Skip to main content

skul

skul is a CLI for applying reusable AI bundles into the directories each coding tool already expects.

It is designed for teams or individuals who want to share:

  • skills
  • slash commands
  • agents
  • root instructions

without checking those generated or tool-local files into Git.

The problem​

Most AI coding tools look for project-local files in their own dot directories:

  • Claude Code uses .claude/...
  • Cursor uses .cursor/...
  • Codex uses .agents/... and .codex/...
  • OpenCode uses .opencode/...
  • GitHub Copilot uses .github/...
  • Kiro uses .kiro/...

That makes reuse awkward. You can copy files manually, commit them into every repo, or maintain custom setup scripts. All three options drift over time.

What skul does​

skul fetches a bundle source, materializes the right files into the current worktree, and records what it wrote so it can update or remove those files safely later.

Key behaviors:

  • Fetch reusable bundle content from a Git repository.
  • Write files into tool-native locations.
  • Install a whole bundle or only selected bundle items.
  • Hide those files with .git/info/exclude instead of modifying .gitignore.
  • Track desired state per repository and materialized state per worktree.
  • Re-apply the same setup in fresh worktrees with skul apply.

For root instructions, skul supports two paths:

  • untracked local composition for tool-specific dotdirs and untracked root files
  • tracked shadow rendering for repository-owned root instruction files

When a root instruction target is already tracked in Git, skul builds a managed overlay on top of the file's HEAD content instead of leaving a visible diff behind.

Typical use​

skul add github.com/sjquant/ai-bundles react-expert
skul add react-expert --agent codex --include skills/diagnose
skul status
skul sync
skul check
skul update

If you work across multiple AI coding tools, skul gives you one bundle format and one command surface instead of separate setup flows per editor or agent.