pinework skill add and pinework skill create. You manage both on the Customize page, under Plugins and Skills.
A plugin brings skills, MCP servers and commands
Pinework reads plugin manifests written for Claude Code, Codex, Cursor or the open agent-plugins format. A manifest can declare many kinds of parts. Pinework installs three of them:- Skills, which agents load into a run.
- MCP servers, which give agents new tools.
- Commands, which you can turn on or off.
A marketplace is a list of plugins
A marketplace is a GitHub repo that publishes an index of plugins. Two are built in: “Anthropic official plugins” and “Cursor official plugins”. You can add your own:Plugin parts are read-only because the plugin owns them
An installed plugin stays tied to its source. Once a day, Pinework checks the branch a plugin follows for a new commit. When the commit changed, it scans the plugin again and replaces its parts in place. You can also runpinework plugin refresh on one plugin.
A local edit would not survive that replacement. So Pinework blocks edits to parts a plugin owns. An edit attempt fails with this message: “This resource is owned by a plugin and is read-only.”
You can still change a few settings on plugin parts:
- Turn the whole plugin on or off for the workspace.
- Turn a plugin skill on or off by default.
- Turn a plugin MCP server on or off, and pick the credential it uses.
To change a plugin skill, make your own copy
Pinework has no fork button. To change what a plugin skill says, read its files, then create a standalone skill with your version. Then turn the plugin’s skill off, so agents do not load both.A skill reaches a run when it is on for the agent
Plugins and their skills apply to the whole workspace. Each agent can override one skill:Install a plugin
Add a plugin from a marketplace.
Add an MCP server
Give agents a tool outside any plugin.