Skip to main content

Skills

Skills come with the Celeris desktop app, with nothing to switch on. The Mac App Store version does not include them.

A skill is a way of doing a job that Celeris has learned from you. "Cut the weekly release notes." "Onboard a new contractor." "Turn this spreadsheet into the invoice format finance wants."

You do not write one. You describe the task roughly, Celeris attempts it, and the attempts that miss get patched a step at a time until it holds.

Growing one​

Open Settings → Tools → Skills and describe the task in a sentence or two. Celeris drafts a plan over the tools it already has and runs it.

  • Every step asks for approval where that tool would ask in your own conversation. Your allowlist and "Always allow" apply, and a step you deny stops the attempt. Try it (dry) skips the steps that change things, and the reads it does run still ask first when they touch something sensitive, such as a key or credentials file.
  • The approval says which skill is asking, whether you or a connected agent started it, and who wrote it. A request from a connected agent never offers "Always allow", and one you deny is not asked again straight away.
  • When a step goes wrong, the fix is applied to that step, not to the whole plan.
  • Each clean run counts. When the plan survives a run without needing changes, it graduates.
  • At that moment Celeris offers to name the skill. Name it something you would say out loud; decline and it keeps the name it drafted.

A graduated skill becomes a tool Celeris can call like any other. That is the point of the exercise: the fixed parts stop being something the model re-derives every time, which is both faster and much harder to get wrong. The reasoning is the same one behind tools that do whole tasks.

Skills are files you can read​

Each skill is a plain markdown file in a folder on your computer. It has a name, a description and a body: the procedure, in words.

Open one in any editor and change it. The next time Celeris loads it, your wording is what it uses, and the version goes up. If a skill is nearly right, editing the prose is usually quicker than running it into shape.

Because they are ordinary files, they travel. Copy one to another machine, or send it to a colleague, and it can be imported there. An imported skill arrives as a starting point rather than a proven one, because it has not run in your setup yet, against your tools.

The Skills page​

The page lists what you have, with the things worth knowing at a glance: how far along each one is, its version, the effort level it runs at, and when it last fired. Two details are worth reading correctly:

  • The run count is a clean-run count, and it restarts after a failure. It is a measure of "is this stable right now", not a lifetime total.
  • A skill never carries its own permission. However well-proven it is, the tools inside it are approved by the same rules as everything else. See Approvals & safety.

Packaged skills​

Celeris also picks up packaged skills: know-how written by someone else, such as a house style, a review checklist, or how a particular framework wants things done.

These add knowledge and no operations. A packaged skill can change how Celeris does something. It can never widen what Celeris is allowed to do, and it never turns an approval prompt into an automatic yes.

Browse plugins​

Open Settings → Tools → Plugins to browse installed plugins and discover available ones.

Select a plugin to inspect its skills, connector requirements, tool references, knowledge files, and source before installing it. The tools available on your computer may differ from the tools a plugin names. Connectors may need their own installation or sign-in.

After installation, check the plugin's connections, update it, or remove it from its details. You can also install a plugin from a Git URL.

Removing one​

Delete a skill from the page, or delete its file. Nothing else is affected; skills do not depend on each other.

Next​