Nanosaur2-670M
well9472 · Kit
Makes illustrations from text prompts with a small 670M-parameter diffusion model, run through ComfyUI. Built as a low-budget research model for people who wan…

Anthropic’s open-source plugins that give Claude a job: sales, legal, finance, data, marketing, support, HR and more. Each bundles skills and tool connectors for Claude Cowork or Claude Code. Not the Agent Skills repository.
Anthropic knowledge-work plugins is an open-source project on GitHub in the Business category, listed by Mindket. It is free. This is version 2, last updated on . Install it from GitHub as its README says. Get Mindket’s setup notes on your Mindket account, as a zip.
claude plugin marketplace add anthropics/knowledge-work-plugins claude plugin install sales@knowledge-work-plugins
Knowledge Work Plugins is Anthropic’s repository of plugins that set Claude up for one job function. A plugin bundles skills, the know-how Claude draws on, with slash commands and connectors. Connectors link Claude to the tools that job uses, such as a CRM, a chat app or a data warehouse. The README says they are built for Claude Cowork and also work in Claude Code. The repository is also a plugin marketplace named knowledge-work-plugins. Nearly all of it is Markdown and JSON, so you can read what a plugin tells Claude to do before you install it.
The README’s table names 11 plugins, but the marketplace file read on 2 November 2026 lists 123. Twenty-two keep their files in the repository: 17 built by Anthropic and five by partners (Slack, Apollo, Common Room, Tribe AI and Zoom). The other 101 point to their makers’ own repositories, each pinned to one commit. Anthropic Agent Skills (anthropics/skills) is a different repository. It holds single skills, such as the Word, Excel, PowerPoint and PDF tools and creative examples, packed into two plugins, and its README says they are for demonstration and education. Here each plugin is a bundle for one role, with connectors to your tools attached.
People who already work in Claude Cowork or Claude Code and want Claude to know their role’s routines, from call prep to month-end close. Sales, legal, finance, data, marketing, support, product, HR, operations, design and engineering each have a plugin, and so do small-business owners and life-science researchers. The README calls the plugins generic starting points that pay off once you add your own tools, terms and processes.
| Industry | Who uses it | How they use it | Example |
|---|---|---|---|
| Sales and revenue | Account executives and managers | Prep calls, research accounts, review the pipeline, draft outreach | An account executive asks for call prep and gets attendees, account history, open deal status and questions to ask on the call. |
| In-house legal | Commercial and product counsel | Check contracts and NDAs against a playbook kept in a local file | Counsel runs a vendor contract through the playbook and gets deviations flagged, with redlines. |
| Finance and accounting | Accountants and controllers | Prepare journal entries, reconcile accounts, explain variances, for a professional to review | An accountant reconciles a ledger account against a bank statement and gets the reconciling items sorted by category. |
| Small businesses | Owners without a finance team | Chase overdue invoices, stage payroll and bills for approval | An owner asks who owes money and gets reminder drafts that wait for a go-ahead. |
Skip it if you want one self-contained skill, since Anthropic Agent Skills fits that better. Skip it too on a free Claude plan: the Help Center says plugins are for paid plans.
The README names two places: Cowork, where plugins install from claude.com/plugins, and Claude Code, where you add the marketplace from the terminal. The Help Center adds that plugins also work in chat on the web and in Claude Desktop’s Chat tab, where skills and commands run but hooks and sub-agents do not. A plugin you add in Claude is saved to your account, so it shows up in Claude Code when you sign in with the same account.
| Plugin | The job it does | Some of its connectors |
|---|---|---|
| productivity | Tasks in a TASKS.md file, a memory of your people and shorthand, a local dashboard | Slack, Notion, Asana, Linear |
| sales | Call prep, account research, outreach drafts, pipeline and forecast | HubSpot, Salesforce, Gong and 17 others |
| customer-support | Ticket triage, reply drafts, escalations, knowledge base articles | Intercom, HubSpot, Guru |
| product-management | Specs, roadmaps, research synthesis, stakeholder updates | Linear, Figma, Amplitude, Pendo |
| marketing | Content, campaign plans, brand-voice review, SEO audits, reports | Canva, HubSpot, Ahrefs, Klaviyo |
| legal | Contract review, NDA triage, compliance checks, briefings | Box, Egnyte, DocuSign |
| finance | Journal entries, reconciliation, variance analysis, close, audit support | BigQuery, Slack |
| data | SQL, exploration, charts, dashboards, checks before you share | BigQuery, Hex, Amplitude, Definite |
| enterprise-search | One query across chat, email, documents and wikis | Slack, Notion, Guru, Atlassian |
| engineering, human-resources, design, operations | Standups and incidents, recruiting and reviews, design critiques, vendor reviews and runbooks | Slack, Notion, Atlassian, GitHub, Datadog, Figma |
| bio-research | Literature, single-cell QC, sequencing pipelines, target ranking | PubMed, bioRxiv, ChEMBL, Open Targets |
| small-business | Invoices, payroll, bills, leads, reviews | 35 servers, among them QuickBooks, Xero, Stripe, Gusto |
Anthropic’s other two plugins have no remote connector. The pdf-viewer plugin opens PDFs in a live viewer, where you annotate, fill and sign them through a local server. The last, cowork-plugin-management, builds a new plugin or changes an installed one.
Get it saves a zip of Mindket’s setup notes in your downloads folder, with the Cowork and Claude Code steps taken from Anthropic’s README and Help Center. Pick the plugin for your job first.
| Need | Detail |
|---|---|
| A Claude plan | Pro, Max, Ultra, Team or Enterprise, per the Help Center. Cowork runs in Claude Desktop and, on most of those plans, on the web and phone too. Claude Code is the terminal route |
| Accounts for your tools | A sign-in for each connector you use. Salesforce and Microsoft 365 in the sales plugin need an admin to enable them first |
| Node.js | Only for pdf-viewer. Its README says 18, and its server package lists 20 or later in the npm registry |
| Docker, Nextflow and Java 11 or later | Only for the bio-research Nextflow skill |
claude plugin marketplace add anthropics/knowledge-work-plugins claude plugin install sales@knowledge-work-plugins
Cowork can also take this repository as a marketplace: Plugins, Add, Add marketplace, Add from a repository, then anthropics/knowledge-work-plugins. Claude Code turns auto-update on by default for Anthropic’s official marketplaces, and its docs name this one as an exception. Plugins from it stay at the version you installed until you update them or turn auto-update on in /plugin, under Marketplaces.
claude plugin update sales@knowledge-work-plugins claude plugin uninstall sales@knowledge-work-plugins claude plugin marketplace remove knowledge-work-plugins
Start with the plugin’s first-run skill where it has one. After that, ask for the job in plain words, or type the command. The README’s own examples are /sales:call-prep and /data:write-query.
| Plugin | Type | What happens |
|---|---|---|
| productivity | /productivity:start | Creates TASKS.md, CLAUDE.md, a memory folder and dashboard.html in your working folder, then tells you to open the dashboard from your file browser |
| sales | /sales:setup | Checks which tools are connected, learns your writing voice from sent email and draws a starter dashboard |
| small-business | get me started | Connects two tools, runs one recipe on your data and asks about your business |
The legal plugin reads your playbook from a file named legal.local.md. In Cowork, save it in a folder you shared with Cowork. In Claude Code, save it in the project’s .claude folder. To tune any installed plugin in Cowork, open it and click Customize. Claude then opens a task that adjusts its skills and connectors with you.
A salesperson sends you an NDA from a prospect. In Cowork, you attach the file and type:
/legal:triage-nda
Claude screens the NDA against the playbook in legal.local.md, or against market-standard defaults when you have none, and says which it used. Back comes a triage report: a GREEN, YELLOW or RED label, the parties, term and governing law, and a table of screening results. Each issue comes with its risk and a suggested fix, then a next step. RED means do not sign. The skill adds that it assists with legal work and gives no legal advice.
If you are new. A plugin is a folder of instructions, not a program. Claude reads them when your request matches. Pick one plugin for the work you do most, connect a single tool and run its first-run skill. The sales skills also work with nothing connected: upload an export or paste your notes.
For experts. Skills name a category such as the CRM or the chat tool, so you can swap a connector by editing .mcp.json without rewriting them. The sales skills read field and picklist names from your live CRM instead of assuming a vendor’s, and they treat email, chat and transcripts as data, never as instructions. A scheduled run takes only the actions you set it up for. In Cowork, the cowork-plugin-management plugin builds a new plugin through a five-phase conversation and hands back a .plugin file. A nightly workflow opens a pull request to move an outside entry’s pin when its maker’s repository has moved on, and a Claude policy scan must pass before the new pin merges.
What a plugin can reach depends on what you connect to it.
# Anthropic knowledge-work plugins: setup notes This kit holds setup notes only. Mindket wrote them. The plugins install from Anthropic's repository: https://github.com/anthropics/knowledge-work-plugins These notes are not made or endorsed by Anthropic. The plugin manifests name Anthropic as the author of the plugins that are its own. Five folders under partner-built name their makers instead: Salesforce (Slack), Apollo.io, Common Room, Tribe AI and Zoom. License: Apache-2.0. The repository's LICENSE file is the Apache License 2.0 text. Its appendix line "Copyright [yyyy] [name of copyright owner]" is the unfilled template, so the file names no copyright holder. After the appendix it carries one more paragraph that is not part of the Apache terms, added in a pull request (#193, April 2026). GitHub reports the license as Apache-2.0. Source: https://github.com/anthropics/knowledge-work-plugins/blob/main/LICENSE The five partner-built folders carry their own LICENSE files: MIT for apollo (Apollo.io), brand-voice (Tribe AI), slack (Slack Technologies) and zoom-plugin (Zoom), and Apache-2.0 for common-room. The bio-research README says its skills are Apache 2.0 and its MCP servers are provided by their respective authors. A partner plugin listed in the marketplace but kept in the partner's own repository has that repository's license. ## What installs A plugin is a folder of Markdown and JSON. The README's "How Plugins Work" lists its parts: `.claude-plugin/plugin.json` (the manifest), `.mcp.json` (tool connections), `commands/` (slash commands you invoke) and `skills/` (knowledge Claude draws on). Source: https://github.com/anthropics/knowledge-work-plugins/blob/main/README.md The marketplace file (https://github.com/anthropics/knowledge-work-plugins/blob/main/.claude-plugin/marketplace.json, read on 2 November 2026) is named knowledge-work-plugins and lists 123 plugins: - 22 keep their files in this repository. - 101 point to their makers' own repositories. Each entry is pinned to one commit. The 17 plugins Anthropic built, with the folder each lives in: - productivity, enterprise-search, cowork-plugin-management, sales, finance, data, legal, marketing, customer-support, product-management and bio-research - engineering, human-resources, design, operations, small-business and pdf-viewer The five partner-built ones kept in this repository: slack-by-salesforce, apollo, common-room, brand-voice and zoom-plugin. The README's own table names 11 plugins, the first eleven above. The marketplace file lists more. Most plugins hold no code to run. The exceptions found in the repository tree on 2 November 2026: - bio-research: Python scripts under skills/. Only the Nextflow skill's scripts make network requests (NCBI, ENA, Docker Hub and nf-co.re addresses). - data: one Python script, package_data_skill.py, in the data-context-extractor skill. - productivity: skills/dashboard.html, a local page. It loads the Inter font from Google Fonts. - small-business: shared/artifact-style.md tells its skills to build their pages with fonts loaded from Google Fonts, and shared/reference/artifact-example.html is a sample page that does so. - pdf-viewer: its .mcp.json starts a local server with `npx -y @modelcontextprotocol/server-pdf --stdio`. ## What you need first - A Claude account on a paid plan (Pro, Max, Ultra, Team or Enterprise) for plugins in Claude and in Cowork, per the Claude Help Center: https://support.claude.com/en/articles/13837440-use-plugins-in-claude - Or Claude Code, for the terminal route. - An account at each service you connect (Slack, HubSpot, Notion and the like). Where the service needs it, the connector asks you to sign in. - For Salesforce and Microsoft 365 in the sales plugin: an admin has to enable the connector in Claude first, per sales/CONNECTORS.md. - For pdf-viewer: Node.js. Its README says 18 or later. The server package's registry entry (https://registry.npmjs.org/@modelcontextprotocol/server-pdf/latest, version 2.0.4 read on 2 November 2026) lists Node >=20, so use 20 or later. - For the bio-research Nextflow skill: Docker, Nextflow and Java 11 or later, per its SKILL.md. ## Main install command In Claude Code: ``` claude plugin marketplace add anthropics/knowledge-work-plugins claude plugin install sales@knowledge-work-plugins ``` Swap sales for any plugin name. In Cowork, the plugins come from Customize, Plugins, Discover. Steps are in setup/claude-cowork.md. ## Per-agent notes - setup/claude-cowork.md: Claude, Cowork and Claude Desktop - setup/claude-code.md: the terminal - setup/connectors.md: which connectors each plugin lists, and what to do with the ones left blank The README names only Cowork and Claude Code. ## What the plugins write and read - productivity: its start skill creates TASKS.md, CLAUDE.md, a memory/ folder and dashboard.html in the current working directory. Source: https://github.com/anthropics/knowledge-work-plugins/blob/main/productivity/skills/start/SKILL.md - legal: it reads a playbook from `legal.local.md`. In Cowork, save it in a folder you shared with Cowork. In Claude Code, save it in the project's `.claude/` directory. Without one, the NDA skill uses market-standard defaults and says so, and the contract skill offers to build a playbook or to review against general commercial standards. Source: https://github.com/anthropics/knowledge-work-plugins/blob/main/legal/README.md ## Update or remove The Claude Code docs say auto-update is on by default for Anthropic's official marketplaces except knowledge-work-plugins and first-party-plugins, so it is off here until you turn it on. Source: https://code.claude.com/docs/en/discover-plugins Update, uninstall and remove commands are in setup/claude-code.md. In Cowork, a marketplace comes out through its menu, Remove. Steps are in setup/claude-cowork.md. ## Usage data and what is sent The repository's files send nothing by themselves. What changes is what Claude can reach once a connector is signed in: - In Cowork, the Help Center says connectors reach outside services through Anthropic's cloud, and the work Claude does on a task, including local files it opens through the desktop app, is processed on Anthropic's servers. Source: https://support.claude.com/en/articles/13364135-use-cowork-safely - In Claude Code, the Help Center says a plugin runs in full on your computer, including its skills, sub-agents, hooks and MCP servers. - The sales skills treat email, chat, transcripts and enrichment results as untrusted data, and show an action that untrusted text asked for before it runs. Source: https://github.com/anthropics/knowledge-work-plugins/blob/main/sales/skills/setup/SKILL.md - What each connector keeps is its operator's policy. Most connectors are run by their own service, and setup/connectors.md names the ones hosted elsewhere. ## Sources - README: https://github.com/anthropics/knowledge-work-plugins/blob/main/README.md - LICENSE: https://github.com/anthropics/knowledge-work-plugins/blob/main/LICENSE - Marketplace file: https://github.com/anthropics/knowledge-work-plugins/blob/main/.claude-plugin/marketplace.json - Use plugins in Claude (Help Center): https://support.claude.com/en/articles/13837440-use-plugins-in-claude - Claude Code docs, discover plugins: https://code.claude.com/docs/en/discover-plugins
README.md 7 KB setup/claude-code.md 4 KB setup/claude-cowork.md 4 KB setup/connectors.md 4 KB
Updated by Mindket to the current version of this kit.
Posts and comments that name this project on Hacker News, Bluesky, X, Reddit, Facebook and YouTube, found by a weekly search and on Trendshift. Mindket doesn’t write or check them. Open one to read it where it was posted.
well9472 · Kit
Makes illustrations from text prompts with a small 670M-parameter diffusion model, run through ComfyUI. Built as a low-budget research model for people who wan…
Waqf-AI · Kit
Reads handwritten Arabic text from images and returns it as text. A fine-tuned PaddleOCR-VL model for people digitizing Arabic handwriting.
Wan-AI · Kit
Makes video clips from a text prompt or an image plus text, at up to 720P and 24 frames per second, in the Diffusers format for a single consumer GPU.
perplexity-ai · Kit
Picks the best option for a decision from text and images, such as which team should handle a request, with a probability for each choice, for developers who b…
numind · Kit
Reads documents as text or images and fills a JSON template with the fields you ask for, or converts the pages to Markdown. A 4B model for teams that process s…
nace-ai · Kit
Scores yes or no, multiple choice and rating questions about a document in one pass, with a probability for every option. An 8B decision model from Nace.AI tha…