DocumentationSkills for enterprise AI

Create your own skill

In anymize the AI writes you a skill in chat; you can also bring one as a file, a folder or from a GitHub repo.

5 Min

The fastest way: say it in chat

anymize has a built-in Skill Creator. It is always there, even if you have not installed a single skill yet. You do not search for it; you just talk to it.

At the end of a conversation that went well:

Save this as a skill. Call it “Client letter”
and make sure it always starts with the deadlines overview.

From that the AI writes name, description and instructions, saves the skill and activates it immediately in this chat. After that it lives under Apps and there Skills in “My Skills”. No editor, no format you have to learn.

If your skill limit is reached, the AI says so in chat and creates nothing. Deactivate a skill you do not need right now and ask again.

What belongs in a good skill

A skill has three parts. The most important is the description, because that is how the AI later recognizes when the skill is due.

PartWhat belongs in it
NameShort and unambiguous, so you find it again in the slash menu.
DescriptionWhat the skill does and when it should be used, ideally with the words you would type yourself.
InstructionsThe actual flow: order, checks, outline of the result, typical mistakes, an example.
  • Write the instructions the way you would give them to a new colleague.
  • Name the desired result precisely: headings, order, length.
  • Also say what the skill should not do. Boundaries save corrections later.
  • One example in the skill does more than three extra rules.

Bring a finished skill

If you already have something that works, you do not have to rewrite it. Under “My Skills” the Import button leads to four paths.

PathWhat it is for
MarkdownYou paste a SKILL.md or pick the file. The open standard that other tools also read.
ZipA whole skill package with reference files and attachments.
FolderA folder from your computer, as it is.
From GitHubPaste the repo address. If it points at a whole repo, anymize finds every SKILL.md in it.

For skills that still change there is cloud sync: you point at a folder in Nextcloud, Google Drive or OneDrive, and anymize pulls the changes on a fixed cadence, from every half hour to daily.

Why is my GitHub import stuck?

Usually it is not anymize, but a counter at GitHub. We fetch public repos without signing in, and for those anonymous requests GitHub allows only 60 calls per hour. That quota sits on our server address, so it is shared. A repo with many files can use it up on its own.

  • Message “Too many imports in a row”: wait a moment, then start the import again.
  • Better: connect your GitHub account once. Then the quota counts against your account, and private repos come along.
  • Very large repo: give the exact path instead of the repo root, for example owner/repo/tree/main/my-skill.
  • If your organization requires SSO or does not allow third-party apps, anymize says so in the dialog and links to the approval.

Change, rename, pass on

Your own skills stay in your hands. Under “My Skills” the card menu opens the edit dialog: on the Details tab you change name, description and icon, on the Files tab the instruction text and every bundled reference file.

That also works in conversation. Ask the AI to change a paragraph in your skill and it rewrites the file for you. It may do that only for skills you own, and only when you ask. If the same skill was changed elsewhere in the meantime, anymize prefers to stop and say so rather than overwrite the other change.

  • Share by email: the recipient gets an invitation and can accept the skill.
  • Share with a group or the whole team: more under “Distribute skills in the team”.
  • Export: takes the skill with you as a file, to another tool or into an archive.
  • Delete: removes it for you. Anyone who already installed it does not lose their copy.

A skill is instruction, not a case file. Do not write client data, HR matters or citizen data into it - write the method. You add the case itself in chat, and that is where anonymization applies as usual.