Claude brainstorming skill review: Superpowers verdict, safety and alternatives

The Claude brainstorming skill is obra's Superpowers design gate for coding agents. It asks clarifying questions, compares approaches and waits for your approval before the agent writes code. We rate the reviewed v6.4.2 Recommended, 76/100, with safety grade B. Superpowers v7.0.0 rebuilt the skill on 10 October 2026.

Get the skill: brainstorming folder on GitHub · SKILL.md at the reviewed commit · skills.sh listing

What is the Claude brainstorming skill?

The Claude brainstorming skill is obra's Superpowers design gate for coding agents. It asks clarifying questions, proposes approaches and waits for your approval before code. Install it with the Superpowers plugin from the official Claude marketplace.

The brainstorming skill is an Agent Skill, and our primer explains what an Agent Skill is. It sorts each request into a spike, bounded or architectural path. Every path keeps the approval gate, and architectural work ends in an approved, dated spec. It runs in Claude Code, Anthropic's terminal coding agent, and in OpenAI's Codex.

Which brainstorming skill is this page about?

This page reviews the brainstorming skill in obra's Superpowers repository. Six other skills and copies share its name. Examples include gakonst's brainstorm, automagik-dev's genie brainstorm and Anthropic's product-brainstorming. GitHub holds 5,040 SKILL.md files named brainstorming, including forks.

What is the Superpowers plugin?

The Superpowers plugin is obra's bundle of Claude Code skills for disciplined coding. The skills.sh directory lists 15 entries: 14 core skills plus diagnosing-superpowers. A bootstrap skill loads at session start and tells the agent when each skill applies. Top installs at review:

• brainstorming: 381.9K
• systematic-debugging: 277.0K
• writing-plans: 264.4K, which takes the approved spec next
• using-superpowers: 257.1K, the bootstrap
• test-driven-development: 241.3K

Our Superpowers systematic debugging verdict covers a sibling skill from the same plugin.

Is the Superpowers brainstorming skill worth installing?

Yes, it is worth installing for feature work in Claude Code. The Superpowers brainstorming skill scores 76/100, Recommended, at Moderate confidence. Codex has skipped its approval gate, and it can over-ask on small changes.

• Strength: an explicit hard gate plus a three-path router scales ceremony while keeping approval.
• Strength: users report consistently good results, and the maintainer ships fixes within days.
• Weakness: adherence varies by harness and model, with gate and question skips on Codex (#2372, #1989).
• Weakness: the default spec path commits files into docs/ (#1690, #1246).

Tessl, a skill registry, measured a 1.81x lift on Claude Sonnet 4.6.

See how we score skills and other agent skills ranked by verdict.

What changed since our review?

Superpowers v7.0.0 shipped on 10 October 2026. The verdict, score and grade on this page describe v6.4.2.

• SKILL.md: 203 lines added, 263 removed (full diff).
• Builder check: replaces the spec-document reviewer prompt.
• Flow: plays back what it heard, sizes work afterwards, asks which build choices you own (#2463, #2482, #2483).
• Companion: $' rendering fix (#2362).
• Antigravity: native plugin (#2479).
• Marketplace: the official Claude plugin marketplace pins bb92a77, which is v7.0.0.
• Issues: #2372, #1690 and #1957 stayed open on 2026-10-21.

How many tokens does the brainstorming skill use?

The skill file measures 4,387 tokens per load. Our estimate for one feature design session is 60,000 input and 8,000 output tokens.

The estimate assumes 5 sessions a week. The architectural path then chains into writing-plans and often subagent execution, which reviewers flag as costly.

Is the brainstorming skill safe to install?

Yes, the brainstorming skill is safe to install with care, at safety grade B. Its core is instructions only. The optional visual companion runs a local Node server and shell-executes a custom open command (#1957).

Scripts run only when you accept the visual companion. Findings at commit 8ca22db:

• Server: Node built-ins only, bound to 127.0.0.1 by default with a per-session URL key.
• Origin checks: added after #1014 reported cross-origin injection.
• Logo ping: loads from primeradiant.com with the version; disclosed, and SUPERPOWERS_DISABLE_TELEMETRY or DISABLE_TELEMETRY disables it.
• Shell exec: BRAINSTORM_OPEN_CMD passes through cp.exec when set (#1957, open).
• Git writes: specs go to docs/superpowers/specs/ and get committed.
• Session files: /tmp or .superpowers/brainstorm/, with the session key.
• Audits: Gen Agent Trust Hub Pass, Snyk Pass, Socket Warn on skills.sh; Tessl's security check passed.

Our guide to checking a skill's scripts before install applies the same steps to any skill.

What does the community report about the brainstorming skill?

The community reports better-scoped features from the brainstorming skill. A 315-session run saw it trigger 15 of 15 times per model with the bootstrap. Reports note over-asking, heavy token use and gate skips on Codex.

These are the 18 dated reports we weighed, from March to September 2026.

What are the brainstorming skill's known failures?

The brainstorming skill's known failures at v6.4.2 cluster around its approval gate. Codex Desktop skips the gate (#2372), and Codex advances questions early (#1989). Specs commit under docs/ by default (#1690), and small changes get heavy ceremony.

Issue states are as of 2026-10-21.

1. Gate skips on Codex Desktop (#2372, open): with 6.4.1, the agent read the skill and skipped the approval gates.
2. Early questions on Codex (#1989): the agent asked the next question when the user only wanted context.
3. Spec commits under docs/ (#1690 open, #1246): Sphinx published one project's specs on Read the Docs.
4. Ceremony on small work (#2079): a one-paragraph rule change grew to 3,061 lines over about 10 hours.
5. Shell exec of a custom open command (#1957, open): an org security audit flagged it.
6. Host-pane breakage in the visual companion (#2421): open at review.

Our evaluation names the worst failure in these words:

The agent skips the approval gate and implements an unapproved or misunderstood design, or buries a small change under spec and plan ceremony.

Does the brainstorming skill work in Codex, Cursor and other agents?

Yes, the brainstorming skill works in Claude Code and Codex, the two agents we verified. Codex shows approval-gate problems. Cursor, Gemini CLI, OpenCode, Pi, GitHub Copilot CLI, Devin CLI and Hermes Agent are publisher claims.

Claude Code holds the strongest evidence. Our hub lists other Claude Code skills we verified.

How do you install the brainstorming skill?

You install the brainstorming skill by installing the Superpowers plugin, which bundles it. In Claude Code, run /plugin install superpowers@claude-plugins-official. The plugin's session-start bootstrap makes the skill trigger reliably, so avoid copying the folder alone.

Directories show npx skills add obra/superpowers@brainstorming, which skips the bootstrap. Without it, Fable 5.1 triggered the skill 6 of 15 times (#2377). See why a skill fails to trigger. The official marketplace now installs v7.0.0.

How do you install it in Claude Code?

1. Run /plugin install superpowers@claude-plugins-official for the official Claude plugin marketplace.
2. Add the publisher marketplace instead with /plugin marketplace add obra/superpowers-marketplace.
3. Run /plugin install superpowers@superpowers-marketplace after adding that marketplace.

Our full Claude Code skills setup guide covers scopes.

How do you install it in Codex?

1. Open Codex CLI and run /plugins.
2. Search for superpowers.
3. Select Install Plugin.

The Codex app offers the same plugin from its Plugins sidebar.

How do you install it in Cursor, OpenCode and Antigravity?

• Cursor: run /add-plugin superpowers.
• OpenCode: fetch and follow .opencode/INSTALL.md from the repository.
• Antigravity: run agy plugin install https://github.com/obra/superpowers.

These routes are publisher-documented and unverified by us.

How do you use the brainstorming skill in Claude Code?

Use the brainstorming skill by asking Claude Code to build a feature, or by naming the skill. Answer its questions, pick an approach and approve the design. Set where specs are saved before you start.

Steps at v6.4.2:

1. Ask Claude Code for a new feature, or name the brainstorming skill.
2. Answer its clarifying questions, one at a time.
3. Pick one of the 2 or 3 approaches it proposes.
4. Approve the design before any code is written.
5. Approve the written spec on the architectural path.
6. Hand the spec to writing-plans for an implementation plan.

Version 7.0.0 changes steps 2 to 5 with a play-back and a builder check.

Good uses:

• New features
• New projects
• UI design with browser mockups

Poor uses:

• One-line fixes
• Config tweaks
• Repositories that publish docs/ (Sphinx, Read the Docs)

Test it yourself:

1. Ask the agent for a small flag change.
2. Ask for a new todo app in the same repo.
3. Check it picks bounded for the flag and architectural for the app.
4. Check it stops for approval before code.
5. Check it commits the spec where you expect.

What is the brainstorming visual companion?

The brainstorming visual companion is an optional local browser page for mockups and diagrams. It starts a Node server on 127.0.0.1 with a session key, only after you accept. Its guide adds about 3,300 tokens. Version 7.0.0 offers it "whenever seeing would help" (release note).

What is the best Claude skill for brainstorming?

For design before code, the best Claude skill for brainstorming we have reviewed is Superpowers brainstorming, at 76/100. Matt Pocock's grill-me is lighter and unrated. Claude Code's plan mode is built in.

Criterion: design-before-code work in a coding agent, judged by verdict score.

Alexrusin compared Superpowers, grill-me and plan mode on one CRUD API test. Every scored skill appears in the full verdict ranking.

Where is the brainstorming skill on GitHub?

The brainstorming skill lives on GitHub in obra/superpowers under skills/brainstorming, MIT-licensed. We reviewed release v6.4.2, commit 8ca22db, on 30 September 2026. The current release is v7.0.0, and the repository has over 300,000 stars.

• Stars: 293,396 on 2026-09-30, 302,529 on 2026-10-21
• Releases: 14 in 12 months
• Contributors: 37 (agenticskills.io shows 6 and "v1.7.0")

Install from obra's repository or the official marketplace, since GitHub holds 5,040 SKILL.md files named brainstorming.

Common questions

Does Claude have a brainstorm mode?

No, Claude has no built-in brainstorm mode. Claude Code has plan mode, and posts about "brainstorming mode" mean this skill.

Is Claude good for general brainstorming?

Yes, Claude itself handles open idea work without this skill, which targets coding design. See our skills for the Claude app.

Is the brainstorming skill made by Anthropic?

No, obra (Jesse Vincent) makes it. Anthropic's official Claude plugin marketplace distributes it.

Where does the brainstorming skill save the spec?

It saves specs under docs/superpowers/specs/ and commits them to git by default. It honours a location you give it, so set one first (#1690).

How do you stop it on small tasks?

Say the change is small, or ask it to skip the questions. Version 7.0.0 offers a skip on its first message (release note).

How long does a brainstorm take?

One HN user types requirements for 10 to 30 minutes first (HN 49856911). No measured session length exists.