Claude Code is powerful just for writing code. However, what truly takes up time in professional work isn't just the implementation itself. It involves searching for past decisions, organizing specifications, writing SQL, reviewing designs, checking before release, reading contracts and PDFs, and creating reports for stakeholders. The work of developers and product teams extends far beyond the code.
That's where Claude's official plugins come in.
This article doesn't cover the development plugins from the Claude Code CLI-specific marketplace. Instead, we are looking at the official Anthropic business-specific plugins that can be selected from the "Directory -> Plugins -> Anthropic" tab in Claude Desktop, as shown in the image. Specifically, these include Engineering, Product Management, Data, Design, Enterprise Search, PDF Viewer, and more.
While these are primarily provided for Claude Desktop's Chat and Cowork, Anthropic's official repository states they also work with Claude Code. By importing the same packages into Claude Code, you can bring skills, slash commands, and MCP connectors into your development workflow. In short, the premise of this article is "using official plugins found on the desktop within Claude Code."
As of July 18, 2026, plugins are available on paid plans: Pro, Max, Team, and Enterprise. Plugins bundle skills, connectors, and sub-agents. Installed skills can be used in the web chat, Claude Desktop Chat, and Cowork. However, since some hooks and sub-agents are exclusive to Cowork, you should check "what runs where" for each plugin.
In this article, I have selected 10 plugins based on whether they can be used repeatedly in practice by developers, product managers, designers, technical leads, and solo developers, rather than just those with many features.
First, the Installation Method You Should Know
In Claude Desktop, open "Customize" in the left sidebar, go to the "Plugins" tab, and press "+" to display the directory. Then, select the "Anthropic" tab and press "+" or "Install" on the target card. After installation, you can check the added skills and commands by typing "/" in the input field or opening the "+" menu.
To use the same plugins in Claude Code, add the official Knowledge Work marketplace and install them individually.
claude plugin marketplace add anthropics/knowledge-work-plugins
claude plugin install engineering@knowledge-work-plugins
You can change engineering to data, design, product-management, etc. Once installed, they are automatically referenced when skills are needed, and explicit commands can be called with namespaces like /engineering:review or /data:write-query.
Note that some plugins start local MCP servers or connect to external services like Google Drive, Slack, GitHub, and Figma. Since local MCPs may run on your terminal with permissions equivalent to normal programs, always check the provider, required permissions, connection destinations, and whether they have write access.
1. Engineering | The Essential Basic Set for Dev Teams
Engineering is the top priority. Its scope is broad, covering stand-ups, code reviews, debugging, architecture decisions, incident response, pre-deployment checks, and technical documentation. It serves as the foundation for using Claude Code not just as an implementation tool, but as a partner supporting the entire development process.
Representative commands include /engineering:review for reviewing changes, /engineering:debug for reproduction/isolation/root cause identification/fixing, /engineering:architecture for organizing technical decisions in ADR format, /engineering:incident for failure response support, and /engineering:deploy-checklist for checking for omissions before release. It includes knowledge of test strategies, technical debt, system design, and documentation, not just code review skills.
It is particularly useful in the stages before and after writing code. For example, before implementation, use /engineering:architecture to decide whether to make a function synchronous or use a queue; after implementation, run /engineering:review; and before production, run /engineering:deploy-checklist. During failures, instead of just staring at logs, you can record reproduction conditions, impact scope, hypotheses, verification results, mitigations, and permanent fixes chronologically.
For a first try, the following request is easy to understand:
/engineering:review
Please check this diff in the order of correctness, security, performance, and maintainability.
List high-severity issues first, and provide a rationale and suggested fix for each point.
However, review results should not replace tests or CI. Engineering is strong as a plugin for structuring decision-making materials; it does not guarantee build success, type checks, or behavior in a real environment. Its maximum value is realized when implementation, testing, and review are integrated into a single flow.
2. Product Management | Turning Vague Ideas into Implementable Specs
If you suddenly throw "make this feature" at Claude Code, it will produce a plausible implementation. However, if who the problem is being solved for, what is out of scope, and how success is measured remain vague, significant rework will occur after completion. Product Management is a plugin that reinforces this upstream process.
It includes /product-management:write-spec for writing functional specs or PRDs, /product-management:roadmap-update for updating roadmaps, /product-management:stakeholder-update for progress reports, /product-management:synthesize-research for summarizing interviews/surveys, /product-management:competitive-brief for competitor comparisons, /product-management:metrics-review for checking indicators, and /product-management:brainstorm for deep-diving into hypotheses. Common PM frameworks like RICE, MoSCoW, Jobs-to-be-Done, and Opportunity Solution Trees are also built-in as skills.
The synergy with Claude Code lies in "connecting specs and implementation in the same conversation." First, organize problems, target users, acceptance criteria, non-functional requirements, success metrics, and pending items with /product-management:write-spec. Then, drop that content into Issues, break it down into implementation tasks, and finally review it with Engineering. This makes it easier to keep the reasons for code changes consistent without the spec document gathering dust elsewhere.
In practice, rather than having it write the final form in one go, it's important to first specify: "Ask about unclear points and clearly state which items are assumptions." The PM plugin is used not just to polish text, but to expose ambiguity. A PRD that reveals questions to be crushed before implementation is more valuable than a pretty PRD.
3. Enterprise Search | Finding Past Decisions Scattered in Slack, Email, and Docs
One of the most wasteful tasks in team development is the time spent searching for "where was that decided?" When specs are in Notion, discussions in Slack, approvals in email, status in Jira, and final docs in Google Drive, just searching drains concentration.
Enterprise Search cross-searches connected chats, emails, cloud storage, wikis, project management, CRM, and ticket management, deduplicating results into a single answer. When you throw a question at /enterprise-search:search, Claude breaks the question into queries for each source and integrates them with citations. With /enterprise-search:digest --daily or --weekly, you can summarize decisions, action items, and mentions by topic.
For developers, it's powerful for finding past ADRs, incident responses, API change histories, owners of specific features, and customer requests. For example, you can ask, "Why did we decide to use an external IdP instead of building our own auth?" or "Who is the owner of this table?" or "What was the previous temporary fix for the payment failure?" It's also effective for onboarding new members.
The weakness is that it can't find information in places it's not connected to, and the search range is as wide as your access rights. Start by connecting only essential sources like Slack, documentation, and project management, rather than indiscriminately expanding to private DMs or confidential folders. Always have it attach source documents to answers and distinguish between "not found" and "does not exist" for safe use.
4. Data | From SQL and Visualization to Analysis Verification
Data is not just a plugin for generating SQL. It treats data exploration, quality checks, statistical analysis, visualization, HTML dashboard creation, and pre-sharing verification as a single analysis process.
Main commands include /data:analyze for proceeding with analysis from questions, /data:explore-data for checking dataset shapes, missing values, and anomalies, /data:write-query for writing SQL, /data:create-viz for making graphs with Python, /data:build-dashboard for interactive HTML dashboards, and /data:validate for checking analysis methods and aggregation logic. It can connect to Snowflake, Databricks, BigQuery, etc., via MCP, and can also handle CSV, Excel, or pasted results without a connection.
What's convenient in Claude Code is the ability to go back and forth between analysis SQL and product code. For example, investigate drop-off points before implementing a new onboarding flow, recalculate metrics with the same definition after release, and visualize the difference. In bug investigations, you can quickly verify hypotheses like "is the failure rate high only for users of a specific version?" or "did NULLs increase after data migration?"
The most valuable command is surprisingly /data:validate. Even if AI can write SQL quickly, it can easily err in denominator selection, duplicate rows, time zones, survivorship bias, or inclusion of test users. You should have it validate as a separate step after analysis, specifying the tables used, filters, periods, metric definitions, and exclusion criteria. When connecting to production DBs, limit it to read-only permissions initially.
5. Design | Connecting Design Reviews to Implementation Handoffs
Design is not a plugin for generating images that look "okay." It's a practical set for product design, covering design critique, design system management, UX writing, accessibility audits, user research integration, and developer handoffs.
/design:critique reviews from the perspectives of usability, visual hierarchy, consistency, and accessibility, while /design:design-system audits components, tokens, naming, and patterns. /design:handoff creates implementation specs including dimensions, states, interactions, and edge cases, and /design:ux-copy assists with microcopy for error messages, empty states, and onboarding. /design:accessibility and /design:research-synthesis are also available.
When using it with Claude Code, it's better to find holes in the specs with Design before having it turn Figma visuals directly into code. Have it list states that aren't obvious from screenshots, such as "there's a hover, but what about keyboard focus?" or "how do we show loading, empty states, insufficient permissions, or communication failures?" or "will it break with long Japanese text or 200% zoom?" Then, pass the handoff to Claude Code as implementation requirements.
Regarding accessibility audits, while it can streamline checks from a WCAG perspective, it does not replace verification with real browsers, screen readers, keyboard operations, or real users. The correct use is to create review items with Design and connect them to browser tests and human verification.
6. PDF Viewer | Not Just Reading PDFs, but Fixing While Viewing
PDF Viewer has a different role than Claude's native PDF summarization. It opens local files or direct PDF URLs in an interactive viewer to highlight, annotate, add stamps, fill forms, place signature images, and save the edited PDF.
Use /pdf-viewer:open to display and /pdf-viewer:annotate to reflect annotation suggestions page by page. /pdf-viewer:fill-form fills input fields in order, and /pdf-viewer:sign places a signature or initial image. It runs as a local MCP server using @modelcontextprotocol/server-pdf via npx, with a requirement of Node.js 18 or higher.
For developers, it can be used for reviewing API specs, requirement definitions, security audit reports, vendor proposals, and contracts. Instead of just saying "summarize the problems," asking it to "place annotations where changes are needed, and organize questions in yellow and blockers in red labels" makes it a deliverable you can return to the other party.
On the other hand, if you just want to read the content, Claude's native PDF reading is faster. You should use PDF Viewer when you want to write while visually confirming and take away the final file. Also, note that sign places a visual signature image, not a cryptographic electronic signature using a certificate. For contracts requiring legal force, dedicated electronic signature services must be used.
7. Operations | Turning Individual Tasks into SOPs and Runbooks
Operations looks like it's for business management, but it's quite usable in dev organizations too. It assists with vendor evaluation, business process documentation, change management, capacity planning, management status reports, and runbook creation.
/operations:vendor-review organizes costs, risks, contracts, and renewal decisions, while /operations:process-doc creates flows, RACI, and SOPs. /operations:change-request creates change requests including impact analysis, approval paths, and rollback plans, and /operations:capacity-plan analyzes load and personnel. /operations:runbook turns routine tasks into repeatable documents including procedures, checklists, troubleshooting, and escalation points.
In combination with Claude Code, it's powerful for freeing releases and operations from "memory outside the code." For example, summarize database migration procedures into a change request, creating pre-execution checks, monitoring items, abort conditions, rollback SQL, personnel, and contact messages. After a failure, create a post-mortem with Engineering and reflect it in the runbook with Operations.
One caution is never to make a procedure written by Claude official without trying it once. Runbooks should be executed in a staging environment, with humans verifying commands, permissions, time required, and how to revert. Operations speeds up document creation, but rehearsals are necessary to finish them into procedures that work on the ground.
8. Marketing | Running the "Delivery Work" After Release with Claude Code
Even if you build a good feature, it won't be used if the release notes, blogs, emails, LPs, SNS, and sales explanations are weak. Marketing assists with content creation and campaign design that occur after development.
/marketing:draft-content creates blogs, SNS, newsletters, LPs, press releases, and case studies, while /marketing:campaign-plan creates plans including goals, targets, channels, schedules, and KPIs. /marketing:brand-review checks consistency with the brand voice, and /marketing:competitive-brief, /marketing:performance-report, /marketing:seo-audit, and /marketing:email-sequence are also available. Integration with Slack, Canva, Figma, HubSpot, Amplitude, Notion, Ahrefs, Similarweb, Klaviyo, etc., is expected.
For Claude Code users, the flow of having it create promotional materials from code diffs is convenient. Read the changed features, target users, known constraints, and migration steps from the repository, and create technical release notes, general user announcements, and sales FAQs separately. Since the source of the content is the same, explanations across channels are less likely to conflict.
However, if used without brand settings, it tends to produce safe, AI-like text. Providing forbidden expressions, glossaries, representative past articles, how to address customers, and the range of what can be asserted, and then passing it through brand-review at the end, increases its utility. For performance reports, if the definition of connected data is vague, it will misinterpret conclusions, so KPI definitions and comparison periods should be fixed.
9. Legal | Speeding Up Contract Reviews, but Humans Hold the Final Judgment
Legal handles contract reviews for internal legal departments, NDA primary determinations, compliance, legal briefs, and standard responses. What's particularly important is that instead of reading contracts in general terms, you can set your company's negotiation policy and risk tolerance in legal.local.md and compare against it.
/legal:review-contract finds differences from the company playbook for each clause and organizes risks and suggested revisions. /legal:triage-nda performs primary classification like GREEN, YELLOW, RED, and /legal:vendor-check checks existing NDAs, MSAs, DPAs, deadlines, and key conditions from connection destinations. /legal:brief and /legal:respond can create case summaries and draft responses to standard inquiries.
In development settings, it can be used for preliminary organization of SaaS contracts, cloud terms of use, DPAs, NDAs, outsourcing contracts, and security clauses. It's realistic to use it to extract points of discussion like data location, sub-processors, liability limits, intellectual property, termination, and audit rights before handing them off to legal, and to create a list of questions.
However, the official README also clearly states that this is not legal advice and requires verification by a qualified professional. Also, since initial playbook examples are based on US law and business practices, you must rebuild settings if using it under Japanese law or company policy. It's safer to limit AI's determination to point extraction and primary organization rather than making it the final gate in the approval flow.
10. Small Business | The One That Transforms Most for Solo Devs and Small Businesses
Small Business is a plugin that handles small business operations as a whole, rather than assisting a specific job type. It features 15 basic skills, 15 execution workflows, and a router that guides you to the appropriate process from natural language. If you consult it normally with things like "I'm worried if I can pay salaries," "sales have dropped," "I got an angry email from a customer," or "should I raise prices?", it's designed to guide you to the necessary flow.
It includes /small-business:plan-payroll for checking cash flow and uncollected invoices, /small-business:month-heads-up for looking 30 days ahead, /small-business:close-month for proceeding with monthly closing, /small-business:price-check for comparing profit margins and prices, /small-business:run-campaign for setting up sales campaigns, /small-business:handle-complaint for dealing with complaints, and /small-business:monday-brief for summarizing weekly status. Connections to QuickBooks, PayPal, HubSpot, Canva, Gmail, Microsoft 365, DocuSign, etc., are expected, and the design includes approval checkpoints for processes touching money or customers.
For solo developers and small SaaS operators, it reduces the "problem of putting off everything except development." Build features with Claude Code, and run sales, inquiries, billing, promotion, contracts, and weekly reviews with Small Business. The effect of having standard workflows is greater for businesses where only the owner understands the situation.
On the other hand, permission design should be cautious given the many connection destinations. Instead of connecting accounting, payments, CRM, and email all at once, start with one or two that are read-centric. Always require previews and approvals for refunds, sending, and customer data updates. Also, the official disclaimer that it does not provide professional financial, tax, legal, or HR advice must be taken as a premise.
If Choosing by Purpose, These Combinations are Strong
If you are building a product alone, the combination of Engineering, Product Management, Design, PDF Viewer, and Small Business is easy to handle. You can decide requirements, implement, check UI, process external documents, and connect to business operations. You don't need to use all of them all the time; you can switch, such as using Engineering and Design during development, and Small Business during sales or operations.
For a development team of several people, Engineering, Enterprise Search, Data, Operations, and Legal are strong. You can find past decisions, verify hypotheses with data, leave change procedures, and identify legal or compliance points early. If you add Product Management, it becomes a single flow from requirements to implementation, verification, and internal sharing.
For product launches or major releases, the four plugins of Product Management, Design, Engineering, and Marketing are effective. By having PRDs, design specs, implementation, and announcements inherit the same premises in order, you can reduce the gap between "what was made" and "what was communicated."
4 Cautions When Installing Official Plugins
First, do not confuse the state of being installed in the desktop version with the installation procedure for Claude Code. In Desktop, you add from the directory; in Claude Code, it's most reliable to register the official marketplace and install the target plugin. While the official repository says the same plugins can be used in both Cowork and Claude Code, the available connectors and execution environments are not necessarily identical.
Second, think of connectors not as "convenient search destinations" but as external integrations with permissions. Do not give write permissions for tasks where read-only is sufficient, and narrow the scope particularly for accounting, payments, email, contracts, and personal information. Separate processes that just create results from those involving sending, updating, or approval, and place human verification on the latter.
Third, don't install too many plugins at once. As skills and commands increase, the judgment of which procedure to choose also increases, and similar functions tend to overlap. Start with 2-3 that directly address your bottlenecks. For example, if reviews take time, choose Engineering; if specs are vague, Product Management; if there's a lot of information searching, Enterprise Search.
Fourth, treat plugin output as "verifiable drafts" rather than "finished products." Check SQL execution results and definitions, try designs in a browser, rehearse runbooks, and have experts check contracts. Even if you expand the range of what you leave to AI, you cannot transfer the responsibility for approval.
Conclusion: The First 3 Should Be Engineering, Product Management, and Enterprise Search
The official plugins selectable from the Anthropic directory in Claude Desktop are not just collections of additional prompts. They bundle job-specific knowledge, reusable procedures, and connections to external tools into a single package, serving as a mechanism to align Claude with how you work. Methods to introduce these into Claude Code are also provided on the official repository.
If I were to choose only three to start, I would recommend Engineering, Product Management, and Enterprise Search. Use Engineering to raise the quality of implementation and operations, Product Management to reduce ambiguity before building, and Enterprise Search to recover the organization's past knowledge. With just these three, Claude Code moves significantly closer from being an "AI that writes code" to a "development platform that connects specs, implementation, judgment, and sharing."
Add Data if you handle data, Design if you are UI-centric, PDF Viewer if you exchange many documents, and Small Business if you are a solo business or small SaaS. The important thing is not to put everything in, but to place the one role that fits the process where you are losing the most time daily.
