Catalog Skill
Jeffrey22444/catalog-skill
Descobre, cataloga e roteia tarefas entre skills com índice persistente
Ver no GitHubSkill que descobre e cataloga todos os Agent Skills disponíveis no ambiente, roteando tarefas para o conjunto de skills mais relevante. Persiste um índice local para reutilização entre sessões, eliminando rescaneamentos desnecessários.
Instalação
npx convez-ai Jeffrey22444/catalog-skill1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Descoberta automática de skills em ambientes Codex e Claude
- Roteamento inteligente de tarefas para o skill mais apropriado
- Persistência de catálogo para melhor performance em múltiplas sessões
- Gerenciamento de múltiplas raízes de descoberta em workspaces
Veredito

Skill robusto para ambientes multi-skill que resolve o problema real de descoberta e roteamento inteligente. Implementação limpa em Python puro (stdlib only) com CLI bem pensada, embora seu valor dependa da quantidade de skills disponíveis no ambiente.
README
A tradução para o português deste README ainda está na fila.
Catalog Skill
Catalog discovers all currently available Agent Skills in the environment and
routes a task to the smallest relevant skill set. It can also persist a saved
catalog index for reuse in new sessions.
Catalog Skill is open source under the MIT License.
What This Repository Contains
skills/catalog: the actual runtime skill payload that gets installed into the agent's skills directoryinstall.py: the local installer for Codex and Claude Codebuild_release.py: the ZIP release buildertests/: delivery and installer tests
If you are evaluating this repository for GitHub distribution or agent-driven
installation, treat this repository root as the source package and
skills/catalog as the installable skill body.
Install From A Local Checkout
Clone the repository, then run:
python3 install.py --agent codex
python3 install.py --agent claude
python3 install.py --agent both
Preview without writing:
python3 install.py --agent both --dry-run
Upgrade an existing install:
python3 install.py --agent both --upgrade
The installer copies skills/catalog into the target skill directory and
creates a timestamped backup on upgrade. Saved indexes and user configuration
live in ~/.catalog-skill, outside the replaceable skill installation.
Register one or more workspace roots for recursive skill discovery:
python3 install.py --agent codex --discovery-root ~/Documents/workspaces
Repeat --discovery-root to merge additional roots into
~/.catalog-skill/config.json. Existing configured roots are preserved.
Install Via GitHub
The simplest GitHub-based install flow is:
git clone https://github.com/Jeffrey22444/catalog-skill.git
cd catalog-skill
python3 install.py --agent codex
For agent-driven installation, the agent only needs to:
- Clone or download the repository.
- Change into the repository root.
- Run
python3 install.py --agent codexor--agent claude.
No packaging step is required for that path as long as the repository contains the full source package layout.
Slash Commands
/catalog: prefer the saved local index; fall back to a live scan if none is saved/catalog save: rescan live and overwrite the saved local index/catalog save-<skill-name>: find only that exact skill directory name and append new locations to the existing saved index/catalog list: read the saved local index only
After /catalog save, Catalog reports skill names added and removed since the
previous saved index. Targeted save requires an existing index, retains existing
same-name locations, and does not read unrelated skill files.
Live scans always include standard Codex, Claude, plugin-cache, and
current-project roots. They also recursively discover skills directories
below configured workspace roots. CODEX_SKILL_DISCOVERY_ROOTS can temporarily
replace configured workspace roots for one process.
Release Artifacts
Build a clean delivery ZIP with:
python3 build_release.py --dist dist --overwrite
The release ZIP is for end users who receive a packaged artifact instead of a Git checkout. For GitHub installs, the repository itself is the delivery unit.
Notes
- Python 3.9+ only
- Standard library only
- Open source under the MIT License
- For GitHub publishing guidance, see
docs/GITHUB-DISTRIBUTION.md