Agent Skills — Biblioteca de Skills Reutilizáveis para Agentes
NishantVar/agent-skills
Skills multi-agente reutilizáveis em padrão aberto
Ver no GitHubRepositório de skills em formato aberto (SKILL.md) compatível com Claude Code, Codex CLI e Gemini CLI. Inclui skills práticas como pesquisa (Athena), orquestração de agentes (ATeam), fork de agentes (Tfork/Afork) e revisão de código (Judgy), com instalação unificada via shell scripts.
Instalação
npx convez-ai NishantVar/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
- Estender capacidades de Claude Code, Codex e Gemini com funcionalidades customizadas
- Orquestrar múltiplos agentes LLM em paralelo e coordenar execução
- Automatizar workflows de desenvolvimento com skills como fork de agentes e revisão de código
Veredito

Projeto maduro com padrão aberto bem definido (agentskills.io) e skills de alta utilidade para orquestração de agentes. Cobertura multiplataforma (Claude, Codex, Gemini) é um diferencial, mas a documentação técnica do README poderia ser mais acessível para iniciantes.
README
A tradução para o português deste README ainda está na fila.
agent-skills
Write a skill once, use it in Claude Code, Codex CLI, and Gemini CLI.
All three agents use the same core format: SKILL.md with name + description YAML frontmatter (Agent Skills open standard). Each agent reads what it understands and ignores the rest.
Flagship Skills
| Skill | Description |
|---|---|
tfork | Deterministic terminal fork — spawns a verified new terminal window running any agentic coding tool (Claude Code, Codex, Gemini) with a plan or command, using a self-verifying binary that guarantees the window actually opens. |
afork | Front door for forking any coding agent (plain or definition-backed) across runtimes (codex, claude, pi) into a cmux pane — builds a runtime-agnostic launch command and hands it to tfork. Fails closed when a declared permission restriction can't be runtime-enforced. |
p2p | P2P messaging between cmux agents using inline-prompt transport over cmux set-buffer/paste-buffer. Per-surface manifest routing; fire-and-forget. |
Skills
| Skill | Description |
|---|---|
athena | Research companion based on Andrej Karpathy's LLM wiki approach, organized into four tiers: unconfirmed → confirmed → consolidated → design (the one you actually use). |
ateam | (Experimental) Multi-LLM agent team management — spawn Claude, Codex, and Gemini teammates and coordinate them via a hybrid native + sentinel-based protocol. |
find-session | Search and resume previous sessions by keyword (Claude Code only) |
first-principles | Decompose any idea, problem, or decision down to its fundamental truths through collaborative iterative questioning |
judgy | Opinionated code reviewer that spawns a second agent to critique your staged changes |
project-mover | Move a project folder and update all path-dependent metadata for Claude Code, Codex CLI, and Gemini CLI. |
prompt-builder | Interactive prompt crafting with iterative refinement and versioned saves |
smart-grill | Interview the user relentlessly about a plan or design — but only on decisions that actually need their input. |
Install / Uninstall
./install.sh all # All skills → all agents
./install.sh claude my-skill # Single skill, single agent
./uninstall.sh all # Remove everything
Validate
./validate.sh # Checks all skills follow the format rules
Porting a Skill
If a skill lives in a local agent directory (e.g. ~/.claude/skills/) but not yet in this repo, use port.sh to bring it in:
./port.sh claude my-skill # Copy from ~/.claude/skills/my-skill
./port.sh claude my-skill --force # Overwrite if it already exists
port.sh copies the skill, validates it, and automatically adds it to the README skills table.
Adding a Skill
- Create
skills/<skill-name>/SKILL.md:
---
name: my-skill
description: >-
What it does and when to trigger it.
# Claude-specific (optional, ignored by other agents):
user-invocable: true
allowed-tools:
- Bash
---
# My Skill
Instructions for the agent...
-
(Optional) Add
agents/openai.yamlfor Codex UI metadata. -
Run
./install.sh all my-skill
Format Rules
| Rule | Detail |
|---|---|
name | Lowercase + hyphens, max 64 chars, must match directory name |
description | Max 1024 chars, include what + when to trigger |
SKILL.md body | Under 500 lines; split large content into references/ |
| Agent-specific fields | Go directly in SKILL.md frontmatter |
| Codex UI config | Goes in agents/openai.yaml |