Reviewed by Jonathan West · Updated Jul 2, 2026

Claude Skills for HR Teams: 8 Workflows for a Lean People Function (2026)

Concrete Skill recipes for SMB HR and recruiting teams — with the EEOC 'assist, don't decide' guardrail baked in.

Reviewed by Jonathan West · Updated Jul 2, 2026

Claude Skills for HR give a small people team a repeatable way to draft, summarize, and answer the same questions the same way every time. A Skill is a folder Anthropic's Claude loads on demand — it holds the instructions, templates, and reference files your HR function already relies on. For a lean HR team of one to five, that means the tenth job description reads like the first, and the handbook answer an employee gets on a Sunday matches the one HR would give on Monday.

The catch is that HR work is high-stakes YMYL territory. The EEOC has been clear since its 2023 Technical Assistance Document that employers stay liable for AI-assisted employment decisions — the tool does not shift the burden. Skills are a good fit here precisely because they are packaged, reviewable, and version-controlled. You can read the SKILL.md file, audit the template, and keep humans on the decision.

This guide walks through eight Skill recipes an SMB HR team can stand up in a quarter: JD drafting, resume-screening summaries, interview-note synthesis, offer-letter tuning, handbook Q&A, policy-response drafting, onboarding checklists, and exit-interview theme extraction. Each recipe names the guardrails and the failure modes we watch for at Layer3 Labs.

What Claude Skills Are (and Why They Fit HR)

Claude Skills are reusable capability packages Anthropic launched in October 2025. Each Skill is a folder with a SKILL.md file, YAML frontmatter describing when it should activate, and optional scripts or reference files. Claude reads only the frontmatter until a task matches, then loads the full Skill on demand.

For HR that pattern matters. Your handbook, your JD template, your interview scorecard, and your EEOC-compliant screening rubric all live as files a human can read and edit. Unlike a chat prompt you paste every time, a Skill is version-controlled and portable across Claude.ai, Claude Code, and the Anthropic API.

Skills are not the same as MCP servers or Custom GPTs. MCP is a live protocol for tools and data at request time; Skills are packaged how-to's. For an SMB HR team without engineering support, Skills are usually the lower-lift starting point.

  • Skill = folder with SKILL.md + YAML frontmatter + optional templates
  • Loads on demand — you can keep dozens without cluttering the model's context
  • Portable: works on Claude.ai, Claude Code, Claude for Small Business, and the API
  • Auditable: every instruction lives in a file HR and legal can review
  • Successor to Claude Projects, which held only static context
For HR buyers, think of a Skill as your standard operating procedure written in a form Claude can execute — not a black-box model behavior.
Sources
Anthropic — Introducing Claude SkillsClaude Skills — Official Documentation

Want help scoping the first Claude Skills for your HR team — JD drafting, handbook Q&A, or an EEOC-safe screening summary? Layer3 Labs will map your highest-leverage HR Skills and the guardrails they need.

Book a Consultation

The EEOC Rule for AI in HR: Assist, Don't Decide

The EEOC's position is that AI tools used in employment decisions must not violate Title VII, the ADA, or the ADEA, and that the employer — not the vendor — is responsible for compliance. That guidance sits inside the agency's 2022 and 2023 Technical Assistance Documents on AI in hiring and continues to frame the 2026 landscape.

For Claude Skills, the practical rule is simple: a Skill can assist a human by drafting, summarizing, or checking work — but the decision to advance, reject, hire, promote, discipline, or terminate must be made by a qualified person. That means every screening or evaluation Skill needs a "human approves the outcome" step baked into the SKILL.md.

The SHRM 2025 AI in the Workplace report found that 26% of HR pros already use AI in some part of the employee lifecycle, but only a minority have written policies governing it. Skills close that gap because the policy IS the Skill file.

  • Employer is liable for AI-assisted decisions — vendor disclaimers do not transfer risk
  • ADA: candidates must be able to request accommodations if AI is used in screening
  • Disparate impact still applies — test the Skill's output on your applicant pool
  • Keep human-in-the-loop language in every SKILL.md that touches a candidate or employee
  • OFCCP-covered federal contractors face additional record-keeping duties on selection tools
A defensible pattern: Skills produce summaries and drafts. Humans produce decisions and communications. Put that line in writing.
Sources
EEOC — The Americans with Disabilities Act and the Use of Software, Algorithms, and AI to Assess Job Applicants and EmployeesEEOC — Assessing Adverse Impact in Software, Algorithms, and AI Used in Employment Selection Procedures

Drafting Skills: JDs, Offer Letters, and Policy Responses

Drafting Skills are the safest starting point for HR because the output is a document a human edits before sending. The three highest-value drafting Skills we build with SMB HR teams are a JD Skill, an Offer Letter Skill, and a Policy Response Skill.

The JD Skill takes a role card — title, level, team, key outcomes, must-have skills, comp band, location model — and returns a job description in your firm's house voice. The template lives in the Skill folder, so brand and legal disclaimers stay consistent. This is where "the tenth JD reads like the first" pays off for a 40-person company hiring across roles.

The Offer Letter Skill tunes a base template for state-specific wage transparency and at-will language. The Policy Response Skill drafts an initial reply to common employee questions — PTO carryover, remote-work exception, bereavement — using your handbook as the source of truth.

  • JD Skill: role card in, JD out, includes EEO statement and pay range block by default
  • Offer Letter Skill: state-aware template with wage-transparency inserts for CA, CO, NY, WA, IL
  • Policy Response Skill: handbook is a reference file the Skill cites when drafting
  • All three should end with "human sign-off before sending" in SKILL.md
  • Version the templates — a git-tracked folder gives you a compliance audit trail
Sources
Anthropic Skills — Public Examples RepositorySHRM — 2025 State of the Workplace: AI in HR

Summarization Skills: Resume Screens, Interview Notes, Exit Themes

Summarization Skills are where HR teams get the biggest time savings — and where the EEOC guardrails matter most. The three we recommend are a Resume Summary Skill, an Interview Note Synthesis Skill, and an Exit Interview Themes Skill.

The Resume Summary Skill produces a structured summary against the JD's explicit criteria — years of relevant experience, named skills, education if required. It does not score, rank, or reject. The recruiter reads the summary and the resume side by side and makes the advance decision. This is the pattern the EEOC's ADA guidance points toward: a tool that supports human review rather than replacing it.

The Interview Notes Skill takes raw scorecard notes from three or four interviewers and produces a single synthesis organized by competency. The Exit Interview Skill takes the last quarter of exit interview transcripts and returns themes — pay, manager, growth, workload — with anonymized quotes.

  • Resume Summary Skill outputs a structured summary — never a rank, score, or thumbs-up/down
  • Log the JD version and rubric version used so you can rerun the same summary later
  • Interview Notes Skill: pass in the JD's competencies as the synthesis frame
  • Exit Interview Themes Skill: strip PII before summarization to protect the anonymity promise
  • Failure mode we've seen: Skills that infer protected characteristics from names — instruct against it explicitly
A summary Skill that reduces a resume to "strong / weak / neutral" has crossed the line from assist to decide. Keep outputs descriptive, not evaluative.
Sources
EEOC — Technical Assistance Document on Title VII and AI in Employment SelectionAnthropic — Building with Skills (Long-form Docs)

Employee-Lifecycle Skills: Handbook Q&A and Onboarding Checklists

Lifecycle Skills serve employees directly, so accuracy and traceability matter more than speed. A Handbook Q&A Skill lets employees ask questions in plain language and get a cited answer from the current handbook — with a link back to the source paragraph.

The pattern is straightforward. Drop the handbook, benefits summary, and PTO policy into the Skill folder as reference files. Instruct the Skill to answer only from those sources and to say "I don't see that in the handbook — please contact people-ops" when it can't. That last line is the single most important sentence in the SKILL.md.

The Onboarding Checklist Skill takes a role and start date and generates a personalized checklist — hardware, accounts, training, first-week meetings — from your standard template. It is boring, which is the point: consistency for the new hire, less coordination overhead for HR.

  • Handbook Q&A Skill: cites the paragraph, and admits when it does not know
  • Version the handbook in the Skill folder so answers match the effective policy
  • Onboarding Skill: role card in, checklist out, assignable to a hiring-manager runbook
  • Do not let the Handbook Skill answer benefits questions that require plan-year specifics — route those to the broker
  • Log every Q&A interaction — you'll want a record if a policy dispute comes up
Sources
Simon Willison — Notes on Claude SkillsSHRM — Managing Employee Handbooks

A 90-Day Rollout Plan for an SMB HR Team

A defensible Claude Skills rollout for an SMB HR team takes about 90 days from decision to daily use. The first 30 days is inventory and policy; the second 30 days is the two lowest-risk Skills; the third 30 days is the summarization Skills with legal review.

In the first month, list every recurring document HR produces and every question HR answers more than once a month. Adopt or update an AI use policy that names Claude Skills and reaffirms the "assist, don't decide" rule. Assign a Skill owner — usually the HR lead — and a legal or compliance reviewer.

Month two: ship the JD Skill and the Handbook Q&A Skill. These are low-risk, high-frequency, and easy to test. Month three: add the Resume Summary Skill, the Interview Notes Skill, and the Onboarding Checklist Skill after legal signs off on each SKILL.md.

  • Days 1–30: inventory, policy update, assign Skill owner and reviewer
  • Days 31–60: ship JD Skill and Handbook Q&A Skill; measure time saved per JD
  • Days 61–90: ship Resume Summary, Interview Notes, Onboarding Checklist; log every use
  • Quarterly: adverse-impact audit on any Skill that touches candidate selection
  • Store SKILL.md files in a shared drive with change history — this is your compliance record
How we researched this guide: Layer3 Labs reviewed Anthropic's public Skills docs, the EEOC's AI-in-employment technical assistance documents, SHRM's 2025 state-of-the-workplace research, and our own SMB HR client rollouts through Q2 2026.
Sources
EEOC — Artificial Intelligence and Algorithmic Fairness InitiativeAnthropic — Skills Announcement

Frequently Asked Questions

  • No, and they should not. The EEOC's guidance treats employers as liable for AI-assisted employment decisions, so the tool must assist a human reviewer rather than replace one. Design every HR Skill so the output is a summary or draft, and a qualified person makes the advance, hire, or reject call.
  • A Skill is a packaged how-to — a folder with instructions and templates Claude loads on demand. An MCP server is a live protocol that gives Claude access to tools or data at request time, like a real ATS integration. For most SMB HR teams, Skills are lower-lift because they don't require running infrastructure.
  • Yes. Before any Skill touches a candidate or employee, publish an AI-use policy that names the "assist, don't decide" rule and disclosures required in your states. This is both an EEOC and an ADA-accommodation posture, and it is a prerequisite in a growing list of state laws including New York City's Local Law 144 and Illinois's AI Video Interview Act.
  • No. A Handbook Q&A Skill handles the common questions — PTO, holidays, standard policy — and cites the handbook paragraph it drew from. It should refuse questions requiring plan-year benefit specifics, comp adjustments, or accommodation decisions and route those to a human. That refusal is the design, not a failure.
  • Run the Skill on a de-identified sample of the last quarter's applicants and compare the summaries' emphasis across demographic groups where you have data. Look for systematic under-weighting of experience proxies that correlate with protected traits. Document the test — the EEOC expects employers to monitor selection tools they use.
  • Templates, published handbooks, JDs, and de-identified examples are safe. Do not embed live PII — names, SSNs, salary data tied to individuals — in the Skill files themselves. If a Skill needs to process PII at runtime, use Claude for Business or Enterprise where data-use terms and retention settings meet HR confidentiality standards.
  • Start with the JD Skill. It is high frequency, low risk, produces a document a human edits, and forces you to write down your house voice and legal disclaimers in a way every future Skill will reuse. Handbook Q&A is the natural second Skill because it teaches your team the "answer only from source, admit when unknown" pattern.
  • Not for most HR Skills. A Skill is a folder with a Markdown file and templates — an HR ops lead can write one. You may want engineering or a consultant when the Skill needs Python scripts, integrations to an ATS, or automated logging for compliance. Layer3 Labs typically pairs an HR owner with a light technical implementer for the first two Skills.

Build Your HR Team's First Three Claude Skills With Layer3 Labs

We help lean SMB HR teams stand up Claude Skills that pass an EEOC-aware review — JDs, handbook Q&A, and screening summaries with human decisions kept where they belong. Book a free 30-minute AI workflow audit and we'll map your first three Skills.

Book Your Free 30-Min AI Audit