kortecxdocs
Beta tasks

Write a skill

For a beta user: write a skill in the console, see your skills and what each can fire, read one, and remove it.

For a beta user who wants an App to follow the same written procedure each time. After this page you can write a skill, find it in your list, read it and remove it. Not here: attaching a skill to an App.

Skills is family 10, which reads four bars evidenced in the census at revision 3ec0e4f4, 2026-10-20. The reachable bar reads under the older census rule, not the newer one. The census console bar has no reading key. The cloud manifest does not claim that Skills holds four bars; the census finds it evidenced.

As of 2026-10-21, the cloud SDK and command line name two methods for this family: skills.list ("your skills") and skills.get ("one skill"), with the command-line verbs skills list and skills get. They name none for writing or removing a skill, so this page gives the console path. The cloud SDK is also built from source until its first publish.

What a skill is

A skill is instructions plus a list of tools it wishes for. A wish is not a grant: a wish for a tool that cannot be granted is dropped and the run goes on.

Open the tab

MCP in the left rail opens a page titled "Tools a run can be granted." Its Skills tab says "Your skills. Skills are personal: teammates see their own." A skill you write goes into your own catalog in the workspace you are in; one written in another workspace does not show in this one.

Write one

Press New skill. The dialog says "Name it, say what it does, choose the tools it wishes for, and write its instructions. The runtime checks it and answers."

It has a Name, which takes lowercase letters, digits and hyphens, starting with a letter, then What it does, Tools it wishes for and Instructions, which is Markdown. The tools you can pick are the ones the console has read from your tool list. A tool that cannot fire here carries its status word on its chip and can still be picked. Press Create.

In the form, an empty Instructions gets "Write the instructions a run follows." Edit as JSON swaps What it does and Tools it wishes for for a Manifest box, and Back to the form returns.

When the skill is stored the dialog is titled "Skill created" and says "Created" and the name. Using a name you already hold writes over that skill: the dialog is titled "Skill saved" and says "Saved" and the name, or "Already held; nothing new stored." when nothing changed.

Read one

Each skill in the list shows its name and version, its description, a Status word, a Can fire here count such as "1 of 2" or "no wishes", and whether its Instructions are inline or by reference. The Status word is the one the runtime serves, such as ready or blocked.

Details opens the skill's drawer. It shows "M of N can fire here", or "Wishes for no tool", and "blocked by" with the reasons when it is blocked. Each wished tool reads "can fire here" or "can't fire here". An Attach to an App row reads "Can attach here", or "Can't attach here" with the runtime's code beside it when it serves one. The drawer also shows Tags when the skill has some, Updated when the runtime serves a date, and Who can edit: "Your catalog only".

Instructions and How it was checked are sections you open, and each reads when first opened. How it was checked lists the checking steps, any advisories and the phase. If the deployment does not hold the instructions text, Instructions says "This deployment doesn't hold the instructions body. The skill names it by reference."

Remove one

In the drawer, Remove… asks you to type the skill's name to remove it from your catalog. The Remove button stays greyed until the typed text matches, and Cancel backs out. The drawer then says "Removed" and the name, or "Nothing to remove." The runtime answers the same way when the name is not in your catalog. The stored instructions text is not deleted with the skill.

Who can change them

The skills handlers read and write the signed-in person's own catalog in the current workspace. They check no organization role.

What it does not do yet

Dated gaps, as of 2026-10-21. None is a step on this page.

  • A skill from one sentence. The Skills tab has no control that proposes a skill from a sentence.
  • Sharing and attaching. Sharing a skill with a workspace, and attaching a skill to an App, are not steps on this page.
  • Failure, load and scenario checks. The census finds no instrument or receipt for this family for any of the three.
  • Driven live. Everything on this page is read from the console source and the cloud code at the cited revisions. None of it was driven against a deployed console for this page.

What it refuses

The census tenancy check for this family reads PASS on a receipt dated 2026-10-13: another tenant's read of an existing skill answers skill_not_found, the same as a random name's, beside the owner's 200, and its list lacks the skill. That receipt read only. It did not probe a write on another tenant's skill, or a member against an owner inside one org.

A manifest is checked and a refusal shows "Couldn't create this skill" with the runtime's code. In the Manifest box, a key containing warrant, grant, secret, credential or executable is refused outside the top-level tools map, as is a manifest whose name differs from the name you gave. A deployment without the skills catalog says "Skills aren't set up on this deployment", or "Skills aren't on this deployment" where the route is not mounted, and then shows no New skill button.