Hr Resume Screening Copilot

HR Resume Screening

Purpose

Rank resumes against a job description with an explainable 100-point model.

Use this skill for first-pass screening: read the JD, extract or read resume text, score candidates, and produce a ranked interview recommendation list.

Inputs

Accept:

• A JD pasted in the conversation.
• A JD file path, preferably .md or .txt.
• One PDF resume.
• One folder of PDF resumes.
• Multiple folders of PDF resumes.
• Previously extracted .txt or .json resume files.

If the JD or resume source is missing, ask for it.

Preparation Gate

Preparation-only requests still use this skill. Treat "do not start screening" as a workflow gate, not as a request to bypass the skill.

For a preparation-only request:

• Ask for the JD, resume file or folder paths, and any optional output preferences.
• Explain the accepted input formats when useful.
• Do not read resumes, run extraction, score candidates, or write screening results until the user asks to begin.
• Defer the optional LLM Wiki preflight until screening begins unless the user explicitly asks to initialize or check it.

Shared Resources

Use HR Agent Copilot shared files:

• ../scripts/extract_resumes.py
• ../references/scoring-rubric.md
• ../references/report-template.md

If the skill is installed in a flattened environment, locate equivalent scripts/ and references/ paths near the skill root.

Optional LLM Wiki Augmentation

This skill declares its memory contract in scp.yml. Before the main business workflow, read ../references/llm-wiki-integration.md, run the resolve-config preflight, and apply the declared query flow when HR memory is enabled. If any runtime step fails, follow the shared fallback contract and complete the original workflow.

Workflow

1. Read the JD and identify:
   • Role title and level.
   • Must-have requirements.
   • Nice-to-have requirements.
   • Business domain.
   • Deal breakers.
   • Hidden priorities implied by the user's wording.
2. Read or extract resumes.
   • If resumes are PDFs and text is not already available, run scripts/extract_resumes.py.
   • Use recursive folder scanning when the user provides folders.
3. Extract candidate facts:
   • Name.
   • Years of experience.
   • Age, if present and job-relevant only when the user explicitly asks.
   • Education level, school, and major.
   • Company history.
   • Each job segment duration.
   • Recent job duration.
   • Core technologies.
   • Business domain experience.
   • Project role and responsibility depth.
   • Risk signals and missing information.
4. Score candidates with references/scoring-rubric.md.
5. Produce a ranking report using references/report-template.md.
6. Separate resume facts, reasoned judgments, and interview verification points.

PDF Extraction

Example:

python scripts/extract_resumes.py --input "D:/workspace/ai-workspace/jd" --output output/hr-resume-extracts

Multiple folders:

python scripts/extract_resumes.py --input "D:/resumes/backend" "D:/resumes/ai" --output output/hr-resume-extracts

If pypdf is unavailable:

pip install -r scripts/requirements.txt

Output

Always include:

• Candidate ranking table.
• Score breakdown table.
• Recommended interview list.
• Non-prioritized candidates.
• Key risk and verification points.
• Notes about missing or ambiguous information.

Scoring Discipline

Do not score only by keyword matching.

Reward:

• Direct role/domain match.
• Clear project ownership.
• Production system experience.
• Relevant architecture and problem-solving depth.

Penalize:

• Vague claims.
• Keyword-only experience.
• unclear project ownership.
• unexplained frequent short tenures.

The final score reflects fit for this JD, not general candidate ability.