Plugin Components Reference

A plugin bundles components. On the plugin edit page, each component type has its own collapsible section where you add existing items or create new ones inline. This page documents every component type and its fields.

Component	What it provides
Skills	Markdown knowledge files with optional scripts, references, and assets.
Toolkits	Curated selections of tools from your MCP servers, plus prompts and resources.
Commands	Slash-commands (for example, /deploy) with typed arguments.
Rules	Always-apply or glob-scoped rules that guide agent behavior.
Hooks	Event-driven automations triggered on agent lifecycle events.
Sub-Agents	Dedicated agents with their own model, scope, and tool permissions.
MCP Servers	Existing integrations from your organization's MCP server list.
Custom MCP Servers	Raw MCP server JSON included directly in the plugin's .mcp.json.

Skills, Toolkits, and MCP Servers are added by searching your existing organization entities. Skills can also be created inline using the same fields as Create and Publish a Skill. The forms below cover the components you create directly on the plugin.

Commands

Select + in the Commands section to create a slash-command.

Field	Description
Name	Display name of the command.
Slug	The command invoked by the agent (for example, deploy).
Description	What the command does.
Content	The command body.

Rules

Rules guide agent behavior, either always or when editing files that match a glob.

Field	Description
Name	Display name of the rule.
Slug	URL-safe identifier.
Description	What the rule does.
Content (Markdown)	The rule instructions.
Always Apply	When on, the rule applies to every interaction. When off, it applies only to files matching Globs.
Globs (comma-separated)	File patterns that scope the rule, for example *.ts, *.tsx, src/**/*.md.

Hooks

Hooks run a command in response to an agent lifecycle event.

Field	Description
Name	Display name of the hook.
Slug	URL-safe identifier.
Description	What the hook does.
Event	The lifecycle event that triggers the hook (see below).
Type	The hook type, such as Command.
Matcher	Optional filter that narrows when the hook fires (for example, a specific tool).
Command	The command to run when the hook fires.

The Event dropdown includes:

sessionStart, sessionEnd, beforeShellExecution, afterShellExecution, beforeMCPExecution, afterMCPExecution, afterFileEdit, beforeSubmitPrompt, preCompact, stop, beforeTabFileRead, afterTabFileEdit, PreToolUse, PostToolUse.

Sub-Agents

Sub-agents are dedicated agents with their own model, scope, and tool permissions.

Field	Description
Name	Display name of the sub-agent.
Slug	URL-safe identifier.
Description	What the sub-agent is for.
Instructions	The sub-agent's system prompt.
Scope	Global or Local.
Model	The model the sub-agent runs on (for example, claude-4-opus).
Effort	Reasoning effort level.
Memory	Memory mode for the sub-agent.
Max Turns	Cap on the number of turns the sub-agent runs.
Tools (comma-separated)	Tools the sub-agent is allowed to use (for example, Read, Write, Shell).
Disallowed Tools (comma-separated)	Tools the sub-agent is blocked from using.
Background	When on, the sub-agent runs in the background.
Worktree Isolation	When on, the sub-agent runs in an isolated git worktree.

Custom MCP Servers

Use the Custom MCP Servers section to paste raw MCP server JSON (stdio, docker, and similar) that is included directly in the plugin's .mcp.json. Use this for servers that are not in your organization's MCP server list.

What to do next

• Create a Plugin: create one and add components
• Distribute a Plugin: publish, download, and sync