Documentation

What onplate is, how to install it on macOS, Linux and Windows, and what you see on the first launch.

From an empty storage to your first skill installed into your AI tool — in five minutes.

The overview screen - storage state, security, how much context skills take and usage statistics.

The main section - all your skills and packages. Adding, installing into AI tools, tags, favorites, trash and the quality profile.

Sets of skills, commands and MCP servers - import from a folder or a Git repository, install in one button, update and roll back.

Connecting MCP servers - by hand, from JSON or from a registry. Secrets are encrypted, the connection is tested before install.

Subagents for your AI - a narrow job, their own tool set and permissions. Creating, importing, security checks and installing into AI tools.

Editing skills with a working copy, publishing versions, changelogs, comparing versions and rolling back.

Searching for skills and MCP servers on external platforms, a preview with a security check, and import into your storage.

Connected Git sources with skills, packages and subagents - update tracking and re-import.

The skill scanner - prompt injection, dangerous commands, hidden unicode. Scores, working with findings and trusted domains.

Finding contradictions between skill instructions and collisions between MCP servers - with the reason and advice on what to do.

The install map - which skill, MCP server or subagent, in which AI, which version and at which path.

Scanning the system - finding skills installed straight into an AI and taking them under onplate management.

The history of every operation - installs, updates, rollbacks and removals. What, when and where.

The Team section - three ways to hand your setup to other people: a team profile from a Git registry, a project profile in the code repository, and a set as a single file.

Appearance, language, window, behavior and the skill install paths for every AI.

The local helper for the onplate interface - explains the screen, the consequences of actions, errors and confirmed results.

Trying a skill for real - onplate hands it to your AI on an ordinary task and reads the session log to show whether the model reached for it.

The .onplate/profile.json file in a code repository - which skills, MCP servers and subagents this project needs. Discussed in a pull request, applied on a button.