Common Agent Skills
hheydaroff/common-agent-skills
Um repo, um deploy, múltiplos agentes IA sincronizados
Ver no GitHubRepositório unificado para gerenciar skills de agentes IA (Pi, Claude Code, Codex) em um único lugar, evitando duplicação e desvio de edições. Fornece um workflow simples: editar, fazer deploy em todos os tools com um comando, e versionar via git.
Instalação
npx convez-ai hheydaroff/common-agent-skills1 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
- Manter skills em sincronismo entre Pi, Claude Code e Codex
- Gerenciar versões centralizadas de skills com git
- Criar pipelines de skills reutilizáveis entre ferramentas
Veredito

Solução pragmática para quem trabalha com múltiplos agentes IA e precisa manter skills sincronizadas. Reduz friction operacional, mas requer familiaridade com bash e estrutura de Agent Skills standard.
README
A tradução para o português deste README ainda está na fila.
common-agent-skills
A unified home for agent skills that works across AI coding tools — Pi, Claude Code, Codex, and any tool following the Agent Skills standard.
The Problem
When you use multiple AI agents, each tool has its own skills directory. The same skill ends up as 3 copies, edits drift, and backup means backing up N places.
The Solution
One repo. One place to edit. One deploy command.
~/development/common-agent-skills/skills/ ← edit here
↓ deploy.sh
~/.agents/skills/ Pi (native support)
~/.claude/skills/ Claude Code
~/.codex/skills/ Codex
Install
git clone git@github.com:hheydaroff/common-agent-skills ~/development/common-agent-skills
bash ~/development/common-agent-skills/setup.sh
That's it. Your skills are live across all tools.
Workflow
# 1. Add or edit a skill
vim ~/development/common-agent-skills/skills/<skill-name>/SKILL.md
# 2. Deploy to all tools
bash ~/development/common-agent-skills/deploy.sh
# 3. Commit
git add -A && git commit -m "add: skill-name" && git push
Included Skills
| Skill | Description |
|---|---|
skill-manager | Teaches any agent how to add, edit, deploy, and manage skills in this repo |
Adding Your Own Skills
Skills follow the Agent Skills standard — a SKILL.md with name and description frontmatter:
skills/
my-skill/
SKILL.md ← required
scripts/ ← optional helper scripts
references/ ← optional docs loaded on-demand
Design Principle: Skills Should Be Independently Installable
Each skill should work standalone. A user can install solution-ideator without prd-creator, or diagnose without improve-codebase-architecture, and nothing breaks.
In practice this means:
- Describe the next activity, not a named skill. Instead of "use
prd-creatornext", say "formalise the approach into a product spec next". If the user has a skill for that activity, they'll use it; if not, the phrase still makes sense. - No hard dependencies in frontmatter descriptions. A skill's description should trigger on the user's intent, not on the presence of a sibling skill.
- Cross-references use activity language. "Run a structured grilling session" is portable; "use the grill-me skill" is not.
- Sibling skills in a family can name each other casually (e.g.
cmux-browser↔cmux-markdown) because they ship together, but even then prefer the capability framing.
The workflow is implicit: a skill that produces a PRD naturally invites a skill that consumes one. Users who install adjacent skills will discover the pipeline; users who install only one still get a complete, useful tool.
Tool Support
| Tool | How skills are loaded |
|---|---|
| Pi | Reads ~/.agents/skills/ natively |
| Claude Code | Deployed to ~/.claude/skills/ |
| Codex | Deployed to ~/.codex/skills/ (additive, .system/ preserved) |
Requirements
- bash
- rsync
- One or more of: Pi, Claude Code, Codex