feat: enhance AI slug generation with page content

Improves AI-generated URL slugs by incorporating page content alongside the URL itself. Fetches page markdown and includes it in the AI prompt context, enabling more meaningful and descriptive slug generation based on actual page content rather than URL alone.

Updates the AI prompt to clarify that SLUG derivation should consider both URL and page content when available, removing the strict limitation of using only URL information.

Author: miantiao-me

nuxt.config.ts

@@ -28,7 +28,7 @@ export default defineNuxtConfig({
     cfApiToken: '',
     dataset: 'sink',
     aiModel: '@cf/qwen/qwen3-30b-a3b-fp8',
-    aiPrompt: `You are a URL shortening assistant, please shorten the URL provided by the user into a SLUG. The SLUG information must come from the URL itself, do not make any assumptions. A SLUG is human-readable and should not exceed three words and can be validated using regular expressions {slugRegex} . Only the best one is returned, the format must be JSON reference {"slug": "example-slug"}`,
+    aiPrompt: `You are a URL shortening assistant, please shorten the URL provided by the user into a SLUG. The SLUG information should be derived from the URL and page content (if provided). Do not make any assumptions beyond the given information. A SLUG is human-readable and should not exceed three words and can be validated using regular expressions {slugRegex} . Only the best one is returned, the format must be JSON reference {"slug": "example-slug"}`,
     caseSensitive: false,
     listQueryLimit: 500,
     disableBotAccessLog: false,

server/api/link/ai.get.ts

@@ -36,6 +36,11 @@ export default eventHandler(async (event) => {
   const { aiPrompt, aiModel } = useRuntimeConfig(event)
   const { slugRegex } = useAppConfig()
 
+  const markdown = await fetchPageMarkdown(event, url, AI)
+  const userContent = markdown
+    ? `URL: ${url}\n\nPage content:\n${markdown}`
+    : url
+
   const messages = [
     { role: 'system', content: aiPrompt.replace('{slugRegex}', slugRegex.toString()) },
 
@@ -51,7 +56,7 @@ export default eventHandler(async (event) => {
     { role: 'user', content: 'https://github.com/miantiao-me/sink' },
     { role: 'assistant', content: '{"slug": "sink"}' },
 
-    { role: 'user', content: url },
+    { role: 'user', content: userContent },
   ]
 
   const response = await AI.run(aiModel as keyof AiModels, { messages }) as AiChatResponse

server/utils/markdown.ts

@@ -0,0 +1,52 @@
+import type { H3Event } from 'h3'
+import { ofetch } from 'ofetch'
+
+export async function fetchPageMarkdown(event: H3Event, url: string, AI: Ai): Promise<string | null> {
+  const reqHeaders = Object.fromEntries(
+    Object.entries(getHeaders(event)).filter(([_, v]) => v !== undefined),
+  ) as Record<string, string>
+
+  try {
+    const response = await ofetch.raw(url, {
+      headers: {
+        ...reqHeaders,
+        Accept: 'text/markdown, text/html;q=0.9, */*;q=0.8',
+      },
+      timeout: 5000,
+      responseType: 'text',
+    })
+
+    const contentType = response.headers.get('content-type') || ''
+    const body = response._data as string
+
+    if (!body) {
+      return null
+    }
+
+    // Markdown for Agents: site returns markdown directly
+    if (contentType.includes('text/markdown')) {
+      return body.slice(0, 4096)
+    }
+
+    // Fallback: convert HTML to markdown via AI.toMarkdown()
+    if (contentType.includes('text/html')) {
+      try {
+        const result = await AI.toMarkdown({
+          name: 'page.html',
+          blob: new Blob([body], { type: 'text/html' }),
+        })
+        if (result.format === 'markdown' && result.data) {
+          return result.data.slice(0, 4096)
+        }
+      }
+      catch (err) {
+        console.warn(`[markdown] AI.toMarkdown() failed for ${url}:`, err)
+      }
+    }
+  }
+  catch (err) {
+    console.warn(`[markdown] Failed to fetch ${url}:`, err)
+  }
+
+  return null
+}