jd-scorecard

name	jd-scorecard
description	Turn job descriptions and PDF resumes into structured hiring decisions, interview questions, and Feishu/DingTalk-friendly output.

JD Scorecard Skill

HRClaw turns messy JD text and PDF resumes into recruiter-ready decisions. It keeps screening consistent, fast, and easy to share in team chat.

把 JD 和 PDF 简历变成结构化、可执行的招聘结论。

Use this skill for two related flows:

• JD -> scorecard
• Resume PDF/text -> score against a scorecard

Best for

• high-volume recruiting
• QA / Python / operations roles
• teams that want one repeatable scoring standard
• Feishu / DingTalk collaboration

If the user gives both a JD and a resume, generate the scorecard first and then score the resume.

JD flow

Default to a single JSON object with:

• role_title
• summary
• filters
• must_have
• nice_to_have
• exclude
• weights
• thresholds
• interview_questions
• red_flags
• assumptions
• next_steps

If the user asks for a readable version, format the same content with templates/scorecard.md. If the user asks for a Feishu/DingTalk-friendly chat view, format the same content with templates/chat-scorecard.md.

Resume score flow

Use this flow when the user uploads a resume PDF or pastes resume text together with a scorecard.

If the user only provides a resume, ask for a scorecard or JD before scoring.

1. Extract the resume text from the PDF first.
2. If the PDF is image-only and no readable text is available, set extraction_status to needs_ocr and stop.
3. Normalize the resume into a candidate profile.
4. Score it against the provided scorecard using the same filters, weights, and thresholds.
5. Return one pure JSON object first.

Resume output should include:

• mode
• source_type
• extraction_status
• scorecard_name
• candidate_profile
• hard_filter_pass
• hard_filter_fail_reasons
• dimension_scores
• total_score
• decision
• review_reasons
• matched_terms
• missing_terms
• blocked_terms
• evidence
• summary
• next_steps

If the user asks for a Feishu/DingTalk-friendly chat view, format the same content with templates/chat-resume-score.md.

Candidate profile fields:

• name
• location
• years_experience
• education_level
• current_title
• current_company
• skills
• industry_tags

If the user provides a JD and a resume together, generate the scorecard first, then score the resume against it.

Rules

• Use only explicit evidence from the JD.
• For resume scoring, use only explicit evidence from the resume and scorecard.
• Do not invent requirements or hidden intent.
• Keep one primary role per scorecard.
• If the JD is mixed or vague, add short assumptions instead of guessing.
• Prefer practical screening signals over generic hiring advice.
• Generate 5 to 10 interview questions that test real work.
• If a resume PDF is unreadable and OCR text is not available, say so clearly instead of guessing.

Flow

1. Extract the role, location, years of experience, education, tools, and exclusions.
2. Convert those signals into a scorecard.
3. Add interview questions that verify the must-haves.
4. Add red flags that help a recruiter reject quickly.
5. For resumes, extract the profile, apply the scorecard, and return the scoring JSON first.

References

• references/quickstart.md
• references/faq.md
• references/limitations.md
• prompts/jd-to-scorecard.md
• prompts/resume-score.md
• prompts/interview-questions.md
• templates/scorecard.json
• templates/scorecard.md
• templates/chat-scorecard.md
• templates/resume-score.json
• templates/resume-score.md
• templates/chat-resume-score.md