Skill storage

Storage is your library. Every skill and package you added to onplate lives here. The distinction that matters:

• In storage — the skill is saved in onplate together with its version history, but no AI knows about it yet.
• Installed — a copy of the skill has been written into a particular AI tool's folder, and that tool started using it.

One skill from storage can be installed into several AI tools at once — that is the whole point.

Search, filters and order

At the top is the search box: it looks through skills, packages, author and tags. Next to it:

• Type — All / Skills / Packages;
• AI — only what is installed into the chosen tool;
• Source — all sources or local only;
• Order — By name, Newest first, Dangerous first;
• the star — favourites only.

Favourites always come first, whatever order the rest of the list is in. If the filters leave nothing, onplate offers to reset search and filters — an empty screen must not be a dead end.

Below the filters is the scanner row with counts: how many entries are dangerous, suspicious, safe, and how many exceptions need rechecking. Every count is clickable and works as a filter.

Add

The Add button opens the “Add to storage” window. There are no tabs any more — instead there are five entrances, and all of them lead into one review:

• Choose folder… and Choose file… — a folder, a file or an archive on disk;
• Paste text or a link — clipboard content, a link to a raw SKILL.md or to GitHub, or a path to a file;
• From a repository — a whole Git source (see below);
• Write a skill from scratch.

What goes where is decided by the content

onplate reads what you picked, shows what it found, and only then offers the Add (N) button. The rule is simple:

A folder with a manifest goes to “Packages” as a whole, together with its commands and hooks. A standalone SKILL.md becomes a skill — you can edit it and publish versions of it.

The review shows where each item came from: an onplate manifest, a Claude plugin, a Codex plugin, a Cursor plugin, a SKILL.md without a manifest, a standalone file or an archive. One folder can carry several manifests at once — the official plugins do exactly that; onplate then reads the Claude one and treats the rest as copies of it. Anything that cannot be added is listed separately with the reason: the name is already taken in storage, the name is taken by an entry in the trash, a justification is required.

If you picked the wrong folder level, onplate says so plainly — there is no manifest and no SKILL.md here — and offers to choose another folder.

Write a skill from scratch

The “Write a skill” form: name (a slug, in Latin — for example report-helper), version, title, author, description, target AI tools and the instructions themselves. Above the form is a note that the entry will appear under “Skills”. From there you can develop the skill in the Editor.

Your own text goes through the scanner too. onplate does not ask you to justify your own writing, but the skill still gets a verdict — and it will hit quarantine on install if it contains dangerous constructs.

From a repository

A repository link, a branch, an access token for private ones, and an update mode. The token is stored encrypted. The default is notify only: dangerous updates always require confirmation. Connected sources live in Repositories.

Justification for dangerous content

Everything you add goes through the security check. If dangerous constructs are found, onplate highlights them and requires a justification — why you trust this skill. Without it a dangerous entry is not added, and that holds for every entrance, including a bulk import from a folder: dangerous items are added one at a time, after you have read the findings.

The skill panel

Clicking a row opens a panel to the right of the list. It holds:

• the security verdict — a coloured dot and, if threats were dismissed by hand, a “Reviewed by user · N” mark linking to what was dismissed;
• description, author, compatibility, permissions;
• the content — the full text of the skill without leaving the list;
• tags — free-form labels, added right here;
• where it is installed — the list of installs with AI tool, scope and version, marked outdated when storage has something newer;
• the source and the Verify source button — it compares your copy with the original and says either “matches the source” or “the source changed”.

The panel header holds the star (favourite), the quality Profile, Try it for real (a run), the other-actions menu and the close button.

If a skill references files that are not in its bundle, onplate says so right in the panel and offers to find the files in a folder — the usual cause is that a single SKILL.md was imported instead of the whole skill folder.

Installing into an AI tool

In the panel pick the scope and the AI tool:

• Global — works in every project of that tool;
• Project — only in the folder you choose.

A skill is installed only into the AI tools it declares. A dangerous skill needs a separate confirmation. Every install is visible in Bindings, and project-scoped ones also on the project page.

Bulk actions

The Select all checkbox and the per-row checkboxes turn on the bulk action bar: Install, Uninstall, Move to trash.

The target and scope you pick apply to everything at once. Whatever will not go is listed with a reason rather than silently skipped:

Reason	What it means
Quarantined	a dangerous skill is installed one at a time, with confirmation
Foreign files in the folder	the destination folder holds files that are not in storage
No compatible AI	the skill declares none of the chosen targets
Not installed anywhere	there was nothing to remove

A bulk install does not bypass the security gate — that is a deliberate decision, not an omission.

Quality profile

The Profile button in the panel header expands a rating across six dimensions: security, structure, documentation, maintainability, trust, portability — with an overall score out of 100 and an explanation of each scale. Next to it: the skill's language, its type (Backend, Testing, Documentation and so on), the confidence of the rating and the analyser version.

It helps you tell a careful skill from one thrown together in five minutes.

Tags and favourites

• Tags — free-form labels (“work”, “writing”, “python”), added in the skill panel. Both search and filtering work on them.
• Favourites — a star on the ones you need most; a separate filter shows only those, and they always come first in the list.

Tags and favourites are stored only on your machine.

Deleting and the trash

Two ways to delete, so nothing is lost by accident:

• Delete from storage — the entry goes to the trash. Versions are kept and the skill can be restored. Files already installed into AI tools stay on disk and keep working — onplate warns you that they will be orphaned.
• Delete from storage and disk — every install in every AI tool is removed, and the entry goes to the trash.

Right after a deletion an Undo appears at the bottom, in case you misclicked.

Trash is its own section in the sidebar, right under storage. Deleted skills, packages, subagents and individual versions land there; a counter shows how many. Inside you can see when an entry was deleted and what you can do with it: Restore, Delete permanently or Empty the trash. Permanent deletion cannot be undone — onplate will warn you.

One thing to remember: restoring brings back the entry and its versions, but it does not recreate the installs that were removed from disk. After restoring, install the skill into the AI tools you need again.