Best AI Agent Skills for HR Managers in 2026

HR managers aren't the obvious user for AI agent skills. Recruiters, sure, they live in candidate databases. But the average HR generalist spends most of her week drafting policies, answering benefits questions, reviewing performance docs, and writing Slack announcements that won't get her dragged in #general.

It turns out Claude Code with the right skill set is unusually good at that work. Below are 8 agent skills HR teams have installed in 2026, what each one does, and how to set it up.

A quick definition before we go in: an agent skill in Claude Code is a folder with a SKILL.md file that gives Claude focused instructions and (often) a set of helper scripts. You install one by adding the skill folder to ~/.claude/skills/. The skill loads only when relevant, your other work isn't affected.

1. Internal Comms, for writing announcements without sounding like a robot

The internal-comms skill is built for drafting professional internal communications, announcements, policy changes, memos, leadership updates, all-hands recaps. It's the single skill I'd install first for any HR generalist.

What it does:

• Drafts in your company's tone (you teach it once with 3-4 past examples)
• Handles structure: TL;DR at top, key changes in the middle, FAQ at the bottom
• Generates a Slack-ready version and an email-ready version side by side
• Flags when language is ambiguous, dismissive, or likely to cause a furious DM

Install:

cd ~/.claude/skills
git clone https://github.com/anthropics/internal-comms-skill internal-comms

Use it for: layoff announcements (carefully), benefits enrollment, holiday schedule changes, return-to-office updates, sensitive policy rollouts. The skill is conservative, it tells you when a draft probably needs legal review before sending.

2. Doc Co-authoring, for drafting and revising long policy documents

The doc-coauthoring skill is for the long stuff. Employee handbooks. PTO policies. Performance review guidelines. The 30-page documents that take 2 weeks to update and 4 rounds of feedback to finalize.

What it does:

• Reads your existing handbook and proposes targeted edits (not full rewrites)
• Maintains consistent tone across hundreds of pages
• Tracks revision history with rationale notes so you remember why a clause changed
• Flags conflicting policies (vacation rollover that contradicts the leave policy)
• Generates a redlined version for legal review

The skill is especially helpful when you inherit a handbook from a predecessor and need to update it without rewriting from scratch.

Install:

cd ~/.claude/skills
git clone https://github.com/anthropics/doc-coauthoring doc-coauthoring

3. Deep Research, for benefits, compensation, and compliance research

The deep-research skill turns Claude into an analyst. You give it a question — "What does median engineering compensation look like at 100-200 person SaaS companies in the Bay Area?" — and it returns a structured report with sources.

The 8-phase pipeline (defined in the skill) hits multiple sources, cross-references claims, flags contradictions, and produces a final brief with citations you can verify. For HR research, this replaces 4-6 hours of manual Glassdoor + Levels.fyi + Payscale tab juggling.

What I use it for:

• Compensation benchmarking before salary review cycles
• Researching new benefits (fertility coverage, gender-affirming care, mental health stipends)
• Compliance research when expanding into a new state or country
• Vendor comparison research before signing a 3-year contract

It's overkill for quick questions. Save it for the research that would otherwise eat half a day.

4. Slack MCP, for monitoring and responding to HR channels

The slack-mcp skill connects Claude to your Slack workspace. For HR, the useful pattern is:

• Monitor your #hr-questions channel
• Auto-summarize the morning's questions before standup
• Draft initial responses to FAQ-type questions (benefits enrollment, time-off policies)
• Flag urgent or sensitive threads for you to handle personally

Note on privacy: don't use Slack MCP to read DMs or private channels without explicit team buy-in. The skill respects channel permissions but the surface area for accidentally exposing sensitive info is real. Discuss with your security team first.

Install:

claude mcp add slack -- npx @anthropic-ai/slack-mcp

5. Notion MCP, for managing HR documentation

If your HR team lives in Notion (and most do in 2026), the notion-mcp skill is non-negotiable. It lets Claude search, read, and update your Notion workspace.

Useful HR workflows:

• Pull the latest version of the employee handbook into a chat for reference
• Update onboarding checklists in bulk when policies change
• Generate weekly HR metrics dashboards from your candidate tracker
• Sync interview feedback from a scratch doc into the official candidate page

The Notion API has rate limits, don't run scripts that do thousands of operations at once. The skill handles batching but watch your usage in the Notion admin panel.

6. PDF Tools, for handling offer letters, contracts, and policy archives

HR teams generate and receive massive amounts of PDFs, offer letters, signed contracts, background check reports, benefits enrollment forms. The pdf-tools skill handles them.

What it does:

• Extracts text and tables from PDFs (including OCR for scanned documents)
• Merges multiple PDFs into a single onboarding packet
• Adds watermarks to draft offer letters
• Searches across hundreds of policy PDFs in seconds
• Splits large documents (a 200-page handbook into individual policy chapters)

Real use case: I had a stack of 80 signed contractor agreements from 2023-2025 in PDF format and needed to find every contract that mentioned a specific IP clause. With pdf-tools, this was a 4-minute task. Manually, it would have been a full day.

7. Claude Supermemory, for institutional knowledge across sessions

The claude-supermemory skill solves a specific HR problem: every conversation starts from zero. You explain your company's PTO policy to Claude on Monday. On Tuesday, you have to explain it again.

Supermemory builds a knowledge graph from your past conversations. Once you've told it your benefits provider is Carrot for fertility and Lyra for mental health, it remembers across sessions. After 2-3 weeks of use, Claude knows your company's policies as well as a 6-month-tenure HR generalist would.

Install:

cd ~/.claude/skills
git clone https://github.com/supermemoryai/claude-supermemory claude-supermemory

Be careful with sensitive data. Supermemory stores its knowledge graph locally by default, but verify that, don't let employee data sync to a cloud service without HR-approved encryption.

8. Marketing Skills, for the EVP and employer branding side

The marketing-skills skill bundle (36 sub-skills) includes pieces that HR uses constantly even though the bundle is positioned for marketers. The relevant ones:

• Copywriting for job descriptions, careers page copy, recruitment email templates
• CRO for optimizing your application funnel (where are candidates dropping off?)
• Brand voice for keeping HR comms consistent with company tone
• Launch strategy for announcing new benefits or HR programs

If you're an HR team of one and you're also handling employer brand, this bundle covers a lot of work that would otherwise need a marketing partner.

How to set up your HR skill stack from scratch

If you're starting today, here's the order I'd install:

1. internal-comms (immediate value, 15-minute install) 2. pdf-tools (covers the daily document work) 3. doc-coauthoring (for the next handbook update) 4. deep-research (for the next benefits or compensation review) 5. notion-mcp (if you use Notion) 6. slack-mcp (after you've talked to security) 7. claude-supermemory (after 2 weeks, when you're tired of re-explaining) 8. marketing-skills (when you have a project that needs it)

The install time for all 8 is about 2 hours, mostly waiting for git clones and configuration. The payoff starts in week 1.

A note on confidentiality

HR data is sensitive. Personnel files, compensation, medical accommodations, separation agreements, none of this should be sent to a cloud AI tool without verifying the data handling policy. Claude Code with local skills keeps most of the processing on your machine, which is better than ChatGPT's default behavior, but verify before processing anything PII-heavy.

If in doubt, anonymize before sharing. Replace employee names with role descriptors. Strip salary figures from compensation reviews. The skills work fine on anonymized data.

FAQ

Q: Do I need to be technical to install these skills? A terminal and basic git knowledge is enough. Each skill has a README with the install command. Set aside an hour and you can install all 8.

Q: Are these skills free? Yes. The skills themselves are open source. You're paying for Claude Code (which has a generous free tier) and any underlying services the skills connect to (Slack, Notion).

Q: What if my company forbids AI tools? Check which specific tools are forbidden. Many companies allow Claude Code with restrictions (no PII, no compensation data) but ban ChatGPT entirely. The compliance posture is usually about data handling, not the AI itself.

Q: How do these compare to enterprise HR tools like Workday or BambooHR? They're not competing. Workday and BambooHR are systems of record, they store your employee data. Agent skills are work multipliers, they help you do the writing, research, and analysis work around the data. Use both.

Q: Which skill should I install first if I only install one? Internal-comms. It pays for itself in the first announcement you draft.