What is an Agent Skill?
The concept and anatomy of an Agent Skill — the same open format behind Claude Skills: what's inside a SKILL.md, why skills beat prompts, and how the standard connects Agentman and Claude.
Agent Skills Help
Step-by-step guides for creating, importing, sharing, and governing your team's skills — and using them from Claude, ChatGPT, and Cursor.
The concept and anatomy of an Agent Skill — the same open format behind Claude Skills: what's inside a SKILL.md, why skills beat prompts, and how the standard connects Agentman and Claude.
Create a skill in the wizard, edit it in the full-screen editor, and publish it — or just describe it to Claude and let it write the skill for you.
Bring your Claude Skills into your team library — drag in one .skill package or fifty, or ask Claude to move them for you over MCP.
Connect your team's skills library to Claude from the Connectors Directory or a custom connector URL, with per-member OAuth and the Skills Explorer.
The same skills library, in every AI tool your team uses — one connector URL for Claude Code, ChatGPT, and Cursor.
Share a skill the way you'd share a doc — by name or email, with one of four access levels: use, read, edit, or admin.
Make a skill available to everyone on the team in one click with General access — it moves to the Team Skills bucket for the whole workspace.
Grant use-only access so teammates — or clients — can run your skill without seeing the instructions. Every use is logged with who ran it.
The four shelves of your library and what each one means for isolation and sharing — plus why publishing a skill never makes it public.
Add a teammate to your workspace so you can share skills with them — invite by email right from the share dialog, with seats managed by plan.
Hand a skill to a new owner when someone changes roles or leaves — versions, notes, sharing, and activity history all carry over.
Workspaces are the hard confidentiality boundary — separate libraries, members, permissions, and logs — and Claude can connect to several side-by-side.
Google-Docs-style version history for your Agent Skills — day-grouped versions with authors and notes, plus a highlighted diff between any two versions.
Leave a short note on a version — what changed and why — so your skill's history reads like a changelog the whole team can trust.
Roll back safely: restoring copies an older version's content and files into a new version — nothing is deleted, and history stays intact.
The skill lifecycle: drafts accumulate edits freely, publishing freezes an immutable live version, and deprecating retires a skill (and frees a Free-plan slot).
The Activity tab shows every application of your skill — total applies, distinct users, and a newest-first event trail with each person's identity.
One log across every skill in the workspace — filter by skill, member, or event type to answer 'who used what expertise, when.'
Search your library by name and tags — or ask in plain language ("how do we appeal a denial?") and let semantic search rank what's relevant.
Bundle related skills into a named set — like a new-hire onboarding pack — and attach the whole set to an agent, or browse it from Claude.
Copy any skill you can read — including system skills — into an editable copy with source attribution, even into another workspace.
Bind skills to your Agentman agents — the wizard recommends matches for the agent's goal, and published updates flow to every bound agent.
Download any skill as a portable .skill package — the open SKILL.md format — so your team's expertise is never locked in.
What's metered (published skills, not drafts), what each plan includes, and the size limits that matter — answered in one place.