convez.ai
SkillAutomaçãoAvançado
dividedby

Skills — Coleção de Skills Claude para Engenharia

dividedby/skills

Skills prontos para automação de engenharia, design e auditoria de projetos

Ver no GitHub
InicialPluginScore · de 0 a 100

Conjunto pessoal de skills Python para Claude que cobrem automação de engenharia, design de frontend, auditoria de código e gestão de roadmap. Inclui padrões avançados como loops autônomos com guardrails, context firewall para preservar qualidade em processamentos longos e propostas de melhoria via issues GitHub.

Instalação

$npx convez-ai dividedby/skills

1 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

Estrelas
1
Forks
2
Acompanhando
0
Saúde
57%
Último commit
há 1 dia
Versão
—
Issues abertas
2
Contribuidores
1

Casos de uso

  • Executar loops autônomos com feedback gates e paradas seguras
  • Manter qualidade em processamentos multi-item com context firewall
  • Sincronizar roadmap com estado real de issues GitHub
  • Auditar versões pinadas de ferramentas e dependências
  • Gerar propostas de refatoração e melhorias arquiteturais

Veredito

Coleção madura e pragmática de skills focada em workflows de engenharia repetitivos. Destaca-se por padrões sofisticados (autonomous-loop com HITL→AFK, context-firewall) e integração profunda com GitHub, mas exige compreensão técnica profunda e repositórios bem estruturados para extrair valor máximo.

README

A tradução para o português deste README ainda está na fila.

Skills

My personal Claude Code skills.

Install

npx skills@latest add dividedby/skills

Pick the skills you want when prompted. Re-run to update.

Skills

Engineering

  • autonomous-loop — Take a briefed backlog to a safely-running unattended (AFK) agent loop: runtime selection, stop condition, per-iteration feedback gates, HITL→AFK graduation-by-guardrail with a cap, and monitor/stop/resume. Methodology, not a runtime.
  • context-firewall — Restructure a long, multi-item run so late items don't decay: a fresh discarded sub-agent context per item, between-item budget checkpoints, and intentional compaction. Loop optional.
  • frontend-design — Design, refine, and audit production-grade frontend interfaces across React, Next.js, Tailwind, and vanilla HTML/CSS/JS that avoid generic AI aesthetics.
  • roadmap — Reconcile a repo's roadmap census with live GitHub issue state so the doc stays trustworthy as the single source for picking the next task: bootstrap the pattern, migrate a legacy planning doc, or repair drift in three tiers. Edits the working tree for review and writes additive issue comments; never commits or closes.
  • software-design — Turns a PRD and published backlog into named modules, located seams, and a testing strategy that makes issues TDD-ready.
  • staleness-audit — Audit a repo's pinned toolchain versions (Node now; wider ecosystem later) for staleness and emit a ranked, recommend-only report — the complement to Dependabot's library bumps.

Config

Setting up and auditing a project's Claude config — instruction files and the settings/hooks harness. See the bucket README for the internal seams.

  • project-claude-config — One state-routed pass over a project's Claude config: scaffold what's missing and critique what's present across both the harness (.claude/settings.json) and the instruction files (CLAUDE.md / AGENTS.md), interviewing only for facts the repo can't reveal.

Meta

  • apply-agent-research — Apply an external agent-research knowledge base to a repo's own agent-meta: read a public knowledge mirror plus the repo's own governance docs, then propose its few best improvements (at most five per run, each clearing a high independent bar) as labeled issues — never editing, committing, or merging.

Proposal loops

Scheduled, skill-driven workflows that propose improvements via labeled issues (a human decides — never auto-applied). Onboarding docs for standing them up in other repos:

  • proposal-loop-harness — the shared skeleton both loops follow (fetch-fresh skill install, propose-via-issues, scheduling).
  • consumer-setup — wire a repo up as an apply-agent-research Consumer (agent-meta self-improvement + the cross-repo skill-request / skill-promotion channels).
  • arch-review-setup — the simpler improve-codebase-architecture loop (codebase → refactor proposals).
Ver o original no GitHub

Bom para

Relacionados