Image', |
| 173 | video: 'Video', |
| 174 | audio: 'Audio', |
| 175 | document: 'Document', |
| 176 | } |
| 177 | return names[type] || type |
| 178 | } |
| 179 |