Cascade · Claude Code · Cursor · VS Code · any IDE

Reusable AI coding skills
by M2Lab.io

A curated collection of methodologies, not project rules — eval-driven development, debugging discipline, PR hygiene, spec-before-code — packaged so any AI assistant can apply them consistently across every project.

Get started View on GitHub

Loading skills into a new project

Pick the option that fits how the project is run.

1. Symlink recommended

Single dev machine. Edits propagate instantly.

git clone git@github.com:M2LabOrg/agent-skills.git ~/dev/M2LabOrg-skills

mkdir -p <project>/.claude/skills
ln -s ~/dev/M2LabOrg-skills/debugging-discipline \
      <project>/.claude/skills/debugging-discipline

echo ".claude/skills/" >> <project>/.gitignore

2. Git submodule pinned

Each project pins a commit. Good for teams.

git submodule add \
  git@github.com:M2LabOrg/agent-skills.git \
  .claude/skills

git commit -m "chore: pin skills submodule"

3. Shallow clone CI / RunPod

Ephemeral environments where symlinks don't survive.

git clone --depth 1 \
  https://github.com/M2LabOrg/agent-skills.git \
  .claude/skills

echo ".claude/skills/" >> .gitignore

Any IDE that supports the .claude/skills/ convention — Cascade (Windsurf), Claude Code, Cursor, VS Code with the Claude extension, and others — will pick these skills up automatically. See the Anthropic skills docs.

The skills

One folder, one SKILL.md, one job done well.

/

Authoring conventions

What makes a skill in this collection.

Pushy descriptions

The YAML description is the trigger. Enumerate scenarios and phrasings; don't just state what the skill does.

Progressive disclosure

Keep SKILL.md under ~500 lines. Push detail into references/ so the model only reads it when needed.

One skill = one folder

SKILL.md + optional references/, scripts/, assets/, evals/.

Methodology, not project rules

Skills capture how we like to work, so they apply across any codebase. Project-specific bits stay tagged in the description.