Image: GitHub
Why it mattersA company can read exactly how Anthropic has told its own agent to do each role. The files are plain markdown and JSON.
Anthropic has published the plugin pack behind Claude Cowork as an open-source repository. The pack covers eleven roles, from sales to legal to bio research. Every plugin is a folder of plain markdown and JSON files.
The repository is anthropics/knowledge-work-plugins on GitHub, under Apache-2.0. It was created on 23 January 2026. The GitHub API reports 32,016 stars on 23 October 2026. The repository ranked 25th on Trendshift's daily list on the same day. The main branch was last pushed on 23 October 2026. Feature work landed on 21 October 2026 with version 2.1.0 of the sales plugin.
What is inside
The pack has eleven official plugins. Productivity covers tasks, calendars and workflows across Slack, Notion, Asana, Linear, Jira, Monday, ClickUp and Microsoft 365. Sales handles prospect research, call prep and pipeline review with HubSpot, Close, Clay and ZoomInfo. Customer Support triages tickets with Intercom, HubSpot and Guru. Product Management tracks specs and roadmaps across Linear, Asana, Figma, Amplitude and Pendo.
Four more plugins cover marketing, legal, finance and data. Marketing uses Canva, Figma, HubSpot and Ahrefs. Legal uses Box, Egnyte, Jira and Microsoft 365. Finance and Data both use Snowflake, Databricks and BigQuery. The pack also includes Enterprise Search and Bio Research for life science teams. A Cowork Plugin Management plugin is included for building new ones.
The file layout each plugin follows
Each plugin folder has the same four parts. One is .claude-plugin/plugin.json, the manifest. The second is .mcp.json, which lists the tool connections. The third is commands/, which holds the slash commands a user types to the agent. The fourth is skills/. This folder holds the domain knowledge the agent calls on its own.
The split matters. A command is called by the person. A skill is called by the agent when it reads the task. A plugin carries both so the same tool can be used two ways. The design is also readable. There is no code to build. A company can fork the repository and edit a markdown file. The plugin can then match the house tools and vocabulary.
What it unlocks for a reader
A team using Claude Code or Claude Cowork can read the files line by line. They can see what Anthropic has told its own agent to do for each role. The sales plugin, in version 2.1.0, now asks the agent to show the source of every line it writes. Each source carries a link to check it. The security guidance plugin was bumped three times in the last five days. The commit log makes this visible.
The pack is also a shortcut for building an internal plugin from scratch. The Cowork Plugin Management plugin exists to help with that. The layout is the same for every role. A company can start from the plugin closest to its own job. It can then swap the connectors. The whole repository is 32,016 stars at the time of writing. Trendshift ranked it 25th on the day. A reader whose team already pays for Claude Code can read these files at no extra cost.
Source
anthropics/knowledge-work-plugins on GitHub, stars and dates confirmed against the GitHub API.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
