rate-cv

A Claude Code skill that rates and critiques a CV/resume: it scores the document out of 100 against a weighted engineering-leadership rubric, checks ATS parse-ability, quantified impact, and AI-slop, and can match the CV to a specific job description with a keyword gap list.

The score is a sanity check, not truth. Real ATS platforms don't emit a score a recruiter sees, and commercial checkers disagree by ~20 points on the same file. So the skill leads with an auditable per-category breakdown and specific, coaching-framed fixes, not just a number.

What it does

• Standalone quality (no job description): how strong the CV is for engineering-leadership roles, and what would raise it.
• Targeted match (job description supplied): the above, plus a separate ATS keyword match and qualification-fit read against that role.
• Mechanical signals from scripts/cv_signals.py: ATS parse-ability (format, single-flow vs tables/columns, section headers, dated roles, contact block), quantification density, action-verb usage, em-dash count, AI-slop terms, and JD keyword overlap.
• Writing quality via an optional natural-writing linter pass, if present at ~/.claude/skills/natural-writing.
• Judgement categories (leadership signals, impact specificity, seniority, fit) scored against references/rubric.md, using the bullet formulas and ATS rules in references/craft.md.

It never coaches toward fabrication: if a job wants a keyword the candidate can't truthfully claim, it reports a gap rather than suggesting you invent it.

Install

git clone https://github.com/maccydee/rate-cv.git ~/.claude/skills/rate-cv

Then in Claude Code, run /rate-cv (or just ask it to "rate my CV"). Works on .docx and .txt; for PDFs, extract the text first.

Layout

rate-cv/
  SKILL.md              orchestration + workflow
  scripts/cv_signals.py mechanical signal engine (.docx/.txt, optional --jd)
  references/rubric.md   the weighted 100-point scoring rubric
  references/craft.md    bullet formulas + real-ATS parse rules + red flags

The rubric (100 points)

ATS parse-ability 15 · structure & clarity 12 · quantified impact 20 · engineering-leadership signals 28 · seniority & scope 15 · role fit & communication 10, with deductions (cap −15) for AI-slop, unexplained gaps, missing dates, and unverifiable claims. Bands: 90+ shortlist-strong, 75–89 qualified, 60–74 borderline.

Usage from the command line

The signal engine runs standalone too:

python3 scripts/cv_signals.py path/to/CV.docx
python3 scripts/cv_signals.py path/to/CV.docx --jd job-description.txt
python3 scripts/cv_signals.py path/to/CV.docx --json

Licence

MIT.