create_skill

Create Skill

Use this skill to define a new capability for Antigravity. It enforces a professional structure inspired by industry best practices (conciseness, progressive disclosure, and modularity).

Inputs

• Skill Name: The name of the skill (snake_case).
• Description: A short, action-oriented summary of what the skill does.
• Instructions: The core logic or checklist for the skill.

Tooling Strategy

• Use write_to_file to create SKILL.md and related files.
• Use run_command to verify directory creation if not using write_to_file's auto-create.

Workflow

1. Design & Strategy

Before creating files, ask the user (or deduce from context):

• Trigger: What specifically should trigger this skill? (This goes into the description).
• Complexity:
  • Simple: Just a SKILL.md checklist.
  • Complex: Needs scripts/ (for tools) or references/ (for docs).

2. Create Skill Structure

Create the directory at c:/Users/ADMIN/.gemini/antigravity/skills/[skill_name]/. Based on Complexity, create subfolders if needed:

• [skill_name]/scripts/ (Executable code)
• [skill_name]/references/ (Documentation)
• [skill_name]/assets/ (Templates/Images)

3. Write SKILL.md (The "Concise" Rule)

• Frontmatter:
  • name: snake_case_name
  • description: CRITICAL. This is what the AI router reads.
• Body:
  • Progressive Disclosure: Keep it under 500 words. Refactor long context into references/.
  • Structure:

    ---
    name: [skill_name]
    description: [description]
    ---
    
    # [Human Readable Title]
    
    [Instructions, prompts, and steps provided by the user]
    

4. Register Slash Command (Workflow)

Create the workflow trigger in both locations:

1. c:/Users/ADMIN/.gemini/antigravity/skills/.agent/workflows/[skill_name].md
2. C:\Users\ADMIN\.gemini\antigravity\global_workflows\[skill_name].md

• Content:

  ---
  description: [Short, action-oriented summary]
  ---
  1. Run the [skill_name] skill.
  

5. Update README.md

• Target: c:/Users/ADMIN/.gemini/antigravity/skills/README.md
• Action: Add the new skill to the ## 🛠️ Available Skills section.
• Logic:
  • Determine the appropriate category (header) for the skill (e.g., Development, Documentation, Strategy).
  • If the category exists, append the skill to it.
  • If not, create a new category header ### [Category Name] and add the skill there.
  • Format: * [**[Human Name]**](./[skill_name]/SKILL.md) (/[skill_name]) - [Short description]

6. Final Confirmation

Confirm to the user:

• Structure created in skills directory.
• Slash command /[skill_name] ready in both local and global workflow folders.
• Documentation file README.md updated.

Skill Design Principles (Advanced)

When generating instructions for SKILL.md, enforce these agentic design patterns:

• Single Responsibility: A skill should do exactly one thing exceptionally well. If it's doing two things, it should be two skills.
• Observable Actions: Favor instructions that produce tangible output (files, git commits, diagrams) rather than just returning text in the chat.
• Fail Gracefully: Explicitly instruct the skill to stop and ask the user for clarification if required context, files, or permissions are missing, rather than guessing.
• Safety Constraints: If the skill involves executing code, deleting files, or making network requests, include a clear "Safety & Verification" step.