Agent Skills Repository

This repository serves as a centralized, continuously updated collection of custom skills for agentic AI coding assistants.

These skills define strict behavioral guidelines, workflows, and constraints to ensure the AI operates reliably, safely, and systematically—especially for complex software engineering and machine learning research tasks.

Structure

The skills are located in the .agents/skills/ directory. Each skill has its own folder containing a SKILL.md file.

.agents/
└── skills/
    ├── base-hygiene/
    │   └── SKILL.md
    ├── code-review-research/
    │   └── SKILL.md
    ├── honest-eval/
    │   └── SKILL.md
    ├── paper-fidelity/
    │   └── SKILL.md
    └── training-diagnostics/
        └── SKILL.md

How Skills Work

Skills are loaded into the agent's context dynamically. To ensure token efficiency, the system relies on YAML frontmatter in each SKILL.md file:

1. Automatic Discovery: The agent scans the .agents/skills/ directory on startup.
2. Trigger-Matching: The agent reads only the name and description from the YAML frontmatter.
3. On-Demand Loading: If the agent determines that a skill's description matches the current task, it will dynamically load and read the full markdown body of the skill and strictly adhere to its instructions.

Creating a New Skill

💡 Tip: Automate Skill Creation You can use the provided antigravity-skill-creator.md prompt to instruct any AI agent to generate a perfectly formatted skill for you. Just feed it the instructions in that file and ask it to build your desired skill.

When manually adding a new skill to this repository, follow these rules:

1. Create a new folder under .agents/skills/ (e.g., my-new-skill/).
2. Inside that folder, create a SKILL.md file.
3. The file must begin with YAML frontmatter containing the name and description. The description is critical—it acts as the trigger for the agent to load the skill.

Example SKILL.md Format:

---
name: my-new-skill
description: Use this skill when [specific trigger condition]. It helps the agent do [specific task].
---

# My New Skill

[Actionable instructions, constraints, and step-by-step processes go here...]

Note: Keep the markdown body actionable and under 500 lines. If a skill requires extensive context, place it in a references/ subdirectory and link to it.

Future Updates

This is a living repository. New skills will be added time and again to improve agent workflows, enforce better coding hygiene, and systematize complex research tasks.