The biggest flaws of Large Language Models are "arrogance" and "path dependency". When you ask it to "help me write a user table," even if you just set the rules in GEMINI.md, it often feels "I know this, I'll just output a Prisma schema based on my intuition."
How can we force an interception of this "clever" intuition before every interaction?
The answer lies in establishing .agents/skills/using-superpowers/SKILL.md.
1. What is the Superpowers Skill Tree?
In Antigravity, you can inject various Skills into the Agent (e.g., how to debug a system, how to wrap up a Git commit).
But there is a top-level core skill, known as the "God of all Skills", called using-superpowers. This skill doesn't teach coding; it has only one objective:
"Before you say anything or execute any command, even if there is only a 1% chance that another Skill suitable for the current task exists, you MUST trigger the action to read that skill. NEVER answer directly based on intuition!"
2. Creating the File
Create .agents/skills/using-superpowers/SKILL.md in your editor (or let the Agent execute it autonomously):
---
description: Use when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions
---
# using-superpowers
You are equipped with a set of "Superpowers" (Skills) located in the `.agents/skills/` directory.
Before answering the user, you MUST run the `list_dir` tool to inspect available skills. If a skill matches the user's request, use the `view_file` tool to read the skill's instructions BEFORE writing any code.
3. The Shift in Agent Behavior
Once this file is established, your Agent will no longer rush into coding. It will pause, scan its toolkit, and operate based on the precise procedures you've defined, vastly reducing errors and hallucinations in complex architectural tasks.