ai-tools
pop-dog/ai-tools
CLI for managing Claude Code skills and agents across projects
Instalação
npx convez-ai pop-dog/ai-tools1 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
README
A tradução para o português deste README ainda está na fila.
ai-tools
A catalog of Claude Code skills and agents, installed with a single command.
Overview
This repo holds reusable Claude Code skills (directories with a SKILL.md) and agents (flat .md files) under /skills and /agents. A single install.sh — run straight from GitHub, no clone required — mirrors the catalog to ~/.ai-tools and symlinks your picks into ~/.claude, where Claude Code discovers them. Re-running the installer updates everything in place.
Key Features
- One-command install —
curl | bash, pick what you want from an interactive menu - Dynamic catalog — the menu is whatever is in
/skillsand/agents; adding a skill to the repo is all it takes to publish it - Full local mirror — the whole catalog is cached in
~/.ai-tools; installing is just symlinking, so updates flow to every installed item at once - Agent-friendly — non-interactive flags (
--all,--list, names) for coding agents and scripts - Safe by construction — never touches skills or agents it didn't install (e.g. skills.sh-managed entries)
- Clean uninstall — one flag removes all symlinks and the mirror
Quick Start
# interactive: pick skills/agents from a menu
curl -sSL https://raw.githubusercontent.com/pop-dog/ai-tools/mainline/install.sh | bash
# for coding agents — install everything, no prompts
curl -sSL https://raw.githubusercontent.com/pop-dog/ai-tools/mainline/install.sh | bash -s -- --all
Usage
# see what's available
curl -sSL https://raw.githubusercontent.com/pop-dog/ai-tools/mainline/install.sh | bash -s -- --list
# install specific items by name
curl -sSL https://raw.githubusercontent.com/pop-dog/ai-tools/mainline/install.sh | bash -s -- pdf-to-md teach-me
# update: refresh the mirror; everything installed updates in place
curl -sSL https://raw.githubusercontent.com/pop-dog/ai-tools/mainline/install.sh | bash -s -- --update
# uninstall: remove all installed symlinks and delete ~/.ai-tools
curl -sSL https://raw.githubusercontent.com/pop-dog/ai-tools/mainline/install.sh | bash -s -- --uninstall
How It Works
- The installer downloads the repo tarball and mirrors
/skillsand/agentsinto~/.ai-tools(the store) on every run. - Installing symlinks an item from the store into
~/.claude/skills/<name>or~/.claude/agents/<name>.md. Only installed items are visible to Claude Code; the rest just sit in the cache. - Updating (
--update, or any later run) refreshes the store, so installed items update through their symlinks automatically. - Anything in
~/.claudethat doesn't point into the store is a foreign entry and is never modified or removed. - The store is disposable: local edits to installed items live in
~/.ai-toolsand are overwritten on the next run. Durable changes belong in this repo.
For the project vocabulary see CONTEXT.md; for design decisions see docs/adr/.
Contributing
Add a skill (a directory containing a SKILL.md) under /skills, or an agent (a single .md file with YAML frontmatter) under /agents, and open a pull request. The installer picks up new items automatically.
Testing a skill before committing
dev-link.sh (clone-only) symlinks a working-tree skill or agent straight into ~/.claude, bypassing the store — so your edits are live and you can test before you commit or push.
./dev-link.sh my-skill # link one working-tree item, live
./dev-link.sh --all # link everything in the working tree
./dev-link.sh --list # working-tree items and their link status
./dev-link.sh --unlink my-skill # remove that dev link
./dev-link.sh --unlink --all # remove every dev link from this repo
Dev links point at the working tree, not the store, so install.sh treats them as foreign and never overwrites them. If a name is already installed from the store, dev-link.sh skips it — rm that link or run install.sh --uninstall first.
License
MIT