Agent Skills: Share skills your team can run — without exposing what's inside.102 production-ready AI skills, ready to clone.
Skills capture how your people actually do the work — then make sure Claude, ChatGPT, and your Agentman agents do it the same way, every time. Built once, shared across your team, and managed in one place. Private by default, shared on your terms.
Also works in Claude Code · Cursor · Agentman agents — through one connection.
Watch: Skills in action (2 min)
Watch how skills make Claude and ChatGPT actually useful
Trusted by Teams at
Customer StoryHow the founder of Nested Bean turned her company’s expertise into its operating system
Forty-plus skills now carry the daily work of a consumer brand — freeing its founder to focus on strategy, quality, and growth.
skills in production across the business
less time to produce a month of multi-channel content — from 3–4 weeks to a single command
faster month-end close — a one-week process, completed in a day
new hires running their first skill
“What used to consume weeks of my team’s time now runs on command — and we spend that time on the things only humans do best: engaging with customers, innovating, raising quality, and thinking further ahead.”
Scaling a founder’s expertise
Manasi Gangan built Nested Bean into one of the most trusted names in infant sleep. Her expertise spans the full business — brand, marketing, operations, finance, and analytics — and like every founder, she could only be personally involved in so much of it in a given week. Growth meant finding a way to apply that expertise everywhere at once.
Skills made that possible. Nested Bean codified how the company works — its brand voice, its forecasting logic, the way its founder reads performance — into more than forty skills spanning advertising, content, sales and operations, and executive analytics. Each one runs the way Manasi would run it, every time, whether or not she’s in the room.
Expertise, architected
Nested Bean treats its skill library as a designed system. Foundational skills establish shared context. Reference skills hold the company’s standards. Processing skills do the work, and visualization skills report it back.
Before any new skill is built, the team maps what it depends on and what will depend on it — the discipline that makes each addition compound the value of the library rather than expand it.
Built in Claude. Governed in Agentman.
Every Nested Bean skill starts as an exciting exploration of possibilities in Claude, where it’s drafted, tested, and refined until the outputs are proven. Once completely validated and human approved, it’s promoted to the Agentman library.
Promotion is the moment a skill becomes a company asset. In Agentman, Nested Bean’s skills are versioned, so an improvement made once reaches everyone. They’re organized into skill sets, so multi-step workflows run in the correct sequence. And the full library is visible in one place, giving the founder a complete view of the company’s intelligence layer as new team members begin authoring skills of their own.
“Pushing to Agentman is never a thoughtless step. It’s the fully vetted version — that’s what keeps our library clean and our architecture sound.”
What changed
A month of content, on command
A full month of multi-channel content once took three to four weeks. It now ships from a single command.
A one-week close, in a day
Month-end reconciliation now finishes in a day instead of a week — with Manasi’s review standards built in.
New hires, productive in week 1
New team members learn the framework and run their first production skill in their first week.
A wider view for the founder
Executive analytics skills give Manasi company-wide performance at a glance, with drill-down on demand.
What’s next
Nested Bean is extending its library from on-demand skills toward fully agentic workflows — the same validated expertise, running continuously.
What are agent skills?
The evolution from one-time instructions to reusable intelligence.
Prompts
One-time instructions. No structure. Different every time you ask.
Skills
Reusable intelligence packages. Decision logic, templates, edge cases. Consistent every time.
Agents + Skills
Agents that work like your best employee. Same judgment, any AI platform.
Skills turn AI from a generalist into a specialist—your specialist. Learn more →
Built for ClaudeBuilt for the way you already use Claude
You use Claude because it does the work — drafting, reasoning, running real workflows. Skills are how you make Claude yours: encode your templates, process, and standards once, and Claude runs them the same way every time.
Claude lets you create a skill. Agentman is the platform around it — where skills get discovered, shared, versioned, and reused across your whole team.
What Agentman adds on top of Claude
| Claude | Agentman | |
|---|---|---|
| Skills library | No shared library — build each skill from scratch, every time. | A searchable My/Team/Shared catalog; install ready-made skills in seconds. |
| Team sharing | A skill you make lives with you, in one chat. | Set who can access at create, import, or clone; non-members get invited to the workspace. |
| Access control | A skill you share is a skill you expose. | Grant use-only so people run it without seeing inside — the body is redacted at the API layer, not just hidden in the UI. |
| Versioning | Re-send files to share an update — teammates drift out of sync. | Change a skill once; everyone who uses it gets the new version. |
| Skill sets | One skill at a time — no way to bundle a workflow. | Bundle related skills and roll out a complete workflow in one move. |
| Workspace management | Built for solo users. | Organize by team, invite members, and control who has access. |
Share what it does. Not how you do it.
With use-only access, teammates and clients can run a skill without seeing what's inside it. Your prompts, decision logic, and hard-won process stay yours — even while the whole team gets the benefit.
Run, don’t read
Grant “Use” and the skill executes normally, but its contents stay sealed. No copying, no leakage.
IP stays protected
The logic you spent months refining isn’t exposed the moment you share it.
Proven, not promised
Security whitepaper and independent technical evaluation available on request.
Learn moreHow it works
Find or build a skill
Grab one from the library or package your own workflow. No code required.
Use it in Claude
Install it and Claude runs it on demand, the same way every time.
Share it
Publish to your team or the community. One click, everyone's in sync.
One library. Private by default.
Every skill starts private to you. Promote it to your team or the wider org when it's ready — you decide what's shared and what stays yours.
My · Team · Shared
A three-tier library. Personal drafts, team standards, and org-wide skills, cleanly separated.
Private by default
Nothing is exposed until you choose to share it. Access is set per skill, not all-or-nothing.
Versioned & permissioned
Improve a procedure once; everyone on that version gets the update. Control who can use, read, edit, and publish.
Built for firms that can't get this wrong.
Permission management and auditing designed for legal, private equity, and healthcare — where every decision needs a paper trail.
Per-skill access control
Grant Read, Edit, or Admin per skill or across the whole workspace — plus Use-only on Business, which lets someone run a skill without reading it. Control who can view, edit, and publish from one place.
Secure OAuth 2.1
AI assistants connect through standard OAuth 2.1, not shared secrets. Every connection is per-member, revocable, and visible to admins.
Use-only sharing
Share a skill teammates — or clients — can run but not read, export, or clone. The instructions stay hidden, and every use is logged with who ran it.
See who did what
Who changed what, when, and which agent used which version on which task. Exportable.
Version history & rollback
Every revision kept. Roll back a procedure in one click.
Confidentiality boundaries
Run one workspace per client, matter, or project — separate libraries, permissions, and logs — and connect them to Claude side-by-side. Context never crosses lines it shouldn't.
Human in the loop
Consequential decisions wait for a person. Always reviewable.
Skills Library
Browse by category
16 categories, each with enough depth to be worth a look. Clone a system skill to get started, then customize it for your organization.
See the difference context makes.
Same request. One without your context, one with it.
Contract Review
Review contracts carefully for liability, payment terms, and termination clauses. Ensure all parties understand their obligations and consider consulting an attorney…
Generic. Textbook. Could be anyone's.
Contract Review
Specific. Actionable. Actually yours.
One connection. Every skill. Any AI tool.
However your team works — in chat, in a workflow, in an app — it all runs on the same skill library.
Connect Claude.ai and ChatGPT through secure OAuth 2.1. Each team member controls their own connections — no shared keys, no copied credentials.
Add Plugins and MCP servers
Settings → Connections → Add MCP Server. Connections use OAuth 2.1 — authorize once, and each teammate manages their own.
Add connections →Ask your AI anything
It loads the right skills automatically — same answers in every tool.
Need something custom?
Build your own skills without code.
Describe what you need in plain English. Our skill builder creates the folder structure, formats the SKILL.md, and organizes your references.
Or clone a system skill and customize it for your organization.
- Visual editor with live preview
- Version control built in
- Team sharing and permissions
- Bulk import — drop in multiple .skill or .zip files at once (native .skill support)
- Clone across tenants, live — roll a skill out to every practice or client workspace in one move
- Import/export — compatible with Anthropic's Agent Skills specification, so a skill written here runs in Claude and Claude Code unchanged
--- name: my-custom-skill version: 1.0.0 description: > One paragraph explaining what this skill does and when AI should use it. category: Your Category tags: - tag1 - tag2 --- # Skill Title ## When to Use This Skill Describe the scenarios where this skill applies... ## Instructions ### Step 1: First Major Step Detailed instructions...
Simple, transparent pricing
Start free and scale as your team puts more of its expertise to work. See the plans and what's included.
Which plan is right for you?
Free Plan
BEST FOR
Individuals & small teams
Capture your skills, version them, and use them from Claude — up to 3 members and 10 private skills, free
Team Plan
BEST FOR
Growing teams
A shared skills library with unlimited members and unlimited private skills — $30/mo for 3 seats
Business Plan
BEST FOR
Teams that need control over each skill
Per-skill access control, use-only sharing, and approvals — $125/mo for 5 seats
Free or 14 day free trial for Team plan
FAQ
Common questions about Agent Skills
An Agent Skill is a structured instruction package that teaches an AI how to do a specific job — the procedures, decision logic, templates, and edge cases your best people handle every day. Skills make a general model behave like a domain expert instead of a generalist.
A prompt is a one-off instruction you paste and lose. A skill is a versioned, inspectable, reusable procedure your organization owns and improves over time — so quality compounds instead of resetting with every conversation.
Skills work across the AI tools your team already uses — Claude, Claude Code, ChatGPT, Cursor, and Agentman agents — over a single MCP connection, so the same codified expertise powers every workflow instead of living in one tool.
Plugins are pre-built bundles of skills, commands, and agents for a whole business function (finance, legal, sales, support) — capabilities your AI can do. MCP servers are secure connections to the tools your team already runs on (Notion, Stripe, GitHub, HubSpot, your EHR) — things your AI can reach. You add both from the Connections page, and they travel over one MCP URL.
Create a free account, then add the plugins and MCP servers you want from Connections. Everything you enable is available through a single MCP URL you paste into Claude, ChatGPT, Cursor, Agentman, or anything that speaks MCP — no local install required.
No. If you can describe how a task should be done, you can capture it as a skill. Start from the library of production-ready skills and adapt one, or author your own from scratch — no code required.
Yes. Skills live in a shared library with versioning, per-skill access control, use-only sharing, approvals, and a record of who did what — built for legal, private equity, and healthcare teams where every decision needs a paper trail.
Access is set per skill. Use lets someone run the skill and apply its output — without seeing the instructions inside it, so your proprietary logic stays protected. Read adds the ability to view the skill’s contents and configuration. Edit allows changing the skill itself, and Admin controls sharing and settings. Every skill is private by default; you choose exactly what to expose and to whom. This is what makes it safe to share a skill across a team, with clients, or across tenants without giving away the IP inside it.
When a job needs hundreds of coordinated steps orchestrated end to end, that is an agentic harness. Our Agentic Harnessing team builds custom agent systems on top of your skills for complex, high-stakes work.
You can start free and scale as your team puts more of its expertise to work. See the pricing page for plan details and what is included.