convez.ai
SkillDesenvolvimento WebIntermediário
photostructure

Claude Code Skills

photostructure/claude-code-skills

Workflows disciplinados para planejamento, revisão e commits no Claude Code

Ver no GitHub
InicialPluginScore · de 0 a 100

Conjunto de workflows opinados para Claude Code que enforce planejamento iterativo, revisão baseada em prova e commits limpos. Inclui skills para replanejamento, revisão de código com evidências, e organização de commits Conventional.

Instalação

$npx convez-ai photostructure/claude-code-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
5
Forks
0
Acompanhando
0
Saúde
50%
Último commit
há 2 dias
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Melhorar qualidade de decisões de design iterando múltiplas vezes antes de commitar
  • Realizar revisões de código focadas apenas em bugs reais, eliminando ruído
  • Organizar árvore de trabalho grande em commits coerentes e single-purpose

Veredito

Skills bem fundamentados que endereçam problemas reais do fluxo de trabalho com Claude Code: satisficing em planejamento, ruído em reviews e commits desordenados. Requerem integração no Claude Code marketplace, mas oferecem ganho tangível em disciplina de desenvolvimento.

README

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

Claude Code coding skills

An opinionated Claude Code plugin marketplace with five workflow skills for disciplined planning, proof-based review, and clean commits. Documented in depth at photostructure.com/coding.

SkillWhat it doesRead more
replanIterative critique-and-refine planning. Forces multiple passes before committing to a design.Claude picks the first idea that works. Make it pick the best one.
reviewCode review that requires proof before reporting — a short list of real bugs, not noise.Most AI code reviews are noise. Here's how to fix that.
review-stagedThe same proof-based review, scoped to git diff --cached, then drives a clean commit.Most AI code reviews are noise. Here's how to fix that.
gitplanUntangle a large working tree into coherent, single-purpose Conventional Commits.—
stageStage only the hunks the current session touched — never the whole file — and commit cleanly.—

Install

# Add the marketplace (from GitHub: owner/repo)
/plugin marketplace add photostructure/claude-code-skills

# Install the bundled plugin
/plugin install coding@photostructure

Then invoke any skill — plugin skills are namespaced by the plugin name:

/coding:replan
/coding:review
/coding:gitplan

To try it locally before publishing:

/plugin marketplace add /home/mrm/src/claude-code-skills
/plugin install coding@photostructure

Layout

claude-code-skills/
├── .claude-plugin/
│   └── marketplace.json        # marketplace catalog (name: photostructure)
└── plugins/
    └── coding/                 # the plugin (skills namespaced as /coding:<name>)
        ├── .claude-plugin/
        │   └── plugin.json
        └── skills/
            ├── replan/SKILL.md
            ├── review/SKILL.md
            ├── review-staged/SKILL.md
            ├── gitplan/SKILL.md
            └── stage/SKILL.md

Adapting

These skills are intentionally generic. Each SKILL.md ends with an "Adapting for your project" note — point them at your CLAUDE.md/AGENTS.md, add domain-specific critique or review checks, and tune the strictness to taste.

Further reading

The thinking behind these skills, on photostructure.com/coding:

Versioning

plugin.json has no version field, so while this marketplace is hosted in git every commit counts as a new version and /plugin update picks up changes automatically. Add an explicit version once you want stable, opt-in releases.

License

MIT © Matthew McEachen

Ver o original no GitHub

Bom para

Relacionados