Skills

Jump Skills Ninjas: A Meta-Skill That Routes Agents to 300+ Specialized Skills On Demand

Jump Skills' Ninja meta-skills are best for routing requests in agents like Claude Code, Kiro, and Cursor to curated AWS/Azure/Firecrawl skill libraries, but they cannot be installed via npx skills — you must first clone about 15 source repositories locally (about 500MB) and then run the install scripts.

Official source · fabricioctelles/jump-skills

Install

git clone https://github.com/fabricioctelles/jump-skills.git && cd jump-skills && ./sync-repos.sh && ./install-ninjas.sh

Who It's For

  • For people who need to quickly find the right tool across a large set of cloud skills: users who want to use AWS (230+ skills), Azure (40 skills), or Firecrawl (33 skills) related skills in agents such as Claude Code, Kiro CLI, Cursor, Codex, OpenCode, and VS Code Copilot.
  • Ninja is essentially a meta-skill (orchestrator): it loads only about 500 tokens of metadata and loads the full skill instructions on demand, reducing context usage.
  • When it's not a fit: it cannot be installed via npx skills — this is a meta-repository, and the paths Ninja references point to cloned repositories that must exist locally. If you don't want to clone roughly 15 source repositories locally (about 500MB), it's not suitable.
  • The evidence does not state the specific agent version requirements, OS differences, or text/task-type limitations it supports; these are unknown — refer to the repository README.

First Try in 3 Minutes

The recommended main path is the officially provided clone + script installation (the README explicitly does not support npx skills):

  1. Prerequisites: git and a shell environment installed locally, and at least one supported agent (e.g., Claude Code).
  2. Run in a terminal:
git clone https://github.com/fabricioctelles/jump-skills.git
cd jump-skills
./sync-repos.sh
./install-ninjas.sh
  1. Side effects and checkpoints: sync-repos.sh reads repos.md and clones about 15 source repositories into repos/<ninja-name>/ (about 500MB of disk usage); install-ninjas.sh copies the Ninja SKILL.md files into the detected agent directories (e.g., ~/.claude/skills/, ~/.kiro/skills/, ~/.cursor/skills/), replaces {{JUMP_SKILLS_DIR}} with absolute paths, and generates a .skills-map mapping. This is a file-writing operation — confirm disk space and target directories before running.
  2. Trigger a first low-risk task: in an installed agent (e.g., Claude Code), simply ask the Ninja to perform a diagnostic/recommendation task — for example, ask the AWS Ninja to recommend the most suitable specialized skill for your problem, rather than executing a deployment directly. Success signal: the Ninja names a matching specialized skill from its curated library and gives routing instructions.
  3. The evidence does not state whether running review or recommendation actions involves reading/writing files, executing commands, or accessing the network; these behaviors are unknown — do not assume zero risk.

Use Cases

  • Case 1: Let the Ninja route your AWS tasks. Scenario: you're unsure which specialized skill to use. How to start: in Claude Code, tell the installed AWS Ninja "recommend the most suitable skill for my task." As a meta-skill it loads only about 500 tokens of metadata, then loads the full instructions of the matching skill on demand; the expected result is a concrete recommendation from among the 230+ AWS skills in the library. Note: the evidence does not state whether the Ninja will automatically execute the recommended skill or access AWS afterward — confirm before any action that modifies resources.
  • Case 2: Routing for Firecrawl or Azure scenarios. Scenario: you need specialized skills for web scraping (Firecrawl Ninja, 33 skills) or Azure (40 skills). Start the same way: after installation via supported paths, pose the task to the corresponding Ninja in your agent. What can be confirmed is that the Ninja completes skill routing and loads instructions on demand; the downstream actual scraping/cloud operation results depend on the routed skill — the evidence does not elaborate, so don't assume specific outputs.

Usage Boundaries

  • Files and disk: Installation necessarily clones about 15 repositories (about 500MB) and writes SKILL.md files into the agent's skills directory; the README does not detail overwrite/merge behavior — back up or inspect target directories before running.
  • Command execution: The main path relies on running two local shell scripts — read their contents before executing.
  • Network access: Cloning source repositories requires network access; whether the Ninja accesses external services at runtime is not stated in the evidence and is unknown.
  • Client prerequisites: install-ninjas.sh auto-detects agents and installs into their respective directories (Claude Code: ~/.claude/skills/; Kiro CLI: ~/.kiro/skills/; Cursor: ~/.cursor/skills/, plus Codex, OpenCode, and VS Code Copilot).
  • Explicit limitations: The repository officially cannot be installed via npx skills; the paths referenced by Ninja skills require the cloned repositories to exist locally.
  • Unconfirmed: Supported operating systems, uninstall/update methods, the specific language and quality of skill content, and whether the Ninja has additional side effects once triggered — none have official evidence; be cautious with first-time external operations.