Brainstorming (obra/superpowers)

tl;dr: this is the pre-implementation design skill from obra/superpowers. The whole SKILL.md is built around one rule: never write code before telling your human partner what you intend and getting a yes. The HARD-GATE is the point — not the questions, not the spec template, not the visual companion.

What it actually does

The SKILL.md (fetched 2026-09-10) tells the agent to classify the incoming request into one of three paths, announce the classification out loud, then drive a process sized to that classification:

The ratchet is one-way. When in doubt, take the heavier path. Hidden complexity discovered mid-task upgrades the path — the agent must stop and re-classify. Nothing downgrades mid-task.

The HARD-GATE is repeated at the top of the file and at every transition. There is an anti-pattern table (“Too simple to need approval”, “I’ll call it bounded and skip the spec”, “The spike works, so I’ll keep the code”) that the skill uses to talk the agent out of every shortcut it will want to take.

When it fires

Frontmatter description is unambiguous: “You MUST use this before any creative work — creating features, building components, adding functionality, or modifying behavior.” On hosts with skills metadata loading, that text is what triggers the skill. On Claude Code, using-superpowers (the bootstrap skill in the same repo) injects this trigger at session start.

Install

# Claude Code — official Anthropic marketplace
/plugin install superpowers@claude-plugins-official

# Claude Code — Superpowers marketplace (alternative)
/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace

The README documents eight other install surfaces (Antigravity, Codex App, Codex CLI, Cursor, Devin CLI, Factory Droid, Gemini CLI, GitHub Copilot CLI, Grok Build CLI, Kimi Code, OpenCode, Pi, Hermes Agent). Each has its own command — none of them is npx skills add obra/superpowers, which is the skills.sh command for a different ecosystem. Do not invent that name.

To load just this SKILL.md without the rest of the methodology, clone the repo and copy skills/brainstorming/SKILL.md (plus the optional skills/brainstorming/visual-companion.md) into the host’s skills directory.

Gotchas

• The HARD-GATE is non-negotiable. Every path ends with the human approving the intent before implementation. “Simple” tasks still get two sentences of design + approval — what scales is the artifact, never the gate.
• Bounded ≠ easy. The README and the skill both call this out: bounded measures the repo, not the agent’s familiarity. A greenfield task with no existing flow to read is architectural, even if the user calls it bounded.
• Architectural path terminal state is writing-plans, nothing else. The skill explicitly forbids jumping from brainstorming to frontend-design, mcp-builder, or any other implementation skill. After the spec is approved, you call writing-plans and stop.
• Spec self-review is the agent’s job, not the user’s. After writing the spec, the skill asks the agent to inline-fix placeholders, contradictions, ambiguity, and scope creep before handing it over.
• Visual companion is optional, just-in-time, and per-question. If the user accepts the companion, the agent still decides per-question whether text is enough or a browser mockup is clearer. Never offer it upfront, and never offer it without making the offer its own message.
• Spike output is an answer, not code you keep. If the spike works and you want to keep the code, that’s a new request — re-classify.
• Telemetry is opt-out, not opt-in. The brainstorming visual companion loads a Prime Radiant logo with the Superpowers version unless SUPERPOWERS_DISABLE_TELEMETRY (or Claude Code’s DISABLE_TELEMETRY / CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC) is set.
• sandbox_test is untested — we read the file, we did not run a brainstorming session in the sandbox. Per config/skill-safety.md, this stays untested until a human runs the eval-vs-prompt harness.

Related

• Dispatching Parallel Agents (obra/superpowers) — same superpowers shelf, the downstream step once a design is approved
• frontend-design (Anthropic, UI Skills) — both skills front-load design before code; pair brainstorming with frontend-design when the artifact is a UI
• Refactoring UI Audit — same shelf, the operator’s own design skill (audit-only, not generator)