Plugin Architecture

Good news, everyone! We are about to talk about plugins, and I need you to resist the urge to skip this lesson because you think you already understand what a plugin is. You do not. Not in this context.

Plugins in Co-Work are care packages. Not extensions. Not add-ons. Care packages. That distinction matters because it changes what you expect from them. An extension adds a feature. A care package gives you everything you need to operate in a domain you were not operating in before.

Each plugin contains three components. First: skills — domain knowledge that Claude draws on just-in-time when the context calls for it. You do not invoke skills manually. Claude reads your prompt, recognizes trigger words, and loads the relevant skill automatically. Second: commands — slash commands that invoke specific functions explicitly. Unlike skills, you call commands on purpose. /review-contract, /create-waterfall, /analyze-dataset. Third: connectors — MCP (Model Context Protocol) integrations that give Claude access to external tools and data sources. A connector might link to your Salesforce instance, your Jira board, your company's knowledge base.

Think of it this way. Without a plugin, you start at zero. You write a prompt from scratch, hoping Claude knows enough about your domain to produce useful output. Sometimes it does. Sometimes it produces something that looks right but misses the nuance that a domain expert would catch immediately.

With a plugin, you start at 80%. The skills encode the domain knowledge. The commands encode the workflows. The connectors provide the data access. You customize the last 20% for your specific needs — your company's terminology, your industry's regulations, your team's preferences. That is a fundamentally different starting point.

Anthropic ships plugins aligned to the standard business functions, and they are worth enumerating because the coverage is broader than most people realize.

Productivity — task management, meeting notes, project coordination. Sales — pipeline management, lead scoring, outreach drafting. Customer Support — ticket triage, response templating, escalation workflows. Product Management — PRD generation, competitive analysis, feature prioritization. Marketing — campaign planning, content calendars, performance analysis. Legal — and this one stood out to me — contract review that highlights areas that are acceptable, areas with potential risk, and core risks requiring human review. Not just "here is a summary." A structured risk assessment. Finance & Accounting — waterfall charts from CSV statements, variance analysis, budget modeling. Data Analysis — query data warehouses, visualize results, generate insights. Enterprise Search — unified search across company knowledge bases and document stores. Bio-Research — literature review, experimental design, data interpretation.

Ten domains. Each one a care package. Each one encoding the workflows that domain experts have refined over years. That is not a plugin marketplace. That is an institutional knowledge library.