MLOps Coding Skills

Welcome to the MLOps Coding Skills repository! 🚀

This project contains a collection of Agent Skills designed to accompany the MLOps Coding Course. These skills are tailored to help AI coding assistants (like Antigravity, Gemini CLI, GitHub Copilot, and Claude Code) understand and guide you through the best practices of MLOps, from prototyping to production.

🧠 What are Agent Skills?

Agent Skills are specialized context or instructions that you can load into your AI assistant. Each skill is a folder containing a SKILL.md file (YAML frontmatter + Markdown instructions) that provides the "Know-How" for a specific task, ensuring the AI follows the course's methodology, standards, and patterns.

📚 Skills in this repository

Each skill derives from one chapter of the MLOps Coding Course:

Skill	Course chapter	Focus
mlops-initialization	1. Initializing	Set up a robust project structure with uv, git, and VS Code.
mlops-prototyping	2. Prototyping	Write clean, reproducible Jupyter notebooks and prevent data leakage.
mlops-industrialization	3. Productionizing	Turn prototypes into a distributable package (src layout, hybrid paradigm).
mlops-validation	4. Validating	Add typing (ty), linting (ruff), testing (pytest), and logging.
mlops-automation	5. Refining	Automate tasks, containers, CI/CD, and experiment tracking (MLflow).
mlops-collaboration	6. Sharing	Prepare the project for sharing: license, docs, templates, releases.
mlops-observability	7. Observability	Add reproducibility, lineage, monitoring, alerting, and explainability.

🛠️ Installation & Usage

Skills follow the open SKILL.md standard, so most agents discover them from a well-known folder. The workspace-standard location is .agents/skills/; each tool then reads from its own path (see the table below).

Recommended setup (symlinks)

1. Clone this repository to a location on your machine (e.g. ~/mlops-coding-skills).

2. Create the skills directory in your target project or workspace, then symlink the skills you need:

   mkdir -p .agents/skills
   # Symlink every MLOps skill (or pick individual folders):
   ln -s ~/mlops-coding-skills/mlops-* .agents/skills/

3. Expose them to your agent using the per-tool path below.

Per-tool install paths

Tool	Skills directory	Notes
Antigravity	.agents/skills/ (workspace)	Discovered natively; no extra step once symlinked.
Gemini CLI	.gemini/skills/ (workspace) or ~/.gemini/skills/	Install via gemini skill install <git-url> --scope=workspace.
Claude Code	.claude/skills/ (project) or ~/.claude/skills/	Project skills live in .claude/skills/; personal skills in ~/.claude/skills/.
Claude Code (plugins)	<plugin>/skills/	Skills bundled in a plugin are loaded automatically when the plugin is installed.
GitHub Copilot	any workspace path	Open the relevant SKILL.md to load it into context (Agent Mode).

Antigravity

Antigravity natively supports the .agents/skills standard. Once the symlinks from the Recommended setup are in place, Antigravity detects and uses these skills automatically.

See the Antigravity Skills docs for details.

Gemini CLI

The Gemini CLI manages skills in .gemini/skills. Install this repository as a remote skill set to keep it updated and scoped correctly:

gemini skill install https://github.com/MLOps-Courses/mlops-coding-skills --scope=workspace

See the Gemini CLI Skills docs for details.

Claude Code

Claude Code discovers skills natively — no manual context loading required:

• Personal skills (all your projects): place or symlink the folders under ~/.claude/skills/, e.g. ln -s ~/mlops-coding-skills/mlops-* ~/.claude/skills/.
• Project skills (this workspace only): Claude Code reads workspace skills from .claude/skills/. If you keep them under the standard .agents/skills/, link it once: mkdir -p .claude && ln -s ../.agents/skills .claude/skills.
• Plugin skills: any skill bundled inside an installed plugin's skills/ directory is loaded automatically.

Claude Code then loads a skill on demand whenever its description matches your task.

(For Claude Desktop, add the relevant SKILL.md files to a Project's Knowledge instead.)

See the Claude Skills documentation for details.

GitHub Copilot

For GitHub Copilot (Agent Mode, e.g. in VS Code), reference the skill files explicitly:

1. Open your project in VS Code.
2. Open the relevant SKILL.md file (e.g. .agents/skills/mlops-observability/SKILL.md) to load it into context.
3. Ask Copilot to follow the instructions in the open skill file.

See GitHub Copilot Agent Skills for the latest features.

🤝 Contributing

We welcome contributions! If you have a new MLOps pattern or improvement, please open a Pull Request. See CONTRIBUTING.md and AGENTS.md for the repository conventions and local checks (mise run format, mise run check).

Happy Coding! 🤖✨