How to use the Superior Thinking Skill agent skill

What does the Superior skill make the agent do?

Superior holds the task fixed and replaces the mind solving it. The agent first states the scope it thinks you mean, in one line with its reason, for example: Scope: the auth flow end to end, from a question about one login handler, because the decision lives at that level. It then imagines how an intelligence that knows the field, sees the real goal, invents no limits and is not anchored to our first idea would approach the same task.

The agent compares that approach with its own plan. Each gap becomes either an executable change to the plan or a named real limit, and any constraint you actually set is kept. The skill is built for big problems. On a single-function question the two approaches usually match, so the pass gains nothing.

• Scope line: states the inferred scope, what you asked, and why.
• Superior approach: the answer a far more knowledgeable, unconstrained mind would give.
• Gap table: each difference is classed as knowledge, convention, or invented constraint.
• Output: a changed plan, with any real limits kept as named constraints.

Install the Superior skill

Download the Superior ZIP from this page and unzip it. For a personal install, place the Superior folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder. Alternatively, in the Claude apps, go to Settings → Capabilities → Skills and upload the skill there.

The skill's main file is SKILL.md, and it includes an Intents/Elevate.md file that holds the working procedure. Keep both files together inside the Superior folder.

• Personal: ~/.claude/skills/Superior/
• Project: .claude/skills/Superior/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload

How do I trigger Superior with natural requests?

You can invoke Superior by name, or by asking the kinds of questions the skill is written to match. The skill's trigger phrases include: superior, how would a smarter intelligence do this, what would a genius or superintelligence do, if you knew everything, are we thinking too small, best possible approach, how should this whole system be built, and analyze how this big thing is implemented.

Name the scope when you can. If you say which system, feature or decision you mean, the agent can state the scope quickly. If your request is unclear or too small, the agent asks a short interview first. If you have already fixed the scope, say so, and the agent will run it as given without asking.

• "Superior on how the whole billing system should be built."
• "Are we thinking too small about this data pipeline? Run Superior."
• "Scope is fixed: the notification subsystem only. What would a far more capable mind do?"

Examples of Superior in practice

In a design case, you ask for the sync layer of a notes app to be designed with Superior. The agent maps the app, server API and clients first, then gives a fresh pass the goal and constraints without your draft. If its answer differs from your planned server-authoritative design, the gap table lists the reasons, such as knowledge of mature libraries or a convention that was never required. Your changed plan keeps any real limit, such as an existing server API, as a migration path.

In a too-small case, you ask Superior about a single retry helper that has callers in two places. The agent sees two plausible scopes and asks which one you mean before it runs. That is the intended behaviour: the skill refuses to run on a scope that is too small to be worth it.

What does Superior need, and what does it run?

Superior has no listed requirements. It does not run scripts and does not use the network, according to the skill's own metadata. Its files are SKILL.md and Intents/Elevate.md. Your project's user customizations, if you have any, are read from a LIFEOS customizations folder under your Claude config directory; if that folder does not exist, the skill uses its defaults.

• Requirements: none listed.
• Scripts: none.
• Network: none.
• Customizations: optional, read from <configRoot>/LIFEOS/USER/CUSTOMIZATIONS/SKILLS/Superior/.

Safety notes and tips

Because Superior runs no scripts and uses no network access, the main risks are about the quality of the answer rather than the system. Its own gotchas flag two failures. The first is anchoring: if the pass runs after a plan already exists, the superior answer can come back as your plan with stronger wording. The second is scope: running on the literal name you gave can produce an empty gap table, while jumping to the whole system maps something you never asked about.

Give the agent the sources behind any figure you include in the brief, not just the figures. In one recorded run, a blind agent found a baseline was overstated because the tool reporting it was new. Checking inputs this way catches errors that the reasoning would otherwise carry forward.

• Avoid running Superior on a plan you have already written if you want a genuinely independent approach.
• Provide sources for every number in your brief.
• Pair with FirstPrinciples to test whether a constraint Superior flags was really invented.