Skip to content
/

Anthropic Knowledge Work Plugins

anthropics-knowledge-work-plugins · anthropics/knowledge-work-plugins · ★ 16k · last commit 2026-05-26

Primitive shape 135 total
Commands 25 Skills 110
00

Summary

Anthropic Knowledge Work Plugins — Summary

Anthropic's Knowledge Work Plugins is the official collection of role-specialized plugin bundles built for Claude Cowork (and compatible with Claude Code), each targeting a specific job function (Engineering, Sales, Finance, Legal, Marketing, etc.) by bundling domain skills, slash commands, and MCP connector configurations into a single installable unit.

Problem it solves: Knowledge workers in specific roles (engineer, salesperson, legal professional, etc.) need domain-specific workflows, terminology, and tool integrations pre-loaded for Claude to function as a true role specialist; generic skill packs don't know whether they're helping a finance analyst or an engineer.

Distinctive trait: Every plugin follows a four-part structure (skills + commands + .mcp.json connectors + plugin.json manifest), and the .mcp.json files wire Claude directly to real enterprise tools (Slack, Linear, HubSpot, Asana, Jira, Notion, Snowflake, BigQuery) via HTTP MCP servers — making these the only official skill packs in the corpus that ship production MCP connector configurations.

Target audience: Enterprises deploying Claude Cowork for specific role teams; also usable individually via Claude Code with /plugin install.

Scope: 11+ first-party role plugins (plus 30+ partner-built plugins in the marketplace.json), covering engineering, sales, finance, legal, marketing, data, HR, design, operations, product management, customer support, and bio-research.

Differs from seeds: Unlike superpowers or agent-skills (Archetype 1 — methodology enforcement), knowledge-work-plugins are domain-specific reference packs combining skills + commands + live MCP connector configs. The MCP connector pattern makes them closer to taskmaster-ai (Archetype 3) but the value is domain expertise, not task decomposition.

01

Overview

Anthropic Knowledge Work Plugins — Overview

Origin

Official Anthropic repository. Apache-2.0 license. Stars: 16,165 with 23 contributors. Active as of 2026-05-26 (same-day push). Built for Claude Cowork product.

Purpose

From the README:

"Plugins that turn Claude into a specialist for your role, team, and company. Built for Claude Cowork, also compatible with Claude Code." "Each plugin bundles the skills, connectors, slash commands, and sub-agents for a specific job function. Out of the box, they give Claude a strong starting point for helping anyone in that role. The real power comes when you customize them for your company — your tools, your terminology, your processes — so Claude works like it was built for your team."

Design Philosophy

"Every component is file-based — markdown and JSON, no code, no infrastructure, no build steps."

"These plugins are generic starting points. They become much more useful when you customize them for how your company actually works:

  • Swap connectors — Edit .mcp.json to point at your specific tool stack.
  • Add company context — Drop your terminology, org structure, and processes into skill files.
  • Adjust workflows — Modify skill instructions to match how your team actually does things."

Plugin Structure Standard

Every plugin follows:

plugin-name/
├── .claude-plugin/plugin.json   # Manifest
├── .mcp.json                    # Tool connections (HTTP MCP servers)
├── commands/                    # Slash commands you invoke explicitly
└── skills/                      # Domain knowledge Claude draws on automatically

Partner Plugin Ecosystem

The marketplace.json lists 30+ third-party partner plugins (Figma, Zapier, Zoom, Box, Prisma, CockroachDB, Mintlify, Postiz, Intercom, Cloudinary, etc.) alongside the first-party plugins — making this the most expansive plugin marketplace in the batch.

Cowork vs Code

Built primarily for Claude Cowork (team/enterprise product) but Claude Code users can install via:

claude plugin marketplace add anthropics/knowledge-work-plugins
claude plugin install sales@knowledge-work-plugins
02

Architecture

Anthropic Knowledge Work Plugins — Architecture

Distribution

  • Type: Claude Code plugin marketplace (Claude Cowork primary; Claude Code secondary)
  • Claude Code install:
    claude plugin marketplace add anthropics/knowledge-work-plugins
    claude plugin install sales@knowledge-work-plugins
    claude plugin install engineering@knowledge-work-plugins
    
  • Claude Cowork: Install from claude.com/plugins/

Required Runtime

None — pure markdown and JSON files. MCP connectors reference external HTTP services (Slack MCP at mcp.slack.com, Linear at mcp.linear.app, etc.) that require network access and OAuth credentials.

Repository Structure

anthropics/knowledge-work-plugins/
├── .claude-plugin/
│   └── marketplace.json         # 20+ first-party + 30+ partner plugins
├── engineering/
│   ├── .claude-plugin/plugin.json
│   ├── .mcp.json                # Slack, Linear, Asana, Jira, Notion, GitHub
│   ├── CONNECTORS.md
│   ├── README.md
│   └── skills/
│       ├── architecture/
│       ├── code-review/
│       ├── debug/
│       ├── deploy-checklist/
│       ├── documentation/
│       ├── incident-response/
│       ├── standup/
│       ├── system-design/
│       ├── tech-debt/
│       └── testing-strategy/
├── sales/
│   ├── .claude-plugin/plugin.json
│   ├── .mcp.json                # Slack, HubSpot, Close, Clay, ZoomInfo, etc.
│   ├── skills/
│   │   ├── account-research/
│   │   ├── call-prep/
│   │   ├── call-summary/
│   │   ├── competitive-intelligence/
│   │   ├── create-an-asset/
│   │   ├── daily-briefing/
│   │   ├── draft-outreach/
│   │   ├── forecast/
│   │   └── pipeline-review/
├── finance/
├── legal/
├── marketing/
├── data/
├── product-management/
├── customer-support/
├── productivity/
├── enterprise-search/
├── bio-research/
├── human-resources/
├── design/
├── operations/
├── small-business/
├── cowork-plugin-management/
├── pdf-viewer/
└── partner-built/
    ├── slack/ (Salesforce)
    ├── apollo/ (Apollo.io)
    ├── common-room/
    └── brand-voice/ (Tribe AI)

Engineering Plugin .mcp.json (representative)

{
  "mcpServers": {
    "slack": {"type": "http", "url": "https://mcp.slack.com/mcp", "oauth": {...}},
    "linear": {"type": "http", "url": "https://mcp.linear.app/mcp"},
    "asana": {"type": "http", "url": "https://mcp.asana.com/v2/mcp"},
    "atlassian": {"type": "http", "url": "https://mcp.atlassian.com/v1/mcp"},
    "notion": {"type": "http", "url": "https://mcp.notion.com/mcp"},
    "github": {"type": "http", "url": "https://api.githubcopilot.com/mcp/"}
  }
}

All MCP connections are HTTP (streamable-http / SSE), not stdio.

Plugin Manifest Structure

Each plugin has its own .claude-plugin/plugin.json:

{
  "name": "engineering",
  "version": "1.2.0",
  "description": "...",
  "author": {"name": "Anthropic"}
}

CONNECTORS.md

Each plugin includes CONNECTORS.md listing available integrations with instructions for connecting each tool.

03

Components

Anthropic Knowledge Work Plugins — Components

Commands (per plugin)

Commands are slash commands you invoke explicitly. Representative set from engineering plugin:

Plugin Command Purpose
engineering /engineering:code-review Review code changes for security, performance, correctness
engineering (implied by skills) Architecture docs, incident response, standup, etc.
sales /sales:call-prep Prepare for a sales call
sales /sales:pipeline-review Review sales pipeline
data /data:write-query Write SQL/data queries
finance /finance:reconciliation Run account reconciliation
product-management /product-management:write-spec Write feature specification

Exact command counts per plugin: not fully enumerable without listing each plugin's commands/ directory, but the README confirms commands exist for each plugin.

Skills (per plugin, representative)

Engineering Plugin (10 skills)

Name Purpose
code-review Review PRs for security, performance, correctness, style
architecture Architecture decision records and system design
debug Systematic debugging workflows
deploy-checklist Pre-deployment verification
documentation Technical documentation
incident-response Incident triage and resolution
standup Daily standup summaries
system-design System design guidance
tech-debt Tech debt identification and prioritization
testing-strategy Testing approach and coverage

Sales Plugin (9 skills)

Name Purpose
account-research Research prospects and accounts
call-prep Prepare for sales calls
call-summary Summarize sales calls
competitive-intelligence Competitor analysis
create-an-asset Create sales materials
daily-briefing Daily sales briefing
draft-outreach Write personalized outreach
forecast Sales forecasting
pipeline-review Pipeline health review

Other plugins follow similar 8-12 skill counts per domain.

MCP Connectors (per plugin)

This is the unique component. Each plugin's .mcp.json configures HTTP MCP servers for real enterprise tools:

Plugin Connectors
Engineering Slack, Linear, Asana, Atlassian, Notion, GitHub
Sales Slack, HubSpot, Close, Clay, ZoomInfo, Notion, Jira, Fireflies, Microsoft 365
Finance Snowflake, Databricks, BigQuery, Slack, Microsoft 365
Data Snowflake, Databricks, BigQuery, Definite, Hex, Amplitude, Jira
Marketing Slack, Canva, Figma, HubSpot, Amplitude, Notion, Ahrefs, SimilarWeb, Klaviyo
Legal Slack, Box, Egnyte, Jira, Microsoft 365
Product Mgmt Slack, Linear, Asana, Monday, ClickUp, Jira, Notion, Figma, Amplitude, etc.

Partner Plugins in Marketplace (30+)

Representative partner plugins (from marketplace.json): Figma, Zapier, Zoom, Box, Prisma, CockroachDB, Mintlify, Postiz, Intercom, Cloudinary, Nimble, Bright Data, SearchFit SEO, Atlan, Miro, PlanetScale, LSEG, S&P Global, Slack (Salesforce), Apollo.io, Common Room, ZoomInfo, Vanta, Sanity, and more.

Subagents

None defined explicitly; plugin descriptions mention "sub-agents" in the cowork-plugin-management plugin description but no persona files found.

Hooks

None.

05

Prompts

Anthropic Knowledge Work Plugins — Prompts

Prompt 1: engineering/code-review — Supercharged vs. Standalone Pattern

File: engineering/skills/code-review/SKILL.md Technique: Progressive enhancement with connector awareness

## How It Works

┌─────────────────────────────────────────────────────────────────┐ │ STANDALONE (always works) │ │ ✓ Paste a diff, PR URL, or point to files │ │ ✓ Security audit (OWASP top 10, injection, auth) │ │ ✓ Performance review (N+1, memory leaks, complexity) │ │ ✓ Correctness (edge cases, error handling, race conditions) │ │ ✓ Style (naming, structure, readability) │ │ ✓ Actionable suggestions with code examples │ ├─────────────────────────────────────────────────────────────────┤ │ SUPERCHARGED (when you connect your tools) │ │ + Source control: Pull PR diff automatically │ │ + Project tracker: Link findings to tickets │ │ + Knowledge base: Check against team coding standards │ └─────────────────────────────────────────────────────────────────┘


> If you see unfamiliar placeholders or need to check which tools are connected, see [CONNECTORS.md](../../CONNECTORS.md).

Technique analysis: Two-tier capability model (standalone vs. supercharged) clearly communicates which features require connected MCP tools and which work without. The CONNECTORS.md reference creates a discoverable index of available integrations. This pattern handles graceful degradation when tools aren't connected.


Prompt 2: sales — Representative Skill Structure (inferred)

File: sales/skills/call-prep/SKILL.md (structure inferred from code-review pattern) Technique: Connector-aware workflow with @argument-hint

---
name: call-prep
description: Prepare for a sales call...
argument-hint: "<contact name, company, or meeting details>"
---

The argument-hint field in the YAML frontmatter provides Codex/Claude Code with context about what arguments to pass, enabling better slash command UX.


Prompt 3: README Philosophy Excerpt

Source: Repository README Technique: Customization-first framing

## Making Them Yours

These plugins are generic starting points. They become much more useful when you customize them for how your company actually works:

- **Swap connectors** — Edit `.mcp.json` to point at your specific tool stack.
- **Add company context** — Drop your terminology, org structure, and processes into skill files so Claude understands your world.
- **Adjust workflows** — Modify skill instructions to match how your team actually does things, not how a textbook says to.
- **Build new plugins** — Use the `cowork-plugin-management` plugin or follow the structure above to create plugins for roles and workflows we haven't covered yet.

Technique analysis: Positions the plugins as starters, not finished products. The customization instruction is embedded in the product README — encouraging organizational forking rather than plugin-as-given-truth usage.

09

Uniqueness

Anthropic Knowledge Work Plugins — Uniqueness

Differs from Seeds

No seed framework parallels this role-specialized plugin design with live MCP connector bundles:

  • Unlike superpowers and agent-skills (Archetype 1 — development methodology), knowledge-work-plugins are role-specific domain plugins with no development workflow enforcement.
  • Unlike anthropics/skills (reference gallery for developers), knowledge-work-plugins target knowledge workers — sales reps, legal teams, finance analysts, engineers — in enterprise contexts.
  • The MCP connector pattern (.mcp.json with HTTP MCP servers for Slack, HubSpot, Snowflake, etc.) is closest to ccmemory's Archetype 3 (MCP-anchored toolserver), but knowledge-work-plugins don't ship their own MCP server — they configure connections to external hosted MCP services.
  • The marketplace.json with 50+ plugins (first-party + partner) is the largest curated plugin marketplace in the batch.

Positioning

The enterprise-grade, role-specialized counterpart to the more developer-centric skill packs. Designed for non-developer knowledge workers who need Claude to understand their domain, tools, and workflows without configuration.

Most Interesting Component

The .mcp.json HTTP connector pattern — particularly the OAuth-gated Slack connector at mcp.slack.com/mcp — represents production-level infrastructure that doesn't require the user to configure MCP servers themselves. This is "batteries included" for enterprise MCP.

The partner ecosystem marketplace (Figma, Zapier, Zoom, Box, LSEG, S&P Global, Vanta, etc.) converts the repository into a verified plugin registry for enterprises.

Observable Failure Modes

  1. Connector dependency — "Supercharged" features require OAuth setup for each tool. Without connected tools, the skills degrade to standalone text-based workflows.
  2. No session enforcement — skills activate by description matching only; no guarantee the right skill fires for a given enterprise workflow.
  3. Generic starting points — the README acknowledges plugins need customization for each organization. Teams that don't customize get generic domain skills without company-specific context.
  4. Claude Cowork lock — optimal experience requires the Cowork product; Claude Code install works but some Cowork-specific features may not transfer.
  5. Partner plugin SHA pinning — partner plugins are pinned to specific commits in marketplace.json; there's no automatic update mechanism visible in the repo.

Explicit Antipatterns

From README: "The real power comes when you customize them for your company" — implies using without customization is an antipattern.

04

Workflow

Anthropic Knowledge Work Plugins — Workflow

General Pattern

No enforced workflow across plugins. Each plugin activates skills automatically when Claude determines they're relevant, and exposes slash commands for explicit invocation. The pattern is:

  1. Install plugin for your role
  2. Claude draws on skills automatically when relevant to your request
  3. Use slash commands (e.g., /engineering:code-review) for explicit actions
  4. MCP connectors provide live data from connected tools

Engineering Plugin Workflow (Code Review)

┌─────────────────────────────────────────────────────────────────┐
│                      CODE REVIEW                                   │
├─────────────────────────────────────────────────────────────────┤
│  STANDALONE (always works)                                       │
│  ✓ Paste a diff, PR URL, or point to files                      │
│  ✓ Security audit (OWASP top 10, injection, auth)               │
│  ✓ Performance review (N+1, memory leaks, complexity)           │
│  ✓ Correctness (edge cases, error handling, race conditions)    │
├─────────────────────────────────────────────────────────────────┤
│  SUPERCHARGED (when you connect your tools)                      │
│  + Source control: Pull PR diff automatically                    │
│  + Project tracker: Link findings to tickets                     │
│  + Knowledge base: Check against team coding standards           │
└─────────────────────────────────────────────────────────────────┘

Customization Workflow

Plugins are designed to be forked and customized:

  1. Install plugin
  2. Swap connectors — edit .mcp.json for your tool stack
  3. Add company context — add terminology, org structure to skill files
  4. Adjust workflows — modify skill instructions for your team's processes
  5. Build new plugins — use cowork-plugin-management plugin or follow the structure

Partner Plugin Workflow

Partners submit plugins via GitHub URL reference in marketplace.json. The source field supports:

  • "url" — direct GitHub repo URL with SHA pinning
  • "git-subdir" — git subdirectory with SHA pinning
  • Both require explicit SHA for version control

Phase-to-Artifact Map

Phase Artifact
Code review Categorized findings
Call prep Meeting brief
Call summary Meeting notes + action items
Draft outreach Personalized email/message
Reconciliation Reconciliation report
Spec writing Feature spec document

Approval Gates

None at the framework level — MCP connectors may require OAuth flows for connecting tools (Slack OAuth, etc.), which are one-time setup gates.

06

Memory Context

Anthropic Knowledge Work Plugins — Memory & Context

State Storage

  • Type: None at framework level — state lives in connected MCP tool systems (Slack, Notion, HubSpot, etc.)
  • Persistence: External (data persists in connected tools, not in the plugin)

Session Context

No SessionStart hook. Skills activate by description matching. Slash commands require explicit invocation.

Live Data via MCP Connectors

The real memory layer is the MCP connector network. When tools are connected:

  • Slack MCP (mcp.slack.com) — access to message history, channels
  • Notion MCP (mcp.notion.com) — access to company docs and wikis
  • GitHub MCP (api.githubcopilot.com/mcp) — access to code, PRs, issues
  • Jira/Atlassian MCP (mcp.atlassian.com) — access to tickets and boards
  • HubSpot — CRM data for sales context
  • Snowflake/BigQuery/Databricks — data warehouse query access

These connectors convert the Claude session into a read/write interface to live company data, which is the closest thing to "memory" in this system.

Cross-Session Handoff

Indirect — meeting notes saved to Notion, tickets created in Linear, summaries posted to Slack persist across sessions via those external tools.

Context Compaction

No explicit mechanism.

Via connected MCP tools:

  • enterprise-search plugin enables cross-tool search (Slack, Notion, Jira, Asana, email, wikis)
  • Individual connectors provide native search within each tool
07

Orchestration

Anthropic Knowledge Work Plugins — Orchestration

Multi-Agent Support

Implied but not formally defined. The cowork-plugin-management description mentions "sub-agents" but no persona files are found in the repository. The README mentions each plugin bundles "skills, connectors, slash commands, and sub-agents" but sub-agent implementations are not visible in the public repository.

Orchestration Pattern

None at framework level. Individual skills may sequence steps (e.g., code-review collects diff → analyzes → outputs findings), but this is single-agent sequential, not multi-agent.

Isolation Mechanism

None.

Multi-Model

No — single model.

Execution Mode

Event-driven — skills activate by description matching, commands by explicit invocation.

Consensus Mechanism

None.

Prompt Chaining

Weak — the STANDALONE → SUPERCHARGED pattern effectively chains the skill with MCP tool outputs, but this is conditional enrichment rather than formal prompt chaining.

Auto-Validators

No — no hooks, no auto-running tests or linters.

Git Automation

The engineering code-review skill can pull PR diffs via GitHub MCP connector when configured, but does not commit or merge automatically.

External Tool Automation

Via MCP connectors, the skills can:

  • Post summaries to Slack
  • Create tickets in Linear/Jira/Asana
  • Read CRM data from HubSpot
  • Query databases via Snowflake/BigQuery

This is the closest the framework comes to "automation" — but it requires human approval to write to external systems in most workflows.

08

Ui Cli Surface

Anthropic Knowledge Work Plugins — UI & CLI Surface

Dedicated CLI Binary

None.

Claude Code Install

claude plugin marketplace add anthropics/knowledge-work-plugins
claude plugin install sales@knowledge-work-plugins
claude plugin install engineering@knowledge-work-plugins

Claude Cowork

Primary deployment surface: claude.com/plugins/. Full GUI-based plugin browsing and installation through Claude Cowork product.

Local Web Dashboard

None in the repository. The Cowork product has a web UI (claude.com) but this is a hosted product, not a local dashboard.

MCP Connection UI

The .mcp.json HTTP connectors require OAuth flows for tools like Slack (OAuth 2.0 with callbackPort: 3118). This OAuth flow appears in the host application (Claude Code or Cowork), not in a local dashboard.

CONNECTORS.md

Each plugin includes a CONNECTORS.md file that lists available tool integrations with setup instructions — serves as a discovery surface for what can be connected.

Cross-Tool Portability

Low-medium — plugins target Claude Cowork primarily, with Claude Code as secondary. The .claude-plugin/ structure and .mcp.json format are Claude-specific. No Cursor/Gemini/Codex compatibility.

Observability

No audit log, no replay. External tool systems (Slack, Notion, Jira) serve as the operational audit trail.

Related frameworks

same archetype · same primary tool · same memory type

BMAD-METHOD ★ 48k

Provides a full agile delivery lifecycle with named expert-persona AI collaborators that elicit the human's best thinking rather…

Agent OS ★ 4.6k

Extracts implicit codebase conventions into token-efficient markdown standards files and injects them selectively into AI agent…

Claude Conductor ★ 367

Gives Claude Code a persistent, cross-linked, auto-analyzed documentation system so it retains codebase context across sessions.

Spec-Driver (Greenfield Spec-Driven Development) ★ 25

Prevents spec rot in AI-assisted development by making implementation changes flow back into evergreen, authoritative specs via…

Codex Integration for Claude Code (skill-codex) ★ 1.3k

Single Claude Code skill that handles Codex CLI invocation correctly (stdin blocking, thinking token suppression, session resume)…

mini-coding-agent ★ 882

A single-file zero-dependency Python coding agent that demonstrates the six core components of coding agents for educational…