← Back to Dashboard

Research: Anthropic Knowledge Work Plugins

Source: GitHub Date: Feb 9, 2026 Status: Initial Analysis

        Executive Summary: Anthropic has open-sourced a suite of "Knowledge Work Plugins" designed for Claude Cowork and Claude Code. These plugins bundle Skills (domain expertise), Connectors (MCP tools), and Commands (slash commands) to specialize Claude for specific roles like Sales, Engineering, and Product Management.
    

Core Concepts

The repository (anthropics/knowledge-work-plugins) introduces a structured way to enhance AI agents:

Role-Based Specialization: Plugins are pre-packaged for specific job functions (e.g., sales, legal, data).
No-Code Structure: Everything is file-based (Markdown & JSON). No complex build steps.
Components:
- skills/: Domain knowledge and workflows (Markdown).
- commands/: Explicit slash commands (e.g., /sales:call-prep).
- .mcp.json: Connections to external tools via Model Context Protocol.

Available Plugins

Anthropic released 11 initial plugins:

Productivity: Task management (Asana, Jira, Notion).
Sales: Prospect research, CRM updates (Salesforce, HubSpot).
Engineering / Data: SQL queries, code reviews, infrastructure checks.
Legal & Finance: Contract review, reconciliation.
Enterprise Search: Unified search across Slack, Notion, Drive.

Relevance to Optimus Command Centre (OCC)

This architecture strongly aligns with our own Skill-based approach in OpenClaw. We can:

Adopt the Patterns: Structure our own skills (like mahjong-ai) using similar separation of "Skills" (knowledge) vs "Connectors" (tools).
Port Plugins: Adapt the productivity or data plugins to work within our environment, giving us instant capabilities for managing tasks or analyzing data.
Custom Plugins: Create a "Master Optimus Personal" plugin that defines your specific preferences and workflows in a standardized format.

Next Steps

[ ] Deep Dive: Analyze the productivity plugin code to see if we can adapt its logic for our Todo system.
[ ] Experiment: Try to implement a simple "Slash Command" style interaction for OCC based on their patterns.