How Hospitality Teams Are Using Agent Skills in 2026

The Hospitality Labor Crunch Meets Agent Skills

Hospitality is running out of people, and the numbers make the case better than any consultant could. Labor costs climbed more than 11% year-on-year in 2025, and 65% of hotels now report staffing shortages — a squeeze that industry analysts describe as turning AI agents from an efficiency play into a strategic necessity.

The intent to spend is already there. In Canary Technologies' 2026 survey, 82% of hotel decision-makers said they plan to increase AI investment within the next 12 months. EHL's 2026 trends report describes "agentic AI" that manages repairs, orders parts, and adjusts inventory autonomously, freeing staff for the human side of hospitality.

Into this gap has stepped a specific technology: agent skills. These are packaged instruction files — typically a SKILL.md document — that teach an agent how to execute a defined procedure, from revenue forecasting to guest-message triage. The concept went mainstream after Anthropic launched the Agent Skills open standard in December 2025, and it has already reached hospitality in concrete forms. HospitalityOS ships a ten-skill Hotel Intelligence Suite covering revenue management, labor optimization, and owner reporting, while inHotel builds its entire product around Skills as software-independent expert workflows that agents execute.

But adoption is outpacing verification, and that's the tension every hospitality team needs to sit with. A large-scale analysis of 42,447 public skills found 26.1% carry at least one vulnerability, and OWASP's first Agentic Skills Top 10 — published in August 2026 — ranks Malicious Skills as risk #1. In one July 2026 attack, Trojanized skills impersonating a legitimate platform each racked up more than 300,000 installs before anyone noticed.

What the early hospitality "skills" products share is a near-total absence of independent evidence. ROI figures like inHotel's claimed 10–15X return are self-reported vendor numbers, not measured outcomes. Meanwhile, the broader research is clear that how skills are chosen matters more than how many are installed: curated, human-authored skills lifted agent pass rates by an average of 16.2 percentage points, while self-generated skills provided no consistent benefit.

That combination — urgent demand, fast-moving supply, thin verification — is exactly the gap SkillzBench is being built to close. Its upcoming directory will independently scan, version-pin, and benchmark skills in a sandbox, publishing a security and quality trust score alongside a measured Lift figure for every skill, with the free MCP connector letting teams wire verified skills into any agent. For hoteliers about to trust agents with guest data and revenue decisions, that kind of evidence is no longer optional.

Three Ways Hospitality Is Actually Deploying Agent Skills

In 2026, hospitality teams aren't asking whether to use agent skills — they're choosing between three very different deployment models, each with its own tradeoffs.

The first pattern is the purpose-built skill marketplace. HospitalityOS, for example, ships a 10-skill Hotel Intelligence Suite covering revenue management, forecasting, P&L analysis, labor optimization, reputation intelligence, and more, installable directly into Claude via a plugin marketplace command. These skills are designed to connect with hotel systems like Opera, Stayntouch, and QuickBooks, turning a general-purpose agent into a hotel-operations specialist. It's worth noting, though, that the "enterprise-grade" positioning here is self-reported by the vendor, not independently verified.

The second pattern treats skills less like plugins and more like living SOPs. inHotel has built its entire product model around "Skills" as software-independent expert workflows that agents execute — the idea being that your operational standards survive even if your tech stack changes. The company claims 10–15X ROI without adding headcount, but these are unverified vendor figures, so treat them as marketing until third-party data exists.

The third pattern is simply the broader standard catching up. Anthropic launched the Agent Skills open standard in December 2025, and Microsoft adopted it in VS Code and GitHub within weeks, with Atlassian, Canva, Figma, and Stripe all shipping skills. Hospitality teams increasingly plug into this general ecosystem rather than waiting for industry-specific tools.

• Marketplace skills — pre-built hospitality packs like HospitalityOS's suite, fast to install but dependent on vendor quality.
• Skills-as-SOPs — platforms like inHotel that encode your own workflows, keeping standards portable across software changes.
• General-standard skills — the open SKILL.md format now backed by Anthropic, Microsoft, and other major vendors.

Across all three patterns, one pairing is becoming the default: skills plus MCP. The skill encodes the procedure — how to handle a guest complaint or rebalance a housekeeping schedule — while an MCP server connects the agent to live PMS, POS, and CRM data. As Sentry's David Cramer puts it, "If skills teach you to cook, MCP provides the instruments that let you do it." Block's engineering team describes the same relationship: skills are like GitHub Actions YAML, and MCP is the runner.

Whichever pattern a team picks, the vetting question is the same. With roughly 26% of analyzed skills carrying at least one vulnerability, installing a hospitality skill pack sight-unseen is a real risk — especially when a single skill can reach hundreds of thousands of installs before anyone catches a problem. That's the gap SkillzBench is building to close: a free directory that security-scans, version-pins, and benchmarks skills so teams can see measured Lift before anything touches their agent.

Why More Skills Isn't Better — and Why 26% Are Risky

Hospitality teams racing to install every available skill are about to hit two walls at once: a quality ceiling and a security floor. The same research that shows skills can lift agent performance also shows that how you build and curate them determines whether they help or hurt.

The numbers tell a clear story. According to O'Reilly's analysis of SkillsBench data, curated, human-authored skills raised average agent pass rates by 16.2 percentage points across 84 tasks. Self-generated skills — where the model writes its own instructions — delivered no consistent benefit in any configuration tested. The takeaway: "models cannot reliably author the procedural knowledge they benefit from consuming."

Library size matters just as much as authorship. Atlassian recommends fewer than five skills per Rovo agent. Flat libraries past roughly 80 skills cause routing collapse, where the agent fails to pick the right skill reliably. Focused skills with two to three modules consistently outperformed comprehensive documentation. For a hospitality operation, this means picking a handful of high-value workflows (guest messaging, revenue management, housekeeping scheduling) rather than installing every plugin a marketplace offers.

The security picture is starker still. A large-scale analysis of 42,447 skills found 26.1% carry at least one vulnerability; a separate audit put the figure at 36.82%, with 280+ skills documented leaking API keys and PII. OWASP's first Agentic Skills Top 10, published in August 2026, ranks Malicious Skills as the number-one risk.

Real attacks are already happening. In July 2026, a campaign impersonating the Paperclip platform pushed Trojanized skills that each exceeded 300,000 installs before detection, according to Dark Reading's coverage of the OWASP findings. Skill repositories grew by at least a third in the six months prior, giving attackers more surface area to exploit. As OWASP project co-lead Niv Hoffman put it: "we do not recommend to forbid… people from using them, but to use them in a secure-by-default environment."

For hospitality teams, the practical implication is a vetting checklist before any skill touches a PMS, POS, or CRM:

• Pin every skill to a specific commit SHA so updates are intentional, not silent
• Demand a security scan result and license check before installation
• Require evidence of Lift — measured pass-rate improvement with versus without the skill
• Maintain an inventory so a flagged skill can be removed across all agents at once

This is the gap that independent verification fills. SkillzBench is building a free directory that crawls and deduplicates skills from GitHub, pins every version to a commit SHA, security-scans and license-checks each one, and benchmarks every skill in a sandbox to publish a Lift score — always shown separately from ratings, with no advertising money influencing results. The MCP connector and per-agent loadouts let teams toggle exactly which skills each agent sees, so a front-desk agent and a revenue manager never share a skill surface by accident.

A Practical Adoption Playbook for Hospitality Teams

The gap between "we should use AI agents" and "our agents actually work" comes down to process. Hospitality teams seeing real results in 2026 aren't installing everything they find — they're following a deliberate adoption sequence that the research now makes clear.

Start with a handful of high-value workflows. Labor costs rose 11%+ year-on-year and 65% of hotels report staffing shortages, so target the workflows where agents demonstrably move numbers: guest messaging (AI agents resolve up to 80% of inbound inquiries without human involvement), revenue management (AI-driven systems lift RevPAR by 15–20%), and housekeeping scheduling (BCG research found a 20% reduction in room turnaround time). Resist the temptation to build a big library — Atlassian recommends fewer than five skills per agent, and flat libraries of 80+ skills cause "routing collapse" and unreliable retrieval.

Encode your own SOPs — don't let the model write them. The SkillsBench research found curated skills lift agent pass rates by an average of 16.2 percentage points, while self-generated skills provided no consistent benefit in any configuration tested. Your front desk escalation rules, your check-in sequence, your comp policy — that's the organizational knowledge that makes a skill valuable. Write it as a SKILL.md file from your actual procedures, not a model's guess at them.

Vet every skill before it touches an agent. This is non-negotiable: an analysis of 42,447 skills found 26.1% carry at least one vulnerability, and OWASP's first Agentic Skills Top 10 ranks Malicious Skills as the #1 risk. Before installing anything, check that it's version-pinned to a commit SHA, security-scanned, and license-checked — and keep an inventory so a flagged skill can be pulled from every agent at once.

Pair skills with MCP for live data. The emerging default is "use both": the skill encodes the procedure, and an MCP server connects it to your PMS, POS, and CRM. As Sentry's David Cramer puts it, "If skills teach you to cook, MCP provides the instruments that let you do it."

Benchmark before and after — don't trust vendor ROI. inHotel claims 10–15X ROI, but these are self-reported figures with no independent verification. Run each task with and without the skill, on your own workflows, and measure the difference yourself.

This is exactly the gap SkillzBench is building to fill: a free, independent directory where you can browse skills, see each one's security and quality verification, review its published Lift score from sandbox benchmarking, and connect vetted skills to your agents through a free MCP connector. Join the waitlist to get access at launch — and start your adoption playbook on evidence, not vendor promises.

CTA: Join the SkillzBench waitlist — free skills directory, free MCP connector, published Lift scores.