Team

One section answers one question: how your setup reaches other people. There are three ways, and they differ not in convenience but in who is on the other end and through which channel the thing travels. Everything works through your own Git or plain files: there is no central onplate server.

Which way to choose

Way	Who is on the other end	Channel	When
Team profile	the whole company	a Git repository with a registry	everyone must have the same set, with a variation per role
Project profile	whoever works on that particular code	a file in the code repository itself	“this project needs these skills”
A set as a file	one particular colleague	a file over a messenger or email	hand over a selection quickly, with no repositories and no agreements

If you are not sure — start with the set: it requires nothing and commits you to nothing. Profiles are for when the agreement has to outlive a single conversation.

Way 1. The team profile

Solves “everyone on the team must have the same set”. One person publishes the reference into a shared Git repository, everyone else connects to it — and each person’s onplate keeps the set installed and current.

Connect a profile

In the Team registry field pick the Git source with the registry. If there are no sources yet — first connect the team repository in Repositories or import it as a package.
Pick your roles — “analyst” or “support”, for example: a profile can define different sets for different roles. A role inherits the base set and adds only its own.
Pick the sync mode:
- Off — the profile is connected but nothing is checked;
- Notify only — onplate shows the differences, the decision is yours;
- Install automatically — whatever is missing is installed on its own.

Differences from the profile

The table shows how your machine differs from the reference:

Status	What it means
Not installed	The package is in the profile, you do not have it
Outdated	An older version is installed
Current	Everything matches
Outside the profile	The package is installed on your machine but is not in the profile
Not in the registry	The profile points at a package the registry does not have

The Refresh button recalculates the report, Apply installs and updates what you selected, with a confirmation. For every package you see the outcome: installed, updated, skipped, needs manual work or an error.

The difference is counted per AI tool as well: a package of the right version that is not installed into every target AI is honestly shown as missing — not as “current”.

Publish a set for the team

The reverse operation — you assemble the reference:

Press Publish a set for the team.
Give the set a name and, if needed, a role.
Pick the format:
- Folder (repository) — onplate assembles the registry into a folder and you commit it yourself;
- Zip bundle — a single file, easy to send;
- Straight into Git — onplate commits and pushes itself: give it the repository URL, the branch, a token with write access and a commit message.
Dangerous packages are left out by default — the separate “Include dangerous” checkbox lets them in.

If nothing changed since the last publish, onplate says exactly that: “No changes — the registry is already current”.

Way 2. The project profile

An .onplate/profile.json file in the code repository that says “this code needs these skills, MCP servers and subagents”. onplate writes it, and people read it in the diff of a pull request — and that, not the install, is its main value: the agreement is discussed where everything else about the project is discussed.

The details are in Project profile.

The Share with the team button is right here, in the “Ways to share” block.

Way 3. A set as a file

The simplest way: gather a selection into one file and hand it to a colleague any way you like — a messenger, email, a flash drive. No repository, no agreement, no setup required.

A set is an ordinary onplate package. The recipient can open it selectively (“Open a set”) or import it whole, like any other package.

Build a set

Press Build a set in the “Ways to share” block.

The name of the set and who built it — the recipient sees the second one. It is the only place that will say where the file came from: over a messenger it arrives as just an archive.
Tick what you are sharing. The list holds the skills, MCP servers and subagents from your storage. Every row shows the version, the security score and the token weight.
At the bottom is the total: “the set adds N tokens to context”. That is the price the recipient pays on every request to their AI, not decoration.
Show the manifest — the very file that will be written. Open it at least once: trust is given by showing, not by promising.
Format: a single .zip file (the default — a set is usually handed to a person) or a folder.
Save the set — pick where to put it.

What will not go into the set

Secret values. Neither the ones marked as secrets nor a token you typed into an ordinary environment variable: onplate recognizes credentials by their shape (ghp_…, AKIA…, a private key block) and replaces them with a required secret. The recipient enters their own. You are told about every substitution before the file is written.
Dangerous things without a justification. A skill scored “Dangerous” travels only with your explanation of why — and the explanation travels with it.

We also warn you separately if the body of a skill still holds a path from your machine (/Users/you/…): your colleague will not have it. That is a warning, not a ban.

Open a set

Press Open a set and pick the file or folder you were sent. You do not have to unpack a .zip yourself.

First comes a preview, and nothing enters storage until you choose. You see who built the set and when, and for each entry what it means for you personally:

Mark	What it means	Ticked in advance
new to you	there is no such record in your storage	yes
you already have the same	the content matches yours	no
something else with this name	the name is taken, the content differs	no

The comparison goes by content, not by version number: somebody else’s “1.0.0” says nothing about whether the bytes match.

Only what is new is ticked in advance. What you already have, and anything dangerous, you include yourself — a checkbox should not decide for you.

If the name is taken

Your record is never overwritten. onplate offers to take theirs under a new name — go-review-from-alice, for example — and both stay with you. The name can be changed right there in the dialog.

Importing does not install

The records land in storage. Laying them out into the configs of your AI tools is a separate decision you make later, in the Skill storage. Agreeing to take somebody else’s file is not the same as agreeing to put it to work.

What you can see afterwards

A record that arrived in a set is marked on its card: Source: project-profile or the name of the set and its author. A month later the question “where did this come from” is asked of the wrong person — and the answer should live inside the record itself.