Analysis updated 2026-05-18
Develop a new AI assistant skill once and link it into many projects with symlinks
Fork and edit a skill originally installed by an external skill installer
Track which vendored skills came from which upstream source using MANIFEST.json
Publish a stable skill to a separate companion repository once it is proven
| tschuehly/skill-incubator | appleaa123/fable_os | jayvig79/plex-preroll-automation | |
|---|---|---|---|
| Stars | 4 | 4 | 4 |
| Language | Shell | Shell | Shell |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 2/5 | 3/5 | 2/5 |
| Audience | developer | developer | ops devops |
Figures from each repo's GitHub metadata at analysis time.
Requires understanding of Claude Code or Codex skill conventions and symlink-based linking via link.sh.
This repository is a personal workshop for developing Skills, reusable sets of instructions that AI coding assistants like Claude Code and Codex can follow. Rather than editing these skills separately in every project where they get used, the author keeps one master copy here and links it into other projects and into their global assistant setup using symlinks, so a change made in this repository is instantly picked up everywhere it is linked. Skills live in two places inside the repo. The skills folder holds skills the author wrote from scratch here. The vendored folder holds skills that were originally installed from an external skill installer tool, then copied into this repo so they could be edited directly, since the installer's own copies are overwritten every time it runs and any local edits would be lost. A file called MANIFEST.json keeps track of exactly which external skills were copied in this way and where they originally came from, so changes upstream can still be compared later. The basic workflow described in the readme has four steps: incubate a new skill in the skills folder written generally enough to work outside this one project, link it into a project or globally with a script called link.sh, refine it over time as real use reveals problems, and finally publish it to a separate companion repository called jvm-skills once it has proven itself reliable enough for wider use. The repository also documents a handful of specific skills in progress, covering things like coordinating multiple AI agents together, choosing which AI model to use for a task based on available quota, and reviewing documentation for clarity. These are all still described as incubating rather than finished. This project is aimed squarely at developers who already use Claude Code or similar AI coding tools and want to manage their own custom skills across many projects.
A personal repository for developing, linking, and refining custom Claude Code and Codex skills across multiple projects before publishing the stable ones elsewhere.
Mainly Shell. The stack also includes Shell, Claude Code, Codex.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.