Plugins & Marketplace

Plugins are extension packages for Kodik. They can add skills, slash commands, agent rules, lifecycle hooks, sub-agent profiles, and MCP servers. You discover and install plugins through the built-in Marketplace.

Opening the Marketplace

Open the Marketplace with the kodik.openMarketplace command from the Command Palette (Ctrl+Shift+P → “Kodik: Open Marketplace”). From there you can search for plugins, read descriptions, and install with a single click.

What plugins can contribute

A plugin can include any combination of the following:

Component	Description	Learn more
Skills	Specialized workflows and competencies for the agent	Skills
Slash commands	Custom commands invoked with `/` in the chat	Slash commands
Rules	Rules files that govern agent behavior in a project	Rules
Hooks	Shell commands tied to agent lifecycle events	Hooks
Sub-agent profiles	Definitions for specialized sub-agents	Sub-agents
MCP servers	Additional tools and resources via the MCP protocol	MCP

Namespacing

All components contributed by plugins are automatically namespaced by plugin ID:

A skill named my-workflow from plugin acme appears as acme:my-workflow.
A command named deploy from plugin acme is invoked as /acme:deploy (the bare /deploy also works).

This allows multiple plugins to contribute components with the same short name without conflict.

Enabling and disabling plugin items

Enablement is per item, not per plugin: every AI-facing component an installed plugin ships has its own toggle in Kodik’s settings panel, alongside the same kind of component from your workspace or user directories.

MCP servers — toggle in Settings → MCP Servers. The disabled flag is stored in the plugin’s own .mcp.json, and a disabled server is dropped from the agent’s tool discovery.
Skills — the auto-discovery toggle writes disable-model-invocation into the skill’s SKILL.md. The skill leaves the model’s skills index but stays invocable from the / menu.
Rules — toggle in Settings → Rules. A disabled rule is left out of the system prompt.
Sub-agents — toggle in Settings → Sub Agents. A disabled agent stays installed (marked “disabled”) but is not offered to the model.
Hooks — toggle in Settings → Hooks. The enabled flag is stored in the hook file itself.

Commands have no toggle — they only run when you invoke them.

Deleting a plugin-provided MCP server in Settings removes only that server entry from the plugin; the plugin’s Marketplace page can re-add it. Uninstalling a plugin removes its installed files and stored configuration; sensitive values, such as secret MCP user-config fields saved in secure storage, are also deleted.

Plugin layout on disk

Installed plugins live in Kodik’s global plugins directory. Each plugin occupies its own subdirectory with a .kodik-plugin/plugin.json manifest. Components follow a standard subdirectory layout:

<plugin-id>/
  .kodik-plugin/
    plugin.json          # plugin manifest
  .mcp.json              # MCP server configurations
  skills/<skill-dir>/    # skills (one directory each)
  commands/<cmd>.md      # slash commands
  rules/<rule>.md        # rules
  agents/<agent>.md      # sub-agent profiles
  hooks/hooks.json       # hooks

Skills — creating and using skills
Hooks — lifecycle hooks
MCP — MCP protocol and servers
Sub-agents — specialized sub-agents