Phase 1 / Ep 04: Superpower Engine using-superpowers

⏱ Est. reading time: 4 min Updated on 4/13/2026

The biggest flaws of large models are "arrogance" and "path dependence". When it sees you ask "help me write a user table", even if you just told it the rules via GEMINI.md, it often thinks "I know this, I'll just intuitively output the Prisma schema".

How can we forcibly intercept this kind of self-smart intuition before every interaction? The answer is to establish .agents/skills/using-superpowers/SKILL.md.

1. What is the Superpowers Skill Tree?

In Antigravity, you can inject various Skills into an Agent (e.g., how to troubleshoot systems, how to finalize Git commits). But there is a top-level core skill, known as the "Chief of All Skills", called using-superpowers. This skill doesn't teach coding; it has only one goal:

"Before you say anything or execute any command, even if there's only a 1% chance that another Skill suitable for the current task exists, you must first trigger the action of reading skills, and absolutely not answer directly!"

2. Establish the File

By creating .agents/skills/using-superpowers/SKILL.md in the editor (or by letting the Agent execute it autonomously):

name: using-superpowers
description: Use when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions

<EXTREMELY-IMPORTANT>
If you believe there's even a 1% chance that an existing Skill or Workflow is relevant to the current task, you **ABSOLUTELY MUST** first read that detailed skill guide using the file reading tool!

If a relevant skill exists, you have no room for negotiation; you must invoke and execute it!
</EXTREMELY-IMPORTANT>

### Red Flags
If you have the following thoughts in your mind, it means you need to stop immediately; you are making excuses for your laziness:
- ❌ "This is just a simple question, I know how to handle it." -> An action is a task; you must check the skill list first!
- ❌ "I can quickly use a tool to review git or content." -> The file doesn't tell you the full end-to-end logic; you must first read the corresponding skill system!
- ❌ "This skill requirement is too trivial or overkill." -> Small fixes often lead to major disasters. Use the skill!

3. Effect and Power Generated

Once you configure this meta-level skill file in your project, what wonderful chemical changes will occur?

In the future, if you define a skill called database-migration. When you casually tell the Agent: "Add a field to the User table." Previously, the AI would directly generate the SQL to modify the table. Now, it undergoes an interception chain internally:

  1. The Agent was about to answer -> triggered the using-superpowers mechanism requirement.
  2. The Agent jolted: "Whoa, I need to modify a database table; I should first look for any relevant rules."
  3. The Agent silently invoked the tool to read directory files.
  4. The Agent found database-migration.
  5. The Agent, following the migration specification, said: "I will first run checks, generate the schema, and then ask for your approval."

This is the "Anti-Dumbness Interceptor".

With this pitfall-prevention tool in hand, in the next section, we will unveil the ultimate killer system that changes the AI's context memory limits – planning-with-files, inherited from Manus.