firecrawl-api / apps /api /src /lib /branding /extractHeaderHtmlChunk.ts
s-eto-hub's picture
Commit de nettoyage des binaires
097eaf0
Raw History Blame Contribute Delete
1.78 kB
/**
* Extract a small chunk of HTML from the top/header area for LLM context when
* no logo candidates were found. Helps the LLM infer brand name and note that
* a logo may exist but wasn't captured (e.g. inline SVG, shadow DOM).
*
* Strategy: strip noise, find header/nav/body start, take first N chars.
*/
const MAX_HEADER_CHUNK_CHARS = 5500;
/**
* Remove script and style tags and their content, and HTML comments.
* Uses regex; safe for typical page header content.
*/
function stripNoise(html: string): string {
let out = html;
// Comments
out = out.replace(/<!--[\s\S]*?-->/g, "");
// Scripts (non-greedy to first </script>)
out = out.replace(/<script\b[\s\S]*?<\/script>/gi, "");
// Styles
out = out.replace(/<style\b[\s\S]*?<\/style>/gi, "");
return out;
}
/**
* Find the earliest start of header-like content: <header, <nav, or <body.
* Returns index into html, or 0 if not found.
*/
function findHeaderStart(html: string): number {
const lower = html.toLowerCase();
const header = lower.indexOf("<header");
const nav = lower.indexOf("<nav");
const body = lower.indexOf("<body");
const indices = [header, nav, body].filter(i => i >= 0);
return indices.length > 0 ? Math.min(...indices) : 0;
}
/**
* Extract a single chunk of HTML useful for logo/brand context: header/nav
* area, stripped of scripts/styles/comments, length-limited.
* Use only when there are no logo candidates so the LLM has fallback context.
*/
export function extractHeaderHtmlChunk(html: string): string {
if (!html || typeof html !== "string") return "";
const stripped = stripNoise(html);
const start = findHeaderStart(stripped);
const chunk = stripped.slice(start, start + MAX_HEADER_CHUNK_CHARS);
return chunk.replace(/\s+/g, " ").trim();
}