Candidate Assessment

Candidate assessment (resume × JD match)

Score a resume against a job description and produce a clean, professional HTML evaluation report (brand-neutral) — for a top-recruiter-grade hiring assessment.

How it works — division of labor

Same philosophy as resume-parsing: the model does judgment, scripts do the deterministic parts.

• resume-parsing skill turns the resume PDF/DOCX into structured resume.json + clean markdown (no hallucination).
• scripts/read_jd.py reads the JD from any format into plain text.
• You (the model) apply the evaluation model in reference/assessment-prompt.md to the resume + JD and produce assessment.json (scores, analysis, risks, questions).
• scripts/render_report.py renders assessment.json into a self-contained, professionally-styled report.html — consistent visuals every time.

Only python3 is needed; pdfmuse auto-installs on first use.

Workflow

Copy this checklist and track progress:

- [ ] 1. Parse the resume (resume-parsing skill) → resume.json + .extract.md
- [ ] 2. Read the JD (read_jd.py) → jd text
- [ ] 3. Read reference/assessment-prompt.md
- [ ] 4. Evaluate → write assessment.json (per the schema there)
- [ ] 5. Render → render_report.py assessment.json --out report.html
- [ ] 6. Open report.html in the browser

Step 1 — Parse the resume

Use the resume-parsing skill on the candidate's resume to get resume.json and .extract.md. (Directly: python ~/.claude/skills/resume-parsing/scripts/extract.py RESUME.pdf --out out, then map to resume.json per that skill.)

Step 2 — Read the JD

python scripts/read_jd.py JD.pdf --out jd.txt      # .txt/.md/.pdf/.docx/.rtf

If the target position name isn't obvious, take it from the JD title (fallback: the JD filename), and confirm with the user if ambiguous.

Step 3–4 — Evaluate

Read reference/assessment-prompt.md — it defines the recruiter role, the weighted 7-module model (0 准入 / 1 硬实力 30% / 2 经验 30% / 3 胜任力 15% / 4 动机稳定 10% / 5 潜力 10% / 6 文化 5% / + 亮点), the scoring discipline (evidence only, quantify, specific risks), and the exact assessment.json schema. Write assessment.json following it.

Step 5 — Render (HTML, and PDF if wanted)

python scripts/render_report.py assessment.json --out report.html          # HTML
python scripts/render_report.py assessment.json --out report.html --pdf    # + report.pdf

--pdf prints the report to report.pdf via headless Chrome (colors preserved, no browser header/footer). If no Chrome/Chromium/Edge is found, skip --pdf and use the in-page button instead.

Step 6 — Show it

open -a "Google Chrome" report.html    # macOS; falls back to any browser

The HTML has a floating 「⬇ 导出 PDF」 button (Print → Save as PDF) with print-friendly styles (A4, colors kept, no mid-card page breaks, button hidden in the PDF) — so the user can export a clean PDF themselves anytime.

Output

• assessment.json — structured scores + analysis (reusable / for a DB).
• report.html — the 候选人内部评估报告, clean brand-neutral design; export to PDF via the in-page button or --pdf. Design tokens: reference/report-design.md.
• report.pdf — (with --pdf) print-ready A4 report.

Reference files

• reference/assessment-prompt.md — evaluation model + assessment.json schema.
• reference/report-design.md — design tokens the report follows (indigo accent + semantic colors).
• scripts/read_jd.py — JD reader (run it).
• scripts/render_report.py — JSON → HTML renderer (run it).