Claude Code Superpowers Plugin — What, Why, Install/Uninstall, Usage, Examples

Claude Code writes code well out of the box, but it often skips planning, tests, and verification and jumps straight into implementation. Superpowers is a Claude Code plugin that pushes back against that habit and enforces a systematic development workflow — brainstorming → design → implementation plan → TDD → review — via composable skills.

The official repo is obra/superpowers. You can install it from Anthropic’s official marketplace or the author’s marketplace (obra/superpowers-marketplace). It also works with Cursor, Codex, OpenCode, and other agents; this post focuses on Claude Code.

Environment: Claude Code (build with /plugin marketplace support) / internet connection

---

[01] What Plugin Is This

Superpowers is an agentic software development methodology plus a library of composable skills, built by Jesse Vincent / Prime Radiant.

Once installed, Claude Code gets:

Piece	What you get
Skills	brainstorming, TDD, systematic-debugging, writing-plans, and more
Commands	/superpowers:brainstorm, write-plan, execute-plan, etc.
Agents	Preconfigured subagents such as code-reviewer
Hooks	Session-start bootstrap and context-aware skill activation

The important part: these are closer to mandatory workflows than soft suggestions. Before a task, the agent checks for a matching skill and follows it when one applies.

1-1. The Basic Workflow

1. brainstorming — Before code: refine requirements with questions, present design in digestible sections
2. using-git-worktrees — After design approval: isolated branch/worktree and clean test baseline
3. writing-plans — Bite-sized tasks (2–5 minutes each) with file paths and verification steps
4. subagent-driven-development / executing-plans — Per-task subagents or batched execution with checkpoints
5. test-driven-development — Enforce RED → GREEN → REFACTOR
6. requesting-code-review — Review against the plan; Critical issues block progress
7. finishing-a-development-branch — Verify tests, then choose merge / PR / keep / discard

---

[02] Why You Need It

Common Claude Code failure modes:

• Starts editing files without agreeing on design or scope
• Misses files during migrations / large refactors
• Claims “it should work” without running verification
• Debugs with guess-and-patch instead of root-cause investigation

Superpowers inserts skills at those moments so skipping steps is hard.

Request only (stock Claude Code)
  "Add per-user API rate limiting"
  → Immediately edits middleware
  → Easy to miss tests, edge cases, auth interaction

With Superpowers
  → brainstorming: limits, storage, response shape, existing stack
  → writing-plans: per-file tasks + verification commands
  → TDD: failing test first
  → verification-before-completion: evidence before "done"

The philosophy in four lines:

• Test-Driven Development — tests first
• Systematic over ad-hoc — process over guessing
• Complexity reduction — YAGNI, simplicity first
• Evidence over claims — verify before declaring success

---

[03] Installation

Install with slash commands inside a Claude Code session. It is not an npm package.

3-1. Official Marketplace (simplest)

/plugin install superpowers@claude-plugins-official

Also listed on the official Claude plugins page.

3-2. Superpowers Marketplace (author; includes related plugins)

/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace

Pick one path. The core plugin is the same.

3-3. Install via UI

/plugin

Open Discover, search for superpowers, and install.

3-4. Verify

After install, start a new session (skills load at session start).

/plugin list

Or ask:

Do you have superpowers?

If the agent confirms Superpowers and lists skills, you’re good. You may also see a session-start hook message such as SessionStart:startup hook succeeded.

3-5. Update

/plugin update superpowers

---

[04] Uninstall

Remove the plugin:

/plugin uninstall superpowers

Also remove the author marketplace if you added it:

/plugin marketplace remove obra/superpowers-marketplace

claude-plugins-official is Claude Code’s built-in marketplace — for Superpowers alone, uninstall is enough.

---

[05] How to Use

5-1. Automatic Activation (default)

No extra config: matching skills auto-trigger from what you ask.

You say something like…	Skill that often activates
“Design a new auth feature”	brainstorming
“This function crashes intermittently”	systematic-debugging
“Review my recent changes”	requesting-code-review
Implementation work	test-driven-development
“We’re done” / success claims	verification-before-completion

Force a skill explicitly:

use brainstorming skill
use systematic-debugging skill

5-2. Slash Commands (explicit workflow)

For complex features and migrations, calling commands directly is more reliable.

Command	Role
/superpowers:brainstorm	Turn ideas into a design (before code)
/superpowers:write-plan	Write an implementation plan (short tasks)
/superpowers:execute-plan	Run the plan in batches with review checkpoints

Recommended order:

1. /superpowers:brainstorm
2. Approve the design
3. /superpowers:write-plan
4. Review the plan
5. /superpowers:execute-plan

5-3. Skills Library Snapshot

Testing / Debugging

• test-driven-development — RED-GREEN-REFACTOR
• systematic-debugging — 4-phase root-cause process
• verification-before-completion — no “done” without evidence

Collaboration

• brainstorming, writing-plans, executing-plans
• subagent-driven-development, dispatching-parallel-agents
• requesting-code-review, receiving-code-review
• using-git-worktrees, finishing-a-development-branch

Meta

• using-superpowers — intro to the skills system
• writing-skills — create custom skills

---

[06] Usage Examples

6-1. New Feature — brainstorm → plan → execute

/superpowers:brainstorm
I want per-user API rate limiting on the REST API.
Prefer Redis; return 429 when exceeded.

Typical flow:

1. Asks about limits, key design, auth interaction, exceptions
2. Presents design in short sections → you approve
3. /superpowers:write-plan produces file-level tasks + test/verify commands
4. /superpowers:execute-plan runs tasks; stops on Critical review findings

6-2. Bug — Systematic Debugging

The payment webhook handler times out sometimes.
use systematic-debugging skill

Instead of blindly raising timeouts, the agent follows reproduce → logs/traces → hypotheses → minimal fix → verification.

6-3. Large Change — write-plan First

/superpowers:write-plan
To enable cacheComponents in Next.js, list the files to change,
phase the work, and include verification commands per phase.

You get something closer to a real roadmap: paths, before/after patterns, success criteria, rollback — not a vague outline. Plans that live as files also survive session cuts.

6-4. Verify Before “Done”

Rate limiting is implemented. Can we call it done?

With verification-before-completion, the agent should not declare success until it has actually run the relevant tests/build/manual checks and shown the output.

---

[07] Troubleshooting

Symptom	What to try
/plugin not recognized	npm update -g @anthropic-ai/claude-code, restart
Marketplace add fails	Check network; verify obra/superpowers-marketplace; remove → re-add
Install fails	Confirm marketplace is registered; use full name superpowers@...
Skills missing	Start a fresh session, /plugin list, /plugin update superpowers
Auto-skills don’t fire	Call explicitly: "use brainstorming skill"

Optional telemetry opt-out:

export SUPERPOWERS_DISABLE_TELEMETRY=1
# Also honors Claude Code DISABLE_TELEMETRY /
# CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC

---

[08] Summary

Item	Detail
What	Structured skills/workflow plugin for Claude Code
Why	Steer “code first” habits toward design, TDD, and verification
Install	/plugin install superpowers@claude-plugins-official
Uninstall	/plugin uninstall superpowers
Core usage	brainstorm → write-plan → execute-plan, or auto skills
Refs	GitHub · Install docs

Overkill for one-line fixes. It shines on new features, migrations, mysterious bugs, and multi-file refactors — the work where coding first tends to cause accidents.