Using Superpowers

Core Concepts

The original presents the following concepts.

1. The 1% Rule: If there's even a slight doubt about whether a skill might be needed, open that skill. The skill catalog is Claude's "task manual" — consult it immediately when doubt arises.

2. Instruction Priority: When instructions from different sources conflict, follow this order:

   • The user's current message
   • Project-specific CLAUDE.md
   • Global ~/.claude/CLAUDE.md
   • Skill body
   • General best practices

3. How to Access Skills: In the Claude Code environment, skills are automatically injected "like tools". Claude calls the relevant skill via the Skill tool at the moment it judges it to be necessary.

4. Red Flags: When certain words/situations appear, a specific skill must be activated.

   Signal	Skill to Activate
   "debug", "bug", "why isn't it working"	investigate / systematic-debugging
   "new feature", "build this", "add this"	brainstorming
   "plan", "steps", "TODO list"	writing-plans
   "review", "PR check"	review
   "deploy", "ship", "open a PR"	ship

5. Skill Types:

   • Rigid: Skills with fixed steps and checklists. Examples: test-driven-development, brainstorming.
   • Flexible: Skills centered on principles and guidelines. Examples: writing-skills, using-superpowers itself.

One-Line Summary

A meta-skill that tells you when and how to invoke the other skills provided by Claude Code. It governs instructions to use skills, priority between skills, Red Flags, and which skill to use in which situation.

Getting Started

/using-superpowers

SKILL.md location: ~/.claude/skills/using-superpowers/SKILL.md

Copy and modify the SKILL.md content if customization is needed.

Practical Example

Scenario: You open Claude Code in the morning to continue working on a Student Club Notice Board assignment. The user says:

"Let's continue the notice board work from yesterday. The Supabase migration keeps failing."

This single sentence contains multiple skill signals. The using-superpowers skill reasons as follows:

1. "continue from yesterday ..." → executing-plans skill candidate. If there was a plan, it should be resumed.
2. "migration keeps failing" → investigate or systematic-debugging skill candidate. A bug/failure signal.
3. "Supabase" + "failing" → Could be a simple config error, a permissions issue, or a network issue. The symptom alone doesn't tell us.

Conclusion: Claude first uses the investigate skill to confirm the root cause, then returns to executing-plans to continue the remaining tasks once resolved. Both skills are needed simultaneously, but order matters.

# Claude's internal reasoning (pseudocode)
if (userMessage.includes("failing") || userMessage.includes("error")) {
  activateSkill("investigate");
}
if (userMessage.includes("continue") && existsPlan("docs/plans/")) {
  queueSkill("executing-plans");
}

Another scenario: The user says "Now make the login UI look nicer." This might look like a "simple styling" task at first glance, but using-superpowers detects a red flag.

• "now ... make the UI nicer" → There are no clear requirements (specifically what? color? layout? mobile?).
• → Run the brainstorming skill first to agree on the intent. Don't change any code before that.

// Without brainstorming, this kind of code tends to appear
// Style changes without clear intent
<button className="bg-blue-500 hover:bg-blue-700 rounded-lg shadow-lg">
  Login
</button>

// After brainstorming to decide "student club tone" + "accessibility AA"
<button className="bg-emerald-600 hover:bg-emerald-700 text-white font-medium px-4 py-2 rounded-md focus-visible:ring-2 focus-visible:ring-emerald-400">
  Login
</button>

Learning Points / Common Pitfalls

• The existence of a meta-skill itself is important: Claude Code has 50+ skills. It is hard to remember every skill at all times. using-superpowers functions as a map of "what to pull out when" for this catalog.
• Practical effect of the 1% rule: In student assignments, the 0.5-second doubt of "is this a situation to use systematic-debugging?" saves 30 minutes later.
• Instruction priority: If the user says "answer in English" but the global CLAUDE.md says "write in Korean", the user's message takes priority. Knowing this priority makes decisions in conflict situations much easier.
• You can also directly tell Claude to "use" a skill: When Claude doesn't automatically pull out a skill, the user can directly specify it like "use the brainstorming skill to agree on the intent first." That authority belongs to the user.
• Next.js 15 tip: Frontend tasks ("create a component", "add a form") almost always work well with the brainstorming → writing-plans → executing-plans chain. Remember this pattern.

Related Resources

• writing-skills — Creating your own skills
• learn — Managing learning content across sessions
• retro — Retrospecting skill usage patterns

---

Field	Value
Source URL	https://docs.anthropic.com/en/docs/claude-code/skills
Author / Source	Anthropic
License	Commentary MIT, original for reference
Translation Date	2026-04-13