Skills have quietly become the way people codify their expertise for AI. Claude made them mainstream — describe how you do a task once, and every future conversation does it your way. If you've built a few, you know the feeling: the skill is genuinely yours, and it makes the model genuinely useful.
Then you hit the wall we kept hearing about: skills live in one person's account. Your best analyst's playbook helps your best analyst. The partner's contract framework reviews the partner's contracts. Everyone else re-invents, re-prompts, and gets different answers to the same question.
So we rebuilt Agent Skills around one idea: your skills deserve a team library. Today we're relaunching it — six weeks of shipped work, live now.
What's new
1. Google-Docs-style sharing. Share a skill the way you'd share a doc: search a teammate by name or email, pick an access level, done. Four levels — use, read, edit, admin — granted per skill, or share with the whole workspace in one click. Type an email that isn't in your workspace yet and you can invite them from the same dialog. Ownership transfers when people move on.
2. Use-only access. This is the one we're proudest of. Grant someone use and they can run your skill — apply your playbook, your pricing model, your checklist — without ever seeing what's inside. The skill body is redacted at the API layer, not just hidden in the interface, and every use is logged with who ran it. It's how a consultant lets a client's team run their methodology without handing over the methodology.
3. Version history 2.0. Every revision, in a day-grouped timeline with real author names. Compare any two versions visually, annotate the ones that matter, see which reference files changed in each revision, and restore any version in one click — non-destructively, so nothing is ever lost.
4. Governance. Every skill has an Activity view for its owner: total applies, distinct users, and a newest-first event trail. Workspace admins get one access log across every skill — filterable by skill, user, or event type. "Who used what expertise, when" is now a lookup, not an investigation.
5. A reorganized library. My Skills, Team Skills, and Shared with me — private by default. Semantic search finds skills by what they do, not what someone named them.
6. Portability. Bring the skills you already built in Claude: drag one .skill package or fifty into the import dialog, or ask Claude to move them for you over MCP — it can import from a URL or upload files up to 50 MB. Export works the same way, any time. The format is the open SKILL.md standard, so nothing is locked in.
7. New pricing. Free gets you 3 team members and 10 published skills — with unlimited drafts, and deprecating a skill frees its slot. Team is $10 per seat per month with unlimited members and unlimited published skills, starting with a 14-day trial. Enterprise adds SSO, HIPAA/SOC 2/GDPR compliance, and private deployment.
And you keep working in Claude
None of this asks you to change tools. One MCP connection — skills.agentman.ai — puts your team library inside claude.ai, Claude Code, ChatGPT, and Cursor, with per-member OAuth. The same skill that ran in your Claude account yesterday runs there today; the difference is your whole team has it, it's versioned, and every use is accounted for.
If you've been using Claude Skills solo and want your team on them, start here: the team library for your Claude Skills. Or just start free — import a skill, share it with one teammate, and see the difference a library makes.
We'll keep shipping. Follow along on the changelog.
Claude is a trademark of Anthropic, PBC. Agentman is an independent platform and is not affiliated with or endorsed by Anthropic.
