Claude Code Plugins vs Commands: Choosing Your Automation Layer
A slash command is one markdown file under .claude/commands/ that only exists in the project you wrote it in. A plugin is a packaged, versioned bundle — commands, skills, agents and hooks together — installed with one command and reusable across every repo you touch or hand to a teammate. The question isn't which is more powerful; a plugin's commands behave identically to standalone ones. It's whether this workflow needs to leave the repo it was born in. Last updated 2026-10-30 by Rakshit Yadav.
What a plugin actually bundles
A Claude Code plugin is a directory with a manifest that a marketplace can resolve and install in one command. Inside it, you'll typically find some mix of commands/*.md (the slash commands it adds), skills/<name>/SKILL.md (playbooks that load on relevance), agents/*.md (subagents with their own tool grants), and hooks/* (scripts that fire on lifecycle events like a pre-commit check). None of these file types change shape when they move into a plugin — a command inside a plugin is still just a markdown file read on /name. What a plugin adds is purely distributional: a version number, a single install step, and the ability to update all of it at once instead of re-copying files into every repo that needs them.
A lone command has none of that machinery. It's one file, living in one project's .claude/commands/ folder, with no manifest and no install step — you write it, commit it, and it exists only there until someone copies it somewhere else by hand.
Decision checklist: personal/project-only vs shareable
| Question | If yes | If no |
|---|---|---|
| Will more than one repo need this exact workflow? | Lean plugin — one install updates every repo at once | A lone command is enough; packaging adds overhead with no payoff yet |
| Does a teammate need the same setup without copy-pasting files? | Plugin — share a marketplace entry or git URL instead of a zip in Slack | Keep it a command committed to the repo; everyone on the repo already has it |
| Does the workflow need more than one primitive — a command that also needs a skill, a subagent, or a hook? | Plugin — bundling means those pieces install and version together | A single command file is simplest if it's genuinely just a prompt |
| Do you expect to update this workflow often and want every consumer to get the update automatically? | Plugin — a version bump propagates on next install/update | A command only needs to change in the one place it lives anyway |
| Is this workflow specific to one project's conventions (its exact file layout, its exact stack)? | Keep it a project command — a plugin implies reusability this doesn't have | Plugin is fine either way |
Most workflows start on the left side of "no" — a command living in one repo — and only cross into plugin territory once a second repo or a second person needs the same thing. Packaging before that point is premature; the overhead of a manifest and a marketplace entry buys nothing until there's more than one consumer.
Migrating a /ship command into a plugin
- Confirm there's a real second consumer. If
/shiponly ever runs in one repo, there's nothing to migrate yet — keep reading Plugins vs Skills vs MCP for when a plugin earns its complexity. - Create the plugin's directory structure. A manifest file at the root, then subfolders mirroring what the command already relies on:
commands/ship.md(the file, moved as-is), andagents/orskills/for anything/shipdelegates to — see Agents vs Commands for how that delegation works inside a command body. - Pull in what the command already depends on. If
/ship's instructions tell the session to delegate to atech-leadsubagent or a release-checklist skill, those files move into the plugin too — a plugin that only ships the command but not what it calls isn't actually portable. - Write the manifest and version it. Give the bundle a name and a starting version so future changes are something consumers can pull as an update rather than a silent overwrite.
- Test a clean install in a second repo. Install the plugin into a repo that's never seen
/shipbefore and run it end to end — this is where missing dependencies (a hook that assumed a file existed, a skill reference that used a relative path) actually surface. - Scaffold the boilerplate instead of hand-rolling it. The claude-code-plugin-scaffold skill generates the manifest and folder layout from an existing command/skill/agent set, which is faster and less error-prone than copying another plugin's structure and editing it by hand.
Where AgenticKit fits
AgenticKit's 34 subagents already ship as the plugin shape described above — commands, skills, agents and hooks installed together as one unit, versioned, so a new repo gets the whole set in one command instead of someone hand-assembling it from the individual pieces. If you're deciding whether your own /ship-style command is worth packaging, AgenticKit's roster is a working example of the payoff: the same bundle installs identically whether it's your fifth repo or your first.
Related pages
Frequently asked questions
Does putting a command inside a plugin change how it runs?
No. A command inside a plugin is read and triggered the same way as a standalone one — type /name and the current session reads that file as a prompt. The plugin only changes how the file got into the repo and how it's updated later.Can I have a plugin with just one command and nothing else?
Yes, though at that point you're mostly paying the packaging overhead for distribution, not for bundling several primitives together. It's worth it once a second repo or teammate needs that one command; otherwise a committed project command file is simpler.What's the actual trigger for turning a command into a plugin?
A second real consumer — another repo, another teammate, or a workflow that now needs a skill/agent/hook alongside the command. Packaging before that point adds a manifest and marketplace entry with no one yet needing either.If I update the plugin, do all the repos that installed it get the update automatically?
They get it on their next install/update step, not automatically on every session — a version bump is something each consuming repo pulls, the same way any package manager update works.
Skip packaging it yourself
AgenticKit installs 34 agents with the commands, skills and hooks they call as one bundle — the plugin shape, pre-built.