Skills in Claude

The native home of the Agent Skills format: a Skillagram skill already is a Claude skill — it imports as-is.

The short version

Claude is where the skill format was born. A Skillagram skill is a Claude skill: same SKILL.md file, same structure. Nothing to convert, nothing to rebuild — you download it, enable it, and Claude uses it.

This is the opposite of the closed verticals (Harvey, Legora, Lexroom), where a skill has to be rebuilt by hand inside the platform. Here, "import" and "export" are real operations: you bring the file, and it works.

What Claude calls a "skill"

Exactly what you'd expect: a reusable, shareable workflow written as a SKILL.md — a header with name and description (the only two required fields), plus the step-by-step instructions Claude follows, and any supporting material (clause banks, examples, templates) in folders next to the file. It's the same object you use in the other open tools.

Claude surfaces skills in several contexts, each useful for a different need:

How it actually works, under the hood

A skill in Claude loads only when it's needed. Claude reads the description of every skill you've enabled; when a message of yours matches, it loads the skill's body and follows it. Until then, it costs almost no context. That's why you can keep many enabled without weighing the conversation down.

Two triggering modes, complementary:

• Automatic — you write it once with a good description, and Claude calls it on its own when it recognizes the task.
• Explicit — you call it yourself, so you're sure it gets used in that reply. In the Claude apps you force it in the conversation; in Claude Code you launch it with /skill-name.

Importing and using a Skillagram skill in Claude

In the Claude apps (web and desktop):

1. Download the skill's folder from Skillagram — the SKILL.md file and any supporting material. If the skill has multiple files, put them in a .zip with the folder inside.
2. Open Settings → Customize → Skills and upload/enable the skill in your workspace. (Exact menu labels can change: if you can't find them, search "Skills" in settings.)
3. Use it: let it fire on its own when the task matches, or call it explicitly to force its use.
4. To work on a specific matter, open a Project, load the skill and the case documents into it, and work in there: the context stays isolated to that matter.

In Claude Code (technical tier):

1. Put the skill's folder in the right place. To use it everywhere: ~/.claude/skills/skill-name/SKILL.md. To limit it to a single project/repository: .claude/skills/skill-name/SKILL.md inside that project.
2. Claude Code detects the skill (changes to the ~/.claude/skills/ folder are picked up even mid-session; for a brand-new skills directory, restart the session).
3. Use it: it triggers on its own when the task matches, or you launch it with /skill-name.

The right intuition: Claude Code and Claude Cowork are agentic. They don't just use the skill — they can read it and propose the changes, adapting it to your context. For example: "Read the skill in ~/.claude/skills/due-diligence/SKILL.md. It's written for a common-law context. Propose the changes to adapt it to Italian law, flagging every legal reference to verify, and show me the diff before applying it." Claude reads the file, proposes the changes, and shows them to you for approval — without you rewriting anything by hand.

Claude Cowork: the agent on your desktop

Claude Cowork is the agentic version of Claude built for people who don't work from a terminal: it lives on your desktop, reads and edits the files in the folders you point it at, and carries out multi-step tasks without you coordinating each step. It's the same engine as Claude Code, but for office work — research, document preparation, file organization. Skills work the same way: same SKILL.md, automatic triggering or adaptation on request. For a lawyer who wants the agent but doesn't want to learn a terminal, it's the way in.

Sharing a skill within your firm

Sharing happens at the workspace level, under admin control. On a Team (or Enterprise) plan, an admin authorizes skills organization-wide and decides who can share and install them. Once authorized, a skill can be distributed to colleagues without each person rebuilding it: it's how you standardize a way of working across the whole firm, instead of passing it around by word of mouth.

To publish on Skillagram, you export the SKILL.md file directly: being an open format, another professional imports it into their own Claude (or ChatGPT, or Codex) with no adaptation.

To confirm for your case: the availability of the Skills feature, the plans that include it, and the exact menu labels change over time. Custom skills are available on the paid plans (Pro, Max, Team, Enterprise), and Projects require at least a Pro or Team plan. If you're a solo professional and don't see the option, it's almost always a matter of plan or workspace settings — not a mistake on your end.

The caveat that always holds

A skill makes Claude consistent, not correct. It will faithfully apply even a flawed method, with the same reliability it applies a good one. Every output must still be re-read by a lawyer before it reaches a client or a courtroom — and, if the skill was written for another jurisdiction, its sources and legal references must be adapted before you rely on it. It's not a limit of the technology: it's the profession.

Product features change quickly. If Claude's interface has moved since this was written, check Anthropic's official documentation or your workspace admin for the current steps.