agentleFS
selamy-labsselamy-labs/agent-skills/skills/plugin-authoring/SKILL.md

plugin-authoring

Skill0 starsChanged 3 days agoMITView on GitHub

Use when packaging skills, commands, agents, or hooks into a Claude Code plugin and publishing to a marketplace. Get the directory structure, plugin.json and marketplace.json manifests, version bump, and local-test loop right so the plugin installs cleanly and is version-pinnable.

6 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Files
SKILL.md

Plugin Authoring

A Claude Code plugin bundles skills (and optionally commands, agents, hooks, MCP servers) into one versioned, installable unit distributed through a marketplace. Getting the structure and manifests right is what makes it install cleanly and pin to a known version — the failure modes are all in the layout and the JSON.

Structure

my-plugin/
  .claude-plugin/
    plugin.json          # the plugin manifest
  skills/<name>/SKILL.md
  commands/<name>.md     # optional
  agents/<name>.md       # optional
  hooks/                 # optional

The marketplace is a separate repo (or root) with a marketplace.json listing the plugins it offers.

Manifests

plugin.json — identity + version (the thing consumers pin):

{
  "name": "my-plugin",
  "version": "0.2.0",
  "description": "One line on what it provides",
  "author": { "name": "...", "email": "..." }
}

marketplace.json — the catalog entry pointing at the plugin:

{ "name": "my-marketplace",
  "plugins": [{ "name": "my-plugin", "source": "./my-plugin" }] }

The loop that prevents broken releases

  1. Build the structure; keep one capability per skill.
  2. Validate locally — install the plugin from a local path and confirm the skills/commands actually load before publishing. A manifest that parses but ships a skill that won't load is the classic invisible break.
  3. Version deliberately — bump plugin.json version on every change; consumers pin to it, so an unversioned change silently shifts everyone.
  4. Publish by updating the marketplace entry, then have a consumer install the pinned version and verify it resolves.

Discipline

  • Version-pin in consumers, never float — reproducibility beats "latest."
  • One bundled unit, declared per workload — each workload's config names the plugin + version it gets, so what's installed is auditable.
  • Don't ship exec scripts inside skills — keep skills as instructions; put any tooling in the plugin's scripts/commands, audited separately.

Adapted from the MIT-licensed softaworks/agent-toolkit plugin-packaging skill (guidance only; upstream's generator script intentionally omitted).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.