convez.ai
Skill
rdeknijf

dotclaude

rdeknijf/dotclaude

Ver no GitHub
InicialPluginScore · de 0 a 100

Model-agnostic Claude Code skills for exploiting temporary access to a smarter model: genius-check reviews one project; genius-queue loops it across your backlog.

Instalação

$npx convez-ai rdeknijf/dotclaude

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 1 semana
Versão
—
Issues abertas
0
Contribuidores
1

README

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

dotclaude

My curated, publicly shareable Claude Code config: the bits of ~/.claude worth sharing. Skills today; custom commands, agents, hooks, statusline, and sanitized settings as I clean more up.

It's a hand-picked mirror of ~/.claude, not a sync. Everything here is scrubbed of private paths, project names, and secrets. Structure mirrors the real thing, so installing is just copying the pieces you want into your own ~/.claude/.

Skills

SkillWhat it does
genius-checkOne brilliant pass over one project. The genius reviews it read-only and writes a prioritized, executable GENIUS-CHECKUP.md a weaker model can act on later. The reusable unit.
genius-queueThe loop. Runs your highest-leverage backlog (global instruction-layer fixes, then gnarly per-project problems, then a breadth checkup sweep) through the genius across many projects, then materializes every plan to disk in one pass at the end.

The genius skills exist to solve one problem. You occasionally get temporary access to a much smarter model (a frontier trial, a temporarily-unlocked higher tier, a better harness), and the window closes before you've extracted its value. The genius model should think; your cheaper day-to-day models should execute its plans afterward. The skills turn that into a routine: leverage-first ordering, run the whole batch then materialize once, every output a concrete artifact a weaker model runs later. Reviews are read-only; proposals are .proposed / GENIUS-*.md files, so nothing outward-facing happens without you. Both are model-agnostic; only the model alias changes.

Install

Easiest: point your agent at this repo and let it do the work. Something like:

Read https://github.com/rdeknijf/dotclaude and install the genius-check and genius-queue skills into my setup, wherever makes sense for me (global ~/.claude/skills/, or this project), and adapt the genius-model alias to whatever top-tier model I actually have.

It reads the SKILL.md files, drops them where your harness looks for skills, and fixes up the model alias for you. That's the point of shipping them as plain, self-describing markdown.

Prefer to do it by hand? Everything mirrors ~/.claude/, so just copy what you want:

git clone https://github.com/rdeknijf/dotclaude.git
cp -r dotclaude/skills/genius-check dotclaude/skills/genius-queue ~/.claude/skills/

Either way, invoke them in Claude Code with /genius-check <path> or /genius-queue.

License

MIT

Ver o original no GitHub

Bom para

Relacionados