convez.ai
SkillProdutividadeIntermediário
givepro91

CC Skills — Marketplace de Skills para Claude Code

givepro91/cc-skills

Gera CLAUDE.md e aprende de erros automaticamente

Ver no GitHub
InicialPluginScore · de 0 a 100

Conjunto de skills JavaScript para Claude Code que gera e mantém um arquivo CLAUDE.md com regras de trabalho versionadas e auto-aprendizagem. Inclui /claude-md para criar/atualizar diretrizes do projeto e /learn para transformar correções em regras duráveis que melhoram a qualidade iterativamente.

Instalação

$npx convez-ai givepro91/cc-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
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 4 meses
Versão
v0.3.0
Issues abertas
0
Contribuidores
1

Casos de uso

  • Padronizar regras de trabalho em repositórios multi-agente
  • Acumular correções como regras duráveis para Claude
  • Configurar workflows de git paralelo (team/solo/none)
  • Manter disciplina de engenharia (Karpathy-style) entre sessões

Veredito

Skill bem pensada que implementa o feedback loop Anthropic (rule → mistake → correction → rule) de forma automática e versionada. Ideia forte de usar CLAUDE.md como repositório canônico compartilhado entre agentes; a detecção de multi-tool (AGENTS.md symlink) é detalhe robusto. Depende do usuário adicionar regras quando Claude falha — não é mágica, mas é multiplicador de qualidade real em equipes.

README

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

cc-skills

A small marketplace of Claude Code skills.

Plugins

claude-md — generate & maintain CLAUDE.md, and learn from mistakes

Two skills that make a repo's AI working rules consistent and self-improving:

  • /claude-md — generate or refresh CLAUDE.md for a new or existing project. Hybrid model: a tailored preamble (scanned stack + build/test/run commands) plus an idempotent managed block holding:

    • Working Discipline — think-before-coding, simplicity-first, surgical changes, goal-driven (Karpathy).
    • Verification — always give yourself a way to verify; report honestly (the biggest quality lever).
    • Parallel Git Workflow — adaptive: team (issue → branch → worktree → PR), solo (branch/worktree per task, no PR ceremony), or none (not a git repo). One branch = one handoff if cc-handoff is installed.
    • Self-Learning Rules — a growing list, fed by /learn.

    Canonical-file aware: writes the block to the file every agent on the repo reads. Default is CLAUDE.md; but if the repo is multi-tool — CLAUDE.md is a symlink to AGENTS.md, or imports it via @AGENTS.md (so Codex reads AGENTS.md too) — the block goes into AGENTS.md instead (with a confirm prompt), and CLAUDE.md stays the thin pointer. The block lives in exactly one file, never split or duplicated.

    Re-runnable: it only ever rewrites its own <!-- CC-RULES:START -->…<!-- CC-RULES:END --> block — your hand-written content (and any OMC block) is left untouched. Accumulated learned rules survive regeneration.

  • /learn — turn a mistake into a durable rule. After Claude does something wrong and you correct it, run /learn; it distills the correction into one concise rule and appends it (deduped) to the Self-Learning Rules. This is the "compounding engineering" loop — every mistake sharpens the rules, and they load every session because they live in CLAUDE.md.

Why

  • Karpathy's discipline — minimum code, surgical changes, state assumptions, define success and loop until verified.
  • Anthropic best practice — a team shares one git-checked CLAUDE.md and adds a rule whenever Claude misbehaves; give Claude a way to verify its work (2–3× quality).
  • Parallel work needs isolation — running several Claudes at once collides on a single branch; a worktree per task (issue/branch/PR for teams) keeps them apart.

Install

/plugin marketplace add givepro91/cc-skills
/plugin install claude-md@cc-skills

Or from a local clone:

/plugin marketplace add /path/to/cc-skills
/plugin install claude-md@cc-skills

Usage

/claude-md            # generate/refresh CLAUDE.md (auto-detects team/solo/none, English)
/claude-md --ko       # Korean prose in the managed block
/claude-md --solo     # force the solo git-workflow variant
/learn                # distill the latest correction into a rule
/learn "always run the full test suite before claiming done"

What the managed block looks like

<!-- CC-RULES:START -->
<!-- Managed by /claude-md. Edits inside this block are overwritten on regen; put custom rules outside it. -->

## Working Discipline
- Think before coding · Simplicity first · Surgical changes · Goal-driven

## Verification
- Always give yourself a way to verify; report honestly.

## Parallel Git Workflow
- One task = one ISSUE = one branch = one worktree; small PRs referencing the issue.

## Self-Learning Rules
<!-- LEARN:ANCHOR -->
- (2026-06-08) Before changing a shared function's signature, grep all call sites.
<!-- CC-RULES:END -->

Project structure

cc-skills/
├── .claude-plugin/marketplace.json
└── plugins/claude-md/
    ├── .claude-plugin/plugin.json
    ├── scripts/
    │   ├── apply-block.mjs      # idempotent CC-RULES block upsert (preserves learned rules)
    │   └── append-rule.mjs      # dedup append into Self-Learning Rules
    └── skills/
        ├── claude-md/SKILL.md   # the generator
        └── learn/SKILL.md       # /learn

Scripts are plain Node ESM (no dependencies). Requires a recent Node on PATH.

License

MIT © 2026 Jay (Spacewalk)

Ver o original no GitHub

Bom para

Relacionados