Skip to main content

Skills

The Skills section manages the instruction packs an agent loads on demand. A skill's name and description stand in the agent's prompt; the instructions themselves arrive only when the agent calls the skill tool for it. The concept — the three sources, the file format, how one is written — is Skills; this page is the screen.

The list​

It shows every skill this installation stores, plus the SKILL.md files in the signed-in user's own skills directory. A skill a project keeps in its .mindconnect/skills/ belongs to a session's working directory and shows up in the chat that works there, not here.

  • New Skill — create one. A name another stored skill already has is refused: an agent looks skills up by name.
  • Delete — on a stored skill's row. A skill the app ships comes back on the next start; switch it off instead to keep it away.
  • A row from a file says so and has no buttons: it is edited where it lies, and this screen has no business writing into somebody's repository or home directory.
  • A skill switched off is listed as (off) and offered to no agent.

The form​

FieldWhat it is for
NameWhat the model passes to the skill tool: lower-case letters, digits and dashes.
DescriptionThe one line the model decides on. Say when to reach for the skill — it is in the prompt from the first token, the instructions are not.
InstructionsMarkdown, handed over whole when the skill is loaded. Write it as long as the work needs; it costs nothing until it is used.
Tools it expectsNamed to the model on load. Grants nothing — the agent's own tools still decide what it may call.
EnabledOff keeps the skill here and offers it to nobody.

Saving checks the version the form was opened with: if someone else saved the skill meanwhile, yours is refused rather than silently overwriting theirs.

Giving an agent skills​

On the agent's edit form (Agents):

  • Enable Skills — adds the skill tool and the prompt section.
  • Skills — the skills this agent may load. Select none and it gets every skill there is, so a skill added later is in reach without touching every agent.

The agent's detail view shows the result as one line: off, all skills, or the names.

Taking a skill to a repository​

/admin/api/skills/{id}/markdown returns any skill as a SKILL.md — front matter and body. Put that file under a project's .mindconnect/skills/ and it travels with the code, and the project's version then wins over the stored one for sessions working there.