Share AI artifacts with your team

Version skills, AGENTS.md and a config.json in a Git repository and let every teammate install them with one pair-cli command, for one or several AI assistants.

A walkthrough for teams who want to organize and distribute AI artifacts (skills, AGENTS.md, KB content) through a Git repository, independent of pair's product lifecycle. By the end, every teammate installs and updates your organization's artifacts with a single command.

Prerequisites

pair-cli installed (pair-cli --version); see the Quickstart
Git, and a Git hosting account (GitHub, GitLab, Bitbucket or self-hosted)

What you'll build

An artifact repository with skills, AGENTS.md and a manifest
A versioned distribution workflow using Git tags
A team config.json for consistent installs
A one-command install: pair-cli install --source <git-url>#v1.0.0 --config <config>
Private-repo auth for CI

Estimated time: 30 minutes.

1. Choose a prefix

Pick a short, unique prefix for your organization (for example acme). It prevents name clashes when skills from several sources land in one project: acme-capability-deploy, acme-process-onboard.

2. Create the repository and choose a layout

mkdir acme-ai-artifacts && cd acme-ai-artifacts
git init

Flat: every skill directory at the root, names already prefixed.

acme-ai-artifacts/
  AGENTS.md
  manifest.json
  acme-capability-deploy/SKILL.md
  acme-process-onboard/SKILL.md

Folder-organized: grouped by type, flattened and prefixed on install. Easier to maintain at scale.

acme-ai-artifacts/
  AGENTS.md
  manifest.json
  .skills/
    capability/deploy/SKILL.md
    process/onboard/SKILL.md

3. Write the team config.json

The config tells pair-cli how to distribute the artifacts. For the folder layout, start with one assistant:

{
  "asset_registries": {
    "skills": {
      "source": ".skills",
      "behavior": "mirror",
      "flatten": true,
      "flattenDepth": 2,
      "prefix": "acme",
      "description": "Acme team skills",
      "targets": [{ "path": ".claude/skills/", "mode": "canonical" }]
    }
  }
}

flattenDepth: 2 says an entry is <category>/<name> deep, so a skill's own sub-directories (a references/ folder) install inside the skill, not beside it (bounded flattening). For the flat layout, omit flatten and prefix.

Store the file in the project root (versioned with the code) or in a shared location passed with --config at install time.

4. Write AGENTS.md and manifest.json

AGENTS.md documents the skills and how to install them:

# AGENTS.md
 
This project uses Acme AI artifacts. Skills install to `.claude/skills/`.
 
## Available skills
 
- `/acme-capability-deploy`: deploy to staging or production
- `/acme-process-onboard`: onboard a new team member
 
## Setup
 
pair-cli install --source https://github.com/acme/ai-artifacts.git#v1.0.0 --config acme-config.json

Tool-specific variants of AGENTS.md (a CLAUDE.md generated from it, say) use the transform property: CLAUDE.md generated from AGENTS.md.

{
  "name": "acme-ai-artifacts",
  "version": "1.0.0",
  "description": "Acme team AI artifacts",
  "author": "Acme Engineering"
}

5. Commit, tag, push

git add .
git commit -m "feat: initial artifact package v1.0.0"
git tag v1.0.0
git remote add origin git@github.com:acme/ai-artifacts.git
git push -u origin main --tags

6. Install on a teammate's machine

# HTTPS (public repo)
pair-cli install --source https://github.com/acme/ai-artifacts.git#v1.0.0 --config ./pair-config.json
 
# SSH (private repo, uses your SSH keys)
pair-cli install --source git@github.com:acme/ai-artifacts.git#v1.0.0 --config ./pair-config.json

The CLI shallow-clones the repository at the tag, reads the config you pass, and distributes the content to the configured targets.

Private repos over HTTPS (CI, no SSH keys): set PAIR_GIT_TOKEN.

- name: Install AI artifacts
  env:
    PAIR_GIT_TOKEN: ${{ secrets.ARTIFACTS_TOKEN }}
  run: pair-cli install --source https://github.com/acme/ai-artifacts.git#v1.0.0 --config ./pair-config.json

7. Ship an update

Change the artifacts, bump version in manifest.json.
git commit -am "feat: add deployment skill v1.1.0" && git tag v1.1.0 && git push origin main --tags
Teammates run pair-cli update --source https://github.com/acme/ai-artifacts.git#v1.1.0 --config ./pair-config.json.

8. Scale to more assistants

Add symlink targets to the same registry; the skills do not change:

"targets": [
  { "path": ".claude/skills/", "mode": "canonical" },
  { "path": ".github/skills/", "mode": "symlink" },
  { "path": ".cursor/skills/", "mode": "symlink" },
  { "path": ".windsurf/skills/", "mode": "symlink" }
]

After the next pair-cli update, the new locations are populated; existing ones are untouched. A ZIP on an HTTP server, a network share or an artifact registry works as a source too: install sources.

What you've learned

How to structure an artifact repository: skills, AGENTS.md, manifest
How a team config.json makes installs consistent, with flatten and prefix
How Git tags version the artifacts and PAIR_GIT_TOKEN handles private repos
How one install reaches several AI assistants

Next steps

Team setup: full team onboarding with pair
Own and publish a Knowledge Base: when the artifacts are a whole Knowledge Base
Configuration: every config.json key

Share AI artifacts with your team

On this page