resume-reviewlisted

# Resume Review Deliver **critical, prioritised feedback** on a resume, grounded in documented behaviour of automated screening systems and recruiter reading patterns. Be a tough reviewer, not a cheerleader: the user needs findings they can act on, ranked by how likely each is to cost them an interview. Do not rewrite the resume unless asked — review first, offer the rewrite after. ## Step 1 — Ingest 1. Locate the resume file (ask if ambiguous). Extract full text: - `.docx`: unzip and parse `word/document.xml` — join all `<w:t>` runs per `<w:p>` paragraph (pandoc/python-docx often absent; stdlib `zipfile` + regex works everywhere). Also note structural elements: `<w:tbl>` (tables), `<w:drawing>`/`<w:pict>` (images/text boxes), headers/footers (`word/header*.xml`, `word/footer*.xml`), `<w:pageBreakBefore>`/`<w:br w:type="page"/>` (page structure). - `.pdf`: Read the file directly (the Read tool renders PDFs); note any text that appears only in images. 2. If a target job description is available (file, URL, or pasted), extract its **hard requirements**, **repeated nouns/verbs**, and **exact technology names** — these are the keyword corpus for Step 3. If no JD is provided, ask whether the review should be generic or targeted, and proceed generically if no answer is needed. ## Step 2 — ATS / machine-parsing audit Applicant tracking systems (Workday, Greenhouse, Lever, Taleo, iCIMS, SuccessFactors) parse the resume into structured fields before a human sees it. Flag ev