Skip to content

jup vs. npx skills

Both jup and npx skills are CLI tools designed to manage and install Open Agent Skills (SKILL.md files) for AI coding assistants like Gemini CLI, Claude Code, and Cursor. However, they take different architectural approaches and cater to different workflows.

Feature Comparison

Feature jup npx skills
Language / Runtime Python (via uv or pip) Node.js (via npx)
State Management Central lockfile (~/.jup/skills-lock.json) Global or per-project installation
Multi-Harness Sync
Yes. Symlinks/copies from a central cache to all configured agents via jup sync Yes. Via -a flags during install (e.g., -a gemini-cli -a cursor)
Search / Registry Yes. Built-in search (jup find) backed by skills.sh Yes. Built-in search (npx skills find) backed by skills.sh
Local Development Excellent. Symlinks local directories directly into agent folders Good. Includes an init command for scaffolding
Updates jup sync --update npx skills update / npx skills check

Command Reference

jup

  • jup find <query>: Searches for available skills in the skills.sh registry. Lists results by default; use --interactive (-it) for interactive installation. Supports --limit (-n) and --min-installs (-i) filtering.
  • jup add <repo|path>: Installs a skill from a GitHub repo or a local directory. Supports --path and --skills for sub-selection.
  • jup sync: Syncs managed skills from the central storage into the configured agent directories (e.g., .gemini/skills, .claude/skills).
  • jup list: Shows all installed skills, locations, and last update times.
  • jup remove <target>: Uninstalls a skill or repository.
  • jup config set harnesses gemini,claude: Configures which harnesses receive the synced skills.

npx skills

  • npx skills find [query]: Interactively searches for available skills in the community registry.
  • npx skills add <package>: Installs a skill. Supports --global and targeting specific agents via -a.
  • npx skills list: Lists installed skills.
  • npx skills check / update: Manages skill versioning.
  • npx skills init: Scaffolds a new skill template.

Pros and Cons

jup

Pros: * Centralized Multi-Agent Sync: You configure your target agents once (jup config set agents ...), and every jup sync ensures all agents have exactly the same skills via lightweight symlinks. * Local-First Design: It natively supports adding local skill directories (jup add ./my-skills). Because it uses symlinks by default, you can edit the source skill locally and the changes are immediately available to your agents without needing to re-run an update. * Interactive Search Registry: Deep integration with skills.sh allows you to easily discover and install community skills via jup find. * Python Ecosystem: Ideal for developers already using uv or Python toolchains.

Cons: * Requires Python: Adds a dependency for users strictly in the Node.js ecosystem.

npx skills

Pros: * Interactive Search Registry: Deep integration with skills.sh allows you to easily discover and install community skills via npx skills find. * Zero-Install for JS Devs: If you have Node.js installed, you can just run npx skills without any prior installation step. * Skill Scaffolding: The init command makes it very easy to start writing a new skill from scratch.

Cons: * Requires Node.js: Adds a dependency for users strictly in the Python/Rust/Go ecosystems. * Less Transparent State: Relies on direct installation into .claude or .gemini directories without a highly visible lockfile synchronization step, making local cross-agent iteration slightly more manual if not using global installation flags.