Skip to main content

0. Plugins - LS YT Vid

Note: The video covers material not in the guide below — please watch in full.

Action Step

Complete this before moving on.

Build your first plugin this week. Open Claude Cowork, grab the Plugin Maker (or just tell Claude to look up how Claude builds plugins), and package a workflow your team already does by hand. Keep the first one minimal — the goal is to feel the difference between a plugin and a skill and to try wiring in a hook. If anything here is fuzzy, paste this video's transcript into Claude and ask it to teach you how hooks, connectors, and sub-agents work — and which sub-agents you could bake in to make the plugin more powerful.

Comment in Slack

Post your answer in your onboarding channel.

What was your biggest takeaway(s) from this training?

Training Guide

Every go-to-market team has the same problem: their best AI workflows live in a doc on one person's laptop, they don't run on their own, they're not wired to the tools the team uses, and nobody else ever gets access to them. Claude's plugins feature fixes exactly that — and this is the overview video that ties the whole section together.

Start with what a plugin actually is.

What a Plugin Actually Is​

The best way to think about a plugin is as a packaged workflow. A skill is the instructions for a task. A plugin is an entire workflow — it wraps the skill plus the other pieces that make it run: connectors, hooks, and sub-agents.

A skill is just a markdown file inside a folder, and yes, you can share it. But the difference isn't shareability — it's that a plugin wraps so much more inside the box that when Claude runs it, it has far more functionality. Three major pieces unlock that.

The first is the one a skill can never do on its own.

Hooks​

A hook is an automatic trigger that fires whenever you're chatting with Claude — before and after a message is sent. It's deterministic code running behind the scenes, so you can attach instructions or automations to specific events.

That determinism is the point. LLMs are probabilistic and tend to hallucinate; a hook is discrete code, so it's a way around that. A pipeline hygiene plugin might run a deal scan the moment it launches, so Claude has the data before it responds. A marketing campaign-brief plugin might run a brand-guidelines check before any post or email ships, so nothing off-brand goes out.

The second piece connects the workflow to your tools.

Connectors​

A connector wires the plugin to the tools a workflow needs — Salesforce, Gmail, analytics, whatever. You can ship connectors inside a plugin. So if a teammate downloads your pipeline hygiene plugin but doesn't have the CRM connected yet, the plugin forces them to OAuth into it with their own credentials.

That's why a plugin is more powerful than a skill: a skill only gives Claude the instructions, while a plugin ships all the wiring so nobody has to hunt for setup later. One security rule: never ship the keys or tokens — the user auths with their own credentials.

The third piece hands work to specialists.

Sub-Agents​

A sub-agent is a separate Claude session the main agent hands work to — think of a manager with a bench of specialists. Inside a plugin, you can bake in those sub-agent prompts so the main agent knows from the start which specialists it can deploy.

A pipeline hygiene plugin might include a deal-risk specialist that flags stuck deals and asks why; a campaign-brief plugin might include a brand-voice specialist that reviews every output.

So where does all this pay off?

Where the Payoff Is​

Across every go-to-market function:

  • Sales — post-call (grab the transcript, draft the Gmail follow-up, update the CRM after rep approval), pipeline hygiene, morning brief, pre-call prep.
  • Marketing — pull last week's numbers and draft on-brand angles, plus a brand guardian with a pre-publish checklist.
  • Customer Success — renewal radar, QBR prep, churn watch on the account level.
  • RevOps — CRM dedupe, forecast rollup, routing guardrails applied before records hit the CRM.

And the real unlock is what this means for the whole team.

Build Once, the Whole Team Runs It​

A plugin is one abstraction layer above skills. Skills were huge when they came out in October 2025, but because a plugin can house multiple skills plus all the connective tissue to run a workflow, one person can build something once and the entire team can use it.

The mental model: one person wraps the skill, adds the hook and connectors, then publishes it to a marketplace (we have a separate video on those — it's not a store, it's how you distribute plugins across the org). Everyone adds that marketplace, and from then on there's only one place to update — everyone gets the latest version. At the marketplace level you also get governance: make a plugin required, installed by default, or available to install.

Which is why this compounds.

Why This Compounds​

A plugin is an asset that compounds. Start small — every time you run it you'll spot an improvement, tell Claude to update it, and the next run is smarter. Every teammate who learns something new updates it too.

And these are open standards — just folders with markdown and JSON files. Whether Claude, OpenAI, Gemini, or Copilot leads the race tomorrow, you're not locked in. The edge goes to teams that move early, because every future update builds on the paradigm you understand now.

Wrap-Up​

To bring it home: a skill is one task Claude can do. A plugin is what closes the gap between a clever trick one person shares around the org and a capability the entire organization can run — packaged inside a box. That's a plugin. The next trainings in this section break down building one, the action steps, and how we do it for clients.