Claude Skills — Kit de Workflows Especializados
martin0ne/claude-skills
Skills modulares para automatizar RAG, auditoria, criação de conteúdo e análise com Claude Code
Ver no GitHubConjunto de skills reutilizáveis para Claude Code que transformam um agente genérico em especialista para workflows específicos. Inclui RAG sobre documentos com provenance de fonte, auditoria pré-publicação de repos, geração de case studies e briefs de podcasts — tudo com scripts Python stdlib leve e zero dependências externas.
Instalação
npx convez-ai martin0ne/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
- Retrieval augmented generation (RAG) sobre bases documentais com citações precisas de arquivo:linha
- Auditoria de segurança pré-publicação: detecção de secrets, PII e dados confidenciais em repos
- Geração automática de case studies, workshops de IA e briefs de conteúdo
Veredito

Excelente toolkit para quem constrói agentes Claude com workflows complexos. A modularidade (SKILL.md + scripts + assets) é bem pensada e as implementations (BM25 puro, pre-scan determinístico) são production-ready. Melhor para intermediários que já usam Claude Code.
README
A tradução para o português deste README ainda está na fila.
claude-skills
A toolkit of reusable Claude Code skills — self-contained packages (SKILL.md + bundled scripts/references/assets) that turn a general agent into a specialized one for a specific workflow. Built and used as part of my day-to-day AI build work.
Each skill follows the progressive-disclosure model: a short metadata header decides when the skill fires, the body carries the how, and bundled resources load only when needed.
Skills
| Skill | What it does |
|---|---|
kb-rag | Retrieval over a folder of project/client documents with source provenance (file:line citations) — a small, dependency-light BM25 index in pure Python stdlib. Answers "what do the docs say about X — and where exactly". |
repo-publish-audit | Pre-publication gate for any repo going public: a deterministic secret/PII pre-scan plus an agentic multi-dimension audit (secrets, PII, confidential data, git history, hygiene, licensing) with adversarial verification → a per-repo SAFE / FIX-FIRST verdict. Defense in depth: the agent and the grep catch different classes of risk. |
case-study-builder | Distills a project, product, or tool into a recruiter-readable case study (context → problem → solution → stack → impact → lessons). |
ai-workshop | Generates a ready-to-run "AI adoption" workshop pack from a topic or industry — agenda, exercises, use-case discovery, deck outline. |
weekly-podcast-brief | Turns a dense weekly review into a clean, self-contained source brief plus an audio-overview prompt — re-shaping a long week into a handful of interlocking threads for passive listening. |
A larger, standalone agent — a multi-step ReAct document agent with a human-approval gate and cited reports — lives in its own repo: agent-flow.
Using a skill
These are Claude Code skills. Drop a skill folder into ~/.claude/skills/ (or your project's skills directory) and Claude Code discovers it automatically — the description in each SKILL.md decides when it triggers. Scripts under scripts/ can be run directly; references/ and assets/ load on demand.
skills/<name>/
├── SKILL.md # metadata (name + description) + instructions
├── references/ # docs loaded into context as needed
├── scripts/ # executable helpers
└── assets/ # templates used in output
Running the helper scripts directly
Two skills bundle standalone scripts you can run outside the agent — pure Python/bash stdlib, no installs:
# kb-rag — BM25 retrieval with file:line provenance (Python 3.9+)
python3 skills/kb-rag/scripts/kb_search.py index <docs-folder> # build the index
python3 skills/kb-rag/scripts/kb_search.py search <docs-folder> "query" # ranked hits + citations
python3 skills/kb-rag/scripts/kb_search.py stats <docs-folder> # index summary
# repo-publish-audit — deterministic secret/PII pre-scan (bash + git)
bash skills/repo-publish-audit/scripts/scan_secrets.sh <repo-path>
skills/repo-publish-audit/scripts/audit_workflow.js is a Claude Code Workflow-runtime module (run by the agent via the Workflow tool), not a standalone Node CLI. The other three skills (case-study-builder, ai-workshop, weekly-podcast-brief) are prompt-driven and run inside Claude Code.
Author
Marcin J. Ołdak — AI builder. mjoldak.pl · github.com/martin0ne
License
MIT — use freely, attribution appreciated.