Anthropic Knowledge Work Plugins: Self-Hosted AI Agent Skills for Every Role

Anthropic has open-sourced **11 Knowledge Work Plugins** for Claude that turn the AI agent into a specialist for specific business roles. With **18,300+ GitHub stars** on the knowledge-work-plugins repository and another **28,900 stars** on the official claude-plugins directory, this is one of the fastest-growing trends in the AI agent ecosystem.

These plugins bundle skills, connectors, slash commands, and sub-agents for job functions including **Sales, Marketing, Customer Support, Product Management, Legal, Finance, Data, Enterprise Search, Bio-Research, and Productivity**. They work with both **Claude Cowork** (Anthropic's team collaboration product) and **Claude Code** (the terminal-based AI coding agent).

For the self-hosted AI community, the significance is clear: these plugins demonstrate a structured, open-source pattern for building domain-specific AI agents that any team can adapt.

What Are Knowledge Work Plugins?

Each plugin is a directory of Markdown and JSON files that defines:

• **Skills** — domain expertise and step-by-step workflows that Claude draws on automatically when relevant
• **Slash commands** — explicit actions you trigger, such as `/sales:call-prep` or `/data:write-query`
• **MCP connectors** — Model Context Protocol server configurations that wire Claude to your real tools (Slack, Notion, Jira, Salesforce, and more)

The key insight: every component is **file-based** — Markdown and JSON, no code, no infrastructure, no build steps. This makes them trivially customisable.

The 11 Plugins at a Glance

| Plugin | What It Does | Key Connectors |

|--------|-------------|----------------|

| **productivity** | Manage tasks, calendars, daily workflows, personal context | Slack, Notion, Asana, Linear, Jira, Microsoft 365 |

| **sales** | Research prospects, prep calls, draft outreach, battlecards | Slack, HubSpot, Close, Clay, ZoomInfo, Notion |

| **customer-support** | Triage tickets, draft responses, build knowledge base | Slack, Intercom, HubSpot, Guru, Jira |

| **product-management** | Write specs, plan roadmaps, synthesise research | Slack, Linear, Asana, Jira, Notion, Figma, Amplitude |

| **marketing** | Draft content, plan campaigns, enforce brand voice | Slack, Canva, Figma, HubSpot, Amplitude, Ahrefs |

| **legal** | Review contracts, triage NDAs, navigate compliance | Slack, Box, Egnyte, Jira, Microsoft 365 |

| **finance** | Prep journal entries, reconcile accounts, financial statements | Snowflake, Databricks, BigQuery, Slack |

| **data** | Query, visualise, interpret datasets | Snowflake, Databricks, BigQuery, Hex, Amplitude |

| **enterprise-search** | Find anything across email, chat, docs, wikis | Slack, Notion, Guru, Jira, Asana, Microsoft 365 |

| **bio-research** | Literature search, genomics, target prioritisation | PubMed, bioRxiv, ChEMBL, Benchling |

| **cowork-plugin-management** | Create or customise plugins for your own tools | — |

Installation with Claude Code

To install these plugins with Claude Code (which works with multiple LLM providers):

```bash

Add the Knowledge Work marketplace

claude plugin marketplace add anthropics/knowledge-work-plugins

Install a specific plugin

claude plugin install sales@knowledge-work-plugins

claude plugin install data@knowledge-work-plugins

```

Once installed, skills activate automatically when relevant. Slash commands become available in your session.

For the broader plugin ecosystem, you can also install from the **official claude-plugins-official** directory (28.9K stars):

```bash

claude plugin marketplace add anthropics/claude-plugins-official

claude plugin install example-plugin@claude-plugins-official

```

Running Plugins with Self-Hosted Models

While these plugins are designed for Claude specifically, the **file-based structure** means you can adapt the same patterns for self-hosted AI agents. The skills are Markdown files containing structured instructions — any capable local LLM can use them as context.

For Hermes Agent users, the hermes-agent skill system supports a similar approach: skills are Markdown files that load into the agent's system prompt. See the Hermes Agent documentation for skills integration.

To build a local RAG system that provides similar context-awareness, pair Open WebUI with a vector database and your local LLM.

Customising Plugins for Your Team

The real power of these plugins is customisation. Since everything is file-based:

1. Swap Connectors

Edit `.mcp.json` to point at your specific tool stack:

```json

{

"mcpServers": {

"notion": {

"command": "npx",

"args": ["@notionhq/notion-mcp-server"],

"env": {

"NOTION_API_KEY": "$NOTION_TOKEN"

}

}

}

}

```

2. Add Company Context

Drop your terminology, org structure, and processes into skill files so your AI agent understands your specific environment:

```markdown

Company Context

• **Product names**: Atlas (CRM), Horizon (analytics)
• **Common acronyms**: NPS (Net Promoter Score), ARR (Annual Recurring Revenue)
• **Approval process**: All customer-facing pricing needs VP approval

```

3. Adjust Workflows

Modify skill instructions to match how your team actually works:

```markdown

Bug Triage Process

1. Check severity in the Slack #bugs channel

2. For P0/P1: notify on-call engineer immediately

3. Create Jira ticket with reproduction steps from the original report

4. Link the Slack thread in the ticket

```

4. Create New Plugins

Use the `cowork-plugin-management` plugin or follow the directory structure to build plugins for roles not yet covered.

Building Your Own Agent Skills

If you are not using Claude, you can still benefit from the same design patterns. The plugin structure is essentially a **prompt engineering framework** — and the lessons apply to any AI agent:

• **Separate domain knowledge from conversation** — keep reusable instructions in modular skill files
• **Use structured commands** — define explicit triggers for common workflows
• **Connect real tools** — MCP servers provide a standard interface for tool connectivity
• **Start generic, then customise** — Anthropic's plugins are designed as starting points

For implementing custom AI agent skills with local infrastructure, see our Build Your Own AI Assistant with n8n guide.

The Bigger Picture: Why Plugin Ecosystems Matter

The rapid adoption of these plugins (47,000+ combined GitHub stars in weeks) signals a shift in how AI agents are built and deployed. Instead of monolithic assistants that try to do everything, the trend is toward:

• **Role-specific agents** — specialised knowledge and tools for each job function
• **Modular skill stacks** — composable instruction files that can be mixed and matched
• **Open-source skill marketplaces** — community-driven repositories of reusable agent abilities

This aligns with the self-hosted AI philosophy: keeping control over your tools, data, and workflows while benefiting from shared, open-source patterns.

Conclusion

Anthropic's Knowledge Work Plugins represent a mature, open-source approach to building domain-specific AI agents. The 11 role-based plugins, combined with the broader claude-plugins-official directory, create a rich ecosystem of reusable agent skills. While designed for Claude, the patterns — modular skill files, MCP connectors, slash commands — are applicable to any self-hosted AI agent setup.

**Sources:**

• Anthropic Knowledge Work Plugins GitHub
• Anthropic Claude Plugins Official GitHub
• Claude Code Documentation