Claude Skills para Engenheiros Sênior
pathiec92/claude-skills
Playbooks de principal engineer para arquitetura, code review e ML systems
Ver no GitHubColeção curada de skills reutilizáveis para Claude Code focada em arquitetura cloud-native, MLOps e aplicações LLM/AI. Cada skill é um framework executável que Claude invoca automaticamente quando seu request corresponde ao gatilho, eliminando a necessidade de colar checklists manualmente.
Instalação
npx convez-ai pathiec92/claude-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
- Revisar arquiteturas de sistemas distribuídos e cloud-native
- Code review em nível de principal engineer com foco em observabilidade
- Design de sistemas de ML/RAG/LLM em produção
- Documentar decisões arquiteturais com ADRs
- Análise pós-incidente e postmortems
Veredito

Excelente para engenheiros sênior que precisam de frameworks executáveis em vez de documentação. A seleção de skills é bem pensada com gatilhos explícitos, mas requer adoção em Claude Code (ferramenta específica) — portabilidade limitada a outros ambientes.
README
A tradução para o português deste README ainda está na fila.
claude-skills
A curated set of Claude Code skills for senior engineers and architects working on cloud-native platforms, distributed systems, MLOps, and LLM/AI applications.
Each skill is a focused, reusable framework that Claude invokes automatically when your request matches it — so the right principal-engineer playbook shows up at the right moment, without you having to paste a checklist every time.
What's a skill?
A skill is a SKILL.md file with YAML frontmatter (name + description) and a markdown body. Claude reads the description to decide when to apply the skill, and the body for how. These skills are designed with explicit trigger boundaries (each says when to use it and when to defer to a sibling), so selection stays unambiguous as the catalog grows.
Skill catalog
| Skill | What it does | Triggers when you… |
|---|---|---|
| architecture-review | System/infra design review across reliability, scale, security, ops, cost | ask "review this architecture", discuss SPOFs, DR, 10x/100x scaling |
| principal-code-review | Source/diff review at a principal bar (correctness → observability) | ask to review code, a PR, or hunt for bugs |
| gcp-architecture | GCP service selection, cost, and IAM with a managed-first bias | design on GCP, mention Cloud Run/GKE/Pub/Sub/BigQuery |
| ml-system-design | End-to-end ML systems: data → training → serving → MLOps | design a recommender/ranking system, review an ML pipeline |
| rag-system-design | Production RAG: chunking, retrieval, reranking, grounding, eval | build/debug RAG, semantic search, or Q&A-over-docs |
| llm-app-architecture | Production LLM apps: tools, agents, guardrails, eval, cost | build an LLM feature, agent, or assistant |
| mlops-pipeline-review | Data/training/registry/CI-CD-CT, deployment, drift, rollback | productionize a model or review an ML pipeline |
| observability-review | Metrics, logs, traces, SLIs/SLOs, alerting | add instrumentation, define SLOs, cut alert noise |
| incident-postmortem | Blameless RCA: timeline, contributing factors, action items | write a postmortem after an outage |
| adr-writer | Architecture Decision Records with options and trade-offs | document a significant technical decision |
| leadership-lens | Org/team/governance framing of technical decisions | weigh platform vs point solution, standardization, tech-debt |
| interview-mode | Acts as a Principal interviewer; withholds answers, scores you | practice system-design / ML-system-design interviews |
| learning-coach | Teaches at architect depth: why, trade-offs, failure modes | want a deep dive, not a tutorial |
Install
Skills live in ~/.claude/skills/. Clone this repo and run the installer for your platform, or copy the folders manually.
macOS / Linux:
git clone https://github.com/pathiec92/claude-skills.git
cd claude-skills
./install.sh
Windows (PowerShell):
git clone https://github.com/pathiec92/claude-skills.git
cd claude-skills
./install.ps1
The installer copies each skill into ~/.claude/skills/. Re-run it after pulling updates. Restart Claude Code (or start a new session) to pick up new skills.
Use just one or two
You don't need all of them — each skill is a self-contained folder. Copy only the ones you want:
cp -r skills/rag-system-design ~/.claude/skills/
Design principles
These skills follow a few rules that make automatic selection reliable:
- Single responsibility — one job per skill; no kitchen-sink skills.
- Explicit boundaries — every description states when not to use it and which sibling to use instead (e.g. RAG →
rag-system-design, notml-system-design). - Action-oriented bodies — checklists and frameworks, not prose, so Claude produces structured output.
- Always-on vs on-demand — these are on-demand frameworks; keep persistent preferences in your
CLAUDE.md, not here.
About
Maintained by Pampapathi K — AI Platform / Cloud Architect focused on GCP, MLOps, and large-scale distributed systems. These skills distill the review frameworks and design checklists I use day to day.
Contributions and forks welcome.
License
MIT — use, modify, and share freely.