Claude plugins and Superpowers: making Claude follow a fixed method

Ask Claude Code to "build an app" and it starts building. With the Superpowers plugin, something else happens first: Claude asks questions, one at a time, lays out two or three approaches, writes a design and waits for your approval. Only then comes a plan, and the first code it writes is a failing test. That difference in approach is what plugins are for. They exist for Claude Code, and since January 2026 also for office work such as contracts, bookkeeping and customer support.

As of September 2026. This article first appeared in February 2026 and has been updated: the way you install Superpowers has changed, and Cowork has since merged into the regular Claude app.

What is a Claude plugin?

A plugin is a package you install in one go that equips Claude for a particular kind of work. It can contain four kinds of components:

• Skills: domain knowledge and working methods that Claude applies automatically when a task calls for them.
• Slash commands: actions you invoke yourself, such as /review-contract.
• Connectors: links to external tools such as Slack, Jira or Microsoft 365, via the Model Context Protocol (MCP).
• Subagents: specialised helper agents that run a subtask in their own context.

In Claude Code a plugin can also include hooks: scripts that always run at a fixed moment, for example after every edit.

The advantage over a long system prompt: you don't write the instructions yourself or repeat them in every conversation. And because plugins are plain text files, you can read what they tell Claude before you install them.

Superpowers: a development method for Claude Code

Superpowers is a plugin by Jesse Vincent (Prime Radiant), open source under the MIT licence, with more than 290,000 stars on GitHub. It doesn't teach Claude Code a new programming language but a way of working. The skills that stand out:

Brainstorming before there is code

The brainstorming skill forbids Claude from writing code or scaffolding a project before you have approved a design. Claude first looks at your project, asks questions one at a time, proposes two or three approaches with their trade-offs, and records the approved design in a document. The skill is explicit that this also applies to projects that seem "too simple": that is exactly where unexamined assumptions waste the most work.

Planning in small steps

With writing-plans the design is broken into tasks of two to five minutes. That makes the plan checkable: you can see what will happen at each step.

Test-driven development

The test-driven-development skill enforces the red-green-refactor cycle: first a failing test, then the minimum code to make it pass, then clean-up. The idea: it isn't Claude who says something works, a test shows it.

Subagents with review

In subagent-driven-development, subagents carry out the tasks from the plan, with a review in between before the work counts as done.

Systematic debugging

When something breaks, systematic-debugging makes Claude follow four phases: investigate the root cause, analyse patterns, form and test a hypothesis, and only then fix. Proposing a fix before phase one is complete is not allowed. If three fixes fail, Claude has to question the architecture instead of making a fourth attempt.

That takes more time per task than "just build it". For a typo it's overkill. The method is meant for work where a wrong assumption is expensive: a feature that touches several files, or a bug that has already been "fixed" twice.

Installing Superpowers

Superpowers is in Claude Code's official plugin marketplace. In a Claude Code session, type:

/plugin install superpowers@claude-plugins-official

You don't need to type anything special afterwards. A hook that runs at the start of every session loads the framework, and Claude checks before each task which skill applies.

Plugins for work other than code

On 30 January 2026 Anthropic brought plugins to Cowork, the environment in the Claude app where Claude carries out tasks on its own with your files and tools. Anthropic released eleven of them as open source, covering productivity, sales, marketing, data, product management, biology research and a plugin for building your own plugins, among others. Four examples:

• Legal: reviewing contracts with /review-contract, triaging NDAs, compliance questions and risk assessments.
• Finance: preparing journal entries, reconciling accounts, generating financial statements, analysing variances and managing the month-end close.
• Customer Support: triaging tickets, drafting replies, packaging escalations and turning resolved issues into knowledge base articles.
• Enterprise Search: asking one question across all your company tools at once, such as email, chat, documents and wikis. Think of a question like "what are customers complaining about this week?".

In February 2026 plugins for HR and other functions followed, plus connectors for Google Workspace, DocuSign and WordPress.

A plugin doesn't turn Claude into a lawyer or a controller. It provides a fixed method and the links to your systems. The output is still work that someone who knows the subject needs to check.

Where to find them

Since 16 September 2026, Cowork and regular chat have been merged into one Claude app; what Cowork could do is now available from any conversation. The merge reaches Pro and Max first, then Team and Free. You install plugins via claude.com/plugins. For Claude Code the same plugins live in their own marketplace:

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

Read a plugin before you install it

A plugin determines how Claude works: which steps it makes mandatory, which tools it calls, sometimes which scripts run automatically. Because it is all text files, checking takes little effort. Open the SKILL.md files on GitHub and read what they say. With Superpowers you'll see exactly why Claude suddenly asks questions instead of building, and you can decide whether you want that for your project.