A job match needs the posting's text. If you have it as text, pass it as jobDescription (pasted text). If you have the page's HTML, @veriworkly/ats-engine/job extracts the posting from it. The subpath is 3.5 KB gzipped, has no dependencies and runs in browsers, at the edge and on servers. It parses HTML only. It never fetches a page.

From saved HTML

import { check } from "@veriworkly/ats-engine";
import { extractJobPosting, jobTextFromHtml } from "@veriworkly/ats-engine/job";

export function matchToPage(resume: string, html: string) {
  return check(resume, undefined, {
    jobDescription: jobTextFromHtml(html),
    jobCompany: extractJobPosting(html)?.company, // optional: the employer, never a keyword
  });
}

jobTextFromHtml(html) returns the text to match against. If the page publishes a schema.org JobPosting as JSON-LD with a real description (200 characters or more), it uses that: the employer's own title, company, description and requirement fields, without the navigation, cookie banner and "similar jobs" that make up most of a page's visible text. Otherwise it reads the page's visible text, from <main> or <article> when there is one.

jobCompany keeps every word of the employer's name out of the missing-keywords list, so "Acme" is not reported as a skill you lack. undefined is fine when the page names no employer.

What each helper returns

import {
  MAX_JOB_TEXT_CHARS,
  extractJobPosting,
  jobHtmlToText,
  jobTextFromHtml,
  normalizeJobText,
} from "@veriworkly/ats-engine/job";

export function inspect(html: string, pasted: string) {
  // The JobPosting the page publishes, or null
  const posting = extractJobPosting(html);
  console.log(posting?.title, posting?.company, posting?.requirements);

  // The page's visible text, one line per block, whether or not it has JSON-LD
  const visible = jobHtmlToText(html);

  // The text to match: JSON-LD when it is a real posting, else the visible text
  const text = jobTextFromHtml(html);

  // Pasted or file text, tidied the same way: trimmed lines, blanks dropped, capped
  const tidy = normalizeJobText(pasted);

  console.log(MAX_JOB_TEXT_CHARS, visible.length, text.length, tidy.length);
}

Headings and bullets survive as separate lines, which the matcher uses to tell required skills from preferred ones and to leave out about-us and benefits blocks (job matching). check itself reads the first 20,000 characters of a jobDescription.

Fetching a page

The core and /job make no network calls. If your server fetches a posting URL that a user typed, the request is yours to protect: an address a user chooses can point at a machine on your network. Allow only the hosts you mean to read, require https, do not follow redirects blindly, and bound the time and the size of the response, then pass the HTML to jobTextFromHtml.

From the command line

The CLI can read a posting from Greenhouse, Lever and Ashby by link:

npx @veriworkly/ats-engine check resume.pdf --job-url https://job-boards.greenhouse.io/acme/jobs/123

This is the one network request check makes without --ai, and it lives in the CLI only. It asks the board's public JSON API, never the page, and only the API hosts it lists for the posting's host (boards-api.greenhouse.io, api.lever.co, api.eu.lever.co, api.ashbyhq.com), never the host in your link. It takes https addresses only, follows no redirect, waits at most 20 seconds and reads at most 20 MB. It sends no resume text and no key.

It reads the posting's title and description and the employer's name. Greenhouse names the employer; Lever and Ashby do not, so the board's name in the link is used. A Greenhouse or Lever posting also adds that ATS's documented notes to the advice, as --ats does, unless --ats names another.

Any other site, and Greenhouse's EU boards, are refused with a message to save the page and pass it with --job. --job and --job-url together are refused. See CLI for every flag.

What to watch for

• A posting that publishes only a teaser in its JSON-LD falls back to the page's visible text, which can include page furniture the helpers could not remove.
• The match reads a posting for what it asks of the candidate. It does not judge the posting or the employer.

Next steps

• How terms, requirements and the match score work: job matching.
• Read a posting from a file in Node: files in Node.