Skip to main content

Sharing Skills with the Team

This appendix is the operational companion to 2.2 Build Your First Skill. The lesson covers authoring and treating skills as software; this page covers how a skill reaches teammates at Liatrio — ad-hoc GitHub shares versus contributing to the org library.

Install basics (npx skills add, flags, troubleshooting) live in 1.4 Setup MCP Servers and Skills. Audit unfamiliar skills with the pre-install checklist in 2.2 before you enable them.

Ad-hoc share from GitHub​

A repo is installable the moment it contains a valid SKILL.md (see anatomy of a skill). Push the repo, hand someone the owner/repo string, and they install with the same npx skills add pattern from 1.4. No marketplace registration step is required.

Org default: liatrio-labs/skills​

For anything the team will depend on, don't scatter one-off personal repos — contribute to the org library, liatrio-labs/skills. It's a single monorepo everyone installs from, and every contribution is scanned for security issues before it merges.

Because the repo is internal to the Liatrio GitHub org, the owner/repo HTTPS shorthand can fail unless Git is already authenticated to GitHub over HTTPS. Install with the SSH form (or use the /plugin marketplace path in 1.4):

npx skills add git@github.com:liatrio-labs/skills.git --skill grill-me

Contribute structure​

Each org-owned skill lives in its own directory under skills/. You don't edit the marketplace manifests for Claude Code, Codex, or Cursor — CI regenerates them automatically after your PR merges to main:

liatrio-labs/skills/
├── skills/
│ ├── your-skill/
│ │ └── SKILL.md # plus references/, scripts/, assets/ as needed
│ └── catalog.yaml # entries for vendored third-party skills
├── skills-tests/your-skill/ # optional skill tests
├── .claude-plugin/marketplace.json # Claude Code manifest (generated)
├── .agents/plugins/marketplace.json # Codex manifest (generated)
├── .cursor-plugin/marketplace.json # Cursor manifest (generated)
└── Taskfile.yml # lint and test tasks

Two ways to onboard a skill​

Your skill is…PathWhat you add
Internal to the Liatrio orgDirectthe skills/your-skill/ files (manifests are generated for you)
Public, or owned outside the orgCatalogone entry in skills/catalog.yaml — CI vendors the files for you

Pick one, never both. On the catalog path, pin ref to a full 40-character commit SHA, never a branch or tag — those move under you and break the security audit trail.

Validate, then open a PR​

The repo uses Task and uv for local checks (see its README for setup):

task lint-skills -- skills/your-skill   # validate your skill's SKILL.md frontmatter
task check # full pre-push pass: all linters plus unit tests

Every PR that adds or changes a skill is scanned by two security scanners in parallel (an LLM-based semantic scan and Snyk) that return a verdict such as pass, warn, or block. Scan Gate and Lint Gate are required checks — a skill can't merge until both pass. This is the org-level version of treat skills as software artifacts: version control, CI, and review before anything reaches a teammate.

Which workflow when​

ScenarioDo this
Trying an idea, just for youAuthor locally with skill-creator, keep it in a personal repo
Sharing an experiment ad hocPush to a GitHub repo, share the owner/repo string
Something the team depends onPR it into liatrio-labs/skills and let the scanners gate it