Agent Scripts para Claude Code
mgm702/agent-scripts
Scripts e skills organizados para maximizar a produtividade no Claude Code
Ver no GitHubColeção centralizada de instruções e skills específicas de domínio para potencializar o uso do Claude Code. Inclui workflows de desenvolvimento, automação de tarefas, planejamento e otimizações, estruturados em Ruby para integração direta com o ambiente de codificação.
Instalação
npx convez-ai mgm702/agent-scripts1 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
- Automatizar workflows de desenvolvimento e code review antes de push
- Executar planejamento estruturado e pesquisa profunda de codebase
- Validar ideias, otimizar performance e gerenciar releases com checklists
Veredito

Repositório bem estruturado de skills reutilizáveis em Ruby que cobrem ciclo completo de desenvolvimento (planejamento, codificação, review, deploy). Requer configuração inicial mas oferece automações valiosas para fluxo de trabalho do Claude Code.
README
A tradução para o português deste README ainda está na fila.
agent-scripts
Centralized AI agent instructions and domain-specific skills. Source of truth for coding workspace at ~/code/.
Structure
agent-scripts/
├── AGENTS.md # Core protocol: git safety, code style, quality rules
├── setup/
│ └── notion-mcp.rb # One-time Notion MCP setup for Claude Code
└── skills/
├── ai-cost-comparision/ # LLM provider cost comparison & model selection
├── code-review-and-quality/ # Multi-axis review of local changes before push
├── coding-standards/ # Core guardrails to read every time before coding
├── create-cli/ # CLI tool scaffolding patterns
├── devops-agent/ # AWS infra assessment & phased launch plan
├── duckduckgo-search/ # Web search & content extraction via DDG Lite
├── find-community/ # Find a community to build a business around
├── github/ # gh CLI workflows & automation
├── idea-refine/ # Sharpen a raw idea into a one-pager for planning
├── implementation-control/ # Controlled execution & progress feedback
├── marketing-plan/ # Minimalist content-first marketing plan
├── mobile-developer/ # React Native, step-by-step with explanations
├── mvp/ # Build the smallest viable product, manual-first
├── object-oriented-design-patterns/ # SOLID + GoF patterns reference
├── performance-optimization/ # Measure-first performance audit & fixes
├── plan-and-annotate/ # Structured planning before implementation
├── pricing-strategy/ # Pricing model, tier design & recommendation doc
├── release-manager/ # Pre-release audit & checklist generation
├── research-first/ # Deep codebase/domain research before coding
└── validate-idea/ # Market research + validate by selling before building
Setup
New machine
Everything below writes into $CLAUDE_CONFIG_DIR, which defaults to ~/.claude.
If your Claude Code config lives anywhere else, export it first — otherwise the
skills and hooks land in a directory Claude Code never reads, and nothing errors:
export CLAUDE_CONFIG_DIR=~/.claude-config/personal # skip if you use ~/.claude
Then run these after cloning the repo:
# Link skills into Claude Code
ruby ~/code/projects/agent-scripts/setup/symlink-skills.rb
# Install hooks into settings.json (coding-standards guardrails on every edit)
ruby ~/code/projects/agent-scripts/setup/install-hooks.rb
# Configure Notion MCP (requires a Notion integration token)
ruby ~/code/projects/agent-scripts/setup/notion-mcp.rb
symlink-skills.rb and install-hooks.rb each print Using config dir: ... on the
first line — check it before moving on. notion-mcp.rb doesn't read the variable
itself; it inherits it through the claude CLI it shells out to.
All three are idempotent — re-run any of them safely. install-hooks.rb merges into
settings.json without touching existing hooks, MCP servers, or plugins.
Usage
Reference from other projects
Symlink AGENTS.md into your project root:
ln -s ~/code/projects/agent-scripts/AGENTS.md ~/code/my-project/AGENTS.md
Or point agents to this repo directly in your project's config.
Reference a skill
Point agents to a specific skill when needed:
See ~/code/projects/agent-scripts/skills/performance-optimization/SKILL.md
Skills
| Skill | Purpose |
|---|---|
ai-cost-comparision | Compare LLM provider pricing and select a model to minimize spend |
code-review-and-quality | Multi-axis review (correctness, readability, architecture, security, perf) of local changes before pushing |
coding-standards | Core guardrails (the four rules) to read every time before writing or editing code |
create-cli | Design & scaffold CLI tools with proper flags, output, and testing |
devops-agent | Assess AWS infra and produce a phased production-readiness launch plan |
duckduckgo-search | Web search & content extraction via DuckDuckGo Lite (no API key) |
find-community | Identify and evaluate communities to build a minimalist business around |
github | gh CLI workflows: issues, PRs, CI checks, releases |
idea-refine | Refine a raw idea into a sharp one-pager, then hand off to plan-and-annotate |
implementation-control | Controlled execution, quality gates, and supervisor-style feedback |
kill-code | Evidence-based dead-code elimination with git-safe deletion batches |
marketing-plan | Content-first marketing plan that builds an audience before spending on ads |
mobile-developer | React Native development, step-by-step with explanations |
morning-metrics | Daily project-scoped business + infra dashboard (Stripe, Mercury, AWS, Mixpanel, Sentry, Firecrawl, scrapers) as one HTML page |
mvp | Build the smallest viable product — manual, then processized, then productized |
object-oriented-design-patterns | SOLID principles and GoF patterns — when to reach for each, and when not to |
performance-optimization | Measure-first audit & fixes across Go, Ruby, React, RN, Python, C |
plan-and-annotate | Structured plan docs with iterative annotation before implementation |
pricing-strategy | Select a pricing model, design tiers, and produce a recommendation doc |
product-launch | Launch-day distribution checklist across PH, HN, Reddit, social, and owned audiences |
release-manager | Pre-release audit and go-live checklist generation |
research-first | Deep codebase/domain research with written findings before planning/coding |
sales-bumps | Detect cold email leads and draft bump messages (never auto-send) |
validate-idea | Automated market research with evidence-graded scoring, then validate by selling, not building |
Adding Skills
Create skills/<name>/SKILL.md. Pure markdown, telegraph style, ~30-50 lines. Sections: Purpose, Rules/Patterns, Requirements.