Skip to main content

Skills (concept)

A skill is a unit of capability the agent uses when relevant. Think of it as a documented competency — "build a presentation," "read a PDF," "generate a spreadsheet" — that the model knows how to trigger.

This page is the concept. To create, test, and publish skills, see the Skills section.

Skill ≠ command ≠ file mention​

Imaginne distinguishes three things:

TypeHow it appearsWhat it is
Skillthe agent triggers it when relevantA capability the model uses when it makes sense.
Command/somethingAn interface (host) command, like /model or /clear.
File mention@fileA reference to a file in your workspace — the agent will read it.
@ is for files, not a "function"

@ points to a file in the workspace (as in AI editors). Functions, on the other hand, are triggered like skills — by the model, when they make sense — not by a prefix you type.

Where skills come from​

  • Bundled (always available). The document skills — docx, excel, pdf, ppt_skill — come built in and are installed automatically. See Bundled skills.
  • Published by the organization. Created in the Skill Studio and synced to clients according to your policy.
  • Personal / project. Skills you place in ~/.imaginne/skills/ or in the project's .imaginne/skills/.

How the agent triggers a skill​

You don't "run" a skill manually — you describe what you want and the agent decides whether and which skill to use. The available skills are presented to the model as capabilities; when one is relevant, it triggers it. The organization controls, via policy, which skills enter that list for you.

Local execution — normal vs protected​

Skills run on your machine (on the local surfaces). There are two modes:

ModeWhat it means
Local normalThe skill's files sit openly in ~/.imaginne/skills/. Transparent and easy to inspect.
Local protectedOnly a summary sits on disk; the content arrives in a verified bundle (signature + digest), extracted to a temporary folder, executed, and removed. It protects against casual inspection and redistribution — but it is not DRM.

Details in Execution modes and Protected skills.

Secrets a skill needs​

A skill can declare required_env — the names of the secrets it needs (an API token for an internal system, for example). The organization registers those secrets and the Engine injects only the declared names, only for that execution. See Env-secrets.

See also​