About Commands, Skills, Agents, and Plugins

In recent days I've been seeing too much content where someone made some plugin for Claude Code, some skills, agents and everything in that spirit. It seems not everyone understands what these words actually mean. So some hype and FOMO is being created out of nothing.

In a nutshell:

• A command is simply a prompt that can be sent with autocomplete.
• A skill is a command that the agent can invoke on its own, understanding from indirect signs.
• An agent is a separate process with a prompt sent to a new context window.
• A plugin is a Claude Code specific way to load all of the above + hooks + MCP in batches. Either from a local folder or from some repo.

Hot take: you almost never need any of this from others, you only need ideas. The only important thing here is prompts.

And if you see somewhere that someone published a Claude Code plugin with an agent for simplifying code, or with skills for working with Android, then in your head you should translate this roughly as: "I generated prompt text with Claude Code, possibly proofread it, and tweaked it so it does what I want."

Just generate your own the same way, but based on the current state of your project or your vision. It's not difficult. At best, use others' prompts as a reference. Otherwise it turns out that the agent for simplifying code assumes you're writing a CLI app in React, and the skill for writing Android features talks about some libraries that aren't in your project and never were. Garbage and contradictions in your context is the last thing you need.

The more specifics in a prompt, the less likely it needs to be shared. The less specifics in a prompt, the less likely it even needs to be written, since the model most likely already knows this.