Tool-agnostic Agent Skills for LLM-based runtimes (Claude, Gemini, Antigravity)

This repository contains LLM Agent Skills: reusable, structured instruction modules designed to be loaded by LLM agent runtimes. Skills are model-agnostic, tool-agnostic, and portable across environments such as Claude Code and Gemini/Antigravity.

Supported Universally

These skills are compatible with:

• Antigravity
• Claude Code
• OpenAI Swarm
• LangChain / LangGraph agents
• Custom in-house agent frameworks

Architecture

graph TD
    A[MODEL Claude, GPT, etc.] <-- raw LLM --> B(INSTRUCTION BUNDLES Skills, prompts, policies)
    B <-- not standardized --> C[RUNTIME / HOST UI, CLI, agent system]
    C --> D[Claude Code]
    C --> E[Antigravity]

Loading

Note: This diagram illustrates the role of skills as instruction bundles bridge between runtimes and models.

┌─────────────────────────────┐
│  MODEL (Claude, GPT, etc.)  │  ← raw LLM
└─────────────────────────────┘
            ▲
┌─────────────────────────────┐
│  INSTRUCTION BUNDLES        │  ← "Skills", prompts, policies
│  (NOT standardized)        │
└─────────────────────────────┘
            ▲
┌─────────────────────────────┐
│  RUNTIME / HOST             │  ← Claude.ai, Claude Code,
│  (UI, CLI, agent system)    │     Antigravity, IDEs, etc.
└─────────────────────────────┘ 

What is a Skill?

A skill is a self-contained package of instructions and logic defined in a SKILL.md file. By adhering to the standard, these skills allow any compatible agent to perform specialized, complex tasks without vendor lock-in.

Available Skills

Skill Name	Description
Co-Dialectic	Bidirectional human-AI co-evolution framework. Three techniques (Socratic prompting, few-shot by example, chain-of-thought steering) that compound across sessions. Inspired by Ethan Mollick's Co-Intelligence.
Crypto Investor L2 Scoreboard	Researches and scores L2 blockchains based on adoption and momentum.
LinkedIn Post Manager	Manages a LinkedIn post queue with browser automation and macOS desktop notifications. Human-in-the-loop before posting.

Installation

Use the provided agent_skills_install.py script to install skills.

Note

Standardization Note: The installation paths and methods documented here reflect the current conventions for Antigravity, Gemini, and Claude Code. As the Agent Skill Standard evolves, these locations may change.

Standard Install (Recommended)

This command installs all skills defined in skills.yaml to all detected standard agent locations (Antigravity, Claude Code, Gemini). It uses a copy/overwrite method by default to ensure maximum compatibility.

python3 agent_skills_install.py

Advanced Options

# Install only for a specific runtime
python3 agent_skills_install.py --claude-code
python3 agent_skills_install.py --antigravity

# Install to a custom directory
python3 agent_skills_install.py --target /path/to/custom/folder

# Use symlinks instead of copying (useful for skill development)
# Note: Some environments (like Claude Code) may not support symlinks.
python3 agent_skills_install.py --symlink

Testing & Verification

After installation, you must verify that the agent has loaded the new skills.

Claude Code

1. Restart Claude Code (Critical: Skills are often only loaded on startup).
2. Use "Code" Mode: You MUST use the Code tab in Claude Desktop (not Chat or Cowork) to test skills.
3. Test with a trigger phrase: Try saying "Create a crypto investor scoreboard".
4. Check if recognized: The skill should activate based on the description trigger.

Antigravity

1. Restart the Agent Session: Reload the window or restart the agent process to ensure the skills cache is rebuilt.
2. Verify Skill Availability: Ask "What skills do you have available?" or check the agent's startup logs.
3. Test Execution: Run a command like "Run the crypto L2 scoreboard".

Troubleshooting

• macOS Permissions: If the agent fails to load the skill, it might be due to macOS "Quarantine" attributes on downloaded files. The installation script now automatically attempts to fix this by running xattr -cr. If issues persist, you can manually run:

  xattr -cr ~/.claude/skills/crypto-investor-l2-scoreboard

Contributing

We welcome contributions! Please ensure your skills follow the Specification.

1. Create a new folder in skills/.
2. Create a SKILL.md with YAML frontmatter + instructions.
3. (Optional) Add a README.md for human readers.
4. Submit a Pull Request.