Claude Open-Sources 11 Office Plugins; Searching Across Slack, Notion, and Jira Is the Focus
Anthropic has open-sourced knowledge-work-plugins, 11 plugins covering enterprise search, customer support, data analysis, and more. About 35.8K stars on GitHub. The discussion focuses on whether cross-software retrieval can bring back the original sources.
Anthropic has open-sourced knowledge-work-plugins, with about 35.8K stars on GitHub. It targets Claude Cowork and is also compatible with Claude Code. Simply put, this is a set of plugins that package workflows, tool connections, and slash commands together—not scattered prompts.
The everyday pain point is very concrete: you talked about it in Slack, wrote about it in Notion, changed it in Jira, and two weeks later you only remember that it happened but can't find the exact words. The enterprise search plugin is aimed squarely at this: finding materials across chats, documents, and tickets. The customer support plugin classifies tickets, drafts replies, and compiles escalation materials; the data analysis plugin writes SQL, creates charts, and verifies results.
This open-source release includes 11 plugins, covering productivity, sales, customer-support, product-management, marketing, legal, finance, data, enterprise-search, bio-research, and cowork-plugin-management. Connectors include Slack, Notion, Jira, Asana, Microsoft 365, Snowflake, Databricks, BigQuery, and more.
Installation is not complicated. For Cowork, install from claude.com/plugins. For Claude Code, first add the marketplace, then install the specific plugin:
claude plugin marketplace add anthropics/knowledge-work-plugins
claude plugin install sales@knowledge-work-plugins
After installation, the plugin activates automatically. Skills trigger in relevant scenarios, and slash commands can be used directly, such as /sales:call-prep and /data:write-query. Each plugin has the same structure: plugin.json is the manifest, .mcp.json connects tools, commands holds explicit commands, and skills holds domain knowledge and steps. Everything is markdown and JSON files, with no build step. Changing connectors, writing company terminology and processes, and adjusting workflows can all be done by editing files directly.
One judgment in the discussion is more worth noting. Some users say that rather than automatic organization, they want cross-software retrieval that can directly locate the original chat, document, and specific paragraph. No matter how good the AI summary is, if you can't find the source, you don't dare fully trust it. This concern is not unfounded. Another concern is that permissions have to be configured one by one, and fragmented group chat messages can easily blow up the context. The proposed approach in the reply is to first retrieve, deduplicate, and aggregate by topic and time, then load the original text on demand while preserving message links and timestamps.
Someone else clarified the relationship between retrieval and summarization: retrieval is the foundation, and weekly meeting materials are downstream. Link Jira task status with decisions and reasons for changes in Slack, while also preserving the original sources. If a key decision is missed, then no matter how complete the weekly report is, the conclusion may still be wrong. Sources are more important than summaries; key conclusions must be traceable to the original messages, document locations, and timestamps, so they can be verified with one click. Otherwise, once the AI misunderstands something, the error will be repeatedly cited in subsequent weekly reports.
Open source solves the starting point. Permissions, deduplication, context budgets, and source traceability are the costs you pay every day. If you could automate only one thing first, would it be finding materials across tools or putting together weekly progress updates?
发布时间: 2026-11-15 14:27