Skip to content

Add plugin support as a first-class resource type #133

@talvaknin744

Description

@talvaknin744

Summary

Skillshare already acts as a single source of truth for skills, agents, and extras across many AI CLIs. I’d like to propose adding plugin support as a first-class resource type, so users can import, normalize, and sync plugin content between tools such as Codex and Claude Code, instead of manually re-creating the same workflows in each ecosystem.

Problem

Today, plugin ecosystems are becoming an important distribution mechanism for reusable AI workflows:

  • Codex plugins can bundle workflows, bundled skills, and MCP-related setup.
  • Claude Code plugins can include skills, agents, hooks, MCP servers, and other plugin-specific components.

Skillshare currently solves “one source, many targets” for skills/agents/extras very well, but plugins still sit outside that model. That creates a few problems:

  1. A useful plugin installed in one tool is often not portable to another.
  2. Shared functionality (skills, agents, MCP config, commands) gets duplicated manually.
  3. Teams lose a canonical source of truth once they move from raw skills into richer plugin packaging.
  4. Migration between ecosystems becomes harder than it should be.

Proposal

Introduce plugins as a new resource kind in skillshare, alongside skills, agents, and extras.

High-level idea

Skillshare would provide a canonical intermediate model for plugin contents, then map that model to each target’s native format.

That means skillshare could:

  • Import plugin contents from supported ecosystems
  • Normalize overlapping concepts into a common internal representation
  • Sync/export them back out into target-specific layouts

Example use cases

  • Import a Claude Code plugin and sync the compatible parts into Codex
  • Import a Codex plugin and sync the compatible parts into Claude Code
  • Keep one repo as the source of truth and generate per-tool plugin layouts
  • Reuse bundled skills/agents/MCP config across multiple AI CLIs without maintaining separate copies

Suggested scope

Phase 1: Read/import plugin contents

Support importing plugin folders/manifests from at least:

  • Claude Code plugins
  • Codex plugins

Imported components could be mapped into a canonical model such as:

  • skills
  • agents
  • commands
  • MCP servers/config
  • metadata
  • optional tool-specific extensions

Phase 2: Sync/export to targets

Allow syncing compatible plugin content into target-native layouts.

Example behavior:

  • supported components are rendered normally
  • unsupported components are skipped with a clear warning/report
  • target-specific metadata is preserved where possible

Phase 3: Optional marketplace/index support

If skillshare continues growing as a hub/index-based system, plugin metadata could later be searchable/installable in a similar way to skills.

CLI ideas

Some possible commands:

skillshare sync plugins
skillshare import plugin /path/to/plugin
skillshare install plugin <source>
skillshare list plugins
skillshare check plugins
skillshare audit plugins

Metadata

Metadata

Assignees

Labels

featureNew feature or request

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions