convez.ai
SkillAutomaçãoAvançado
nima-karami

Agentic Development Lab

nima-karami/agentic-development

Skills prontas para transformar pesquisa em desenvolvimento agnóstico eficiente

Ver no GitHub
InicialPluginScore · de 0 a 100

Laboratório para converter pesquisas em skills reutilizáveis do Claude Code, focado em desenvolvimento agnóstico. Oferece skills prontas como `solidify-repo`, `architecture-critic`, `feature-spec` e `autonomous-build-loop` que melhoram o trabalho cotidiano com agentes de código.

Instalação

$npx convez-ai nima-karami/agentic-development

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
0
Acompanhando
0
Saúde
42%
Último commit
há 3 semanas
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Auditar repositórios para colaboração humano-agente com verificações de segurança em camadas
  • Revisar arquiteturas de software antes da implementação com rubrica objetiva
  • Especificar features de forma estruturada antes de codificar
  • Orquestrar pipelines longas de agentes autônomos com resumibilidade

Veredito

Repositório sólido que traduz pesquisa agentic em skills de produção. A abordagem de duas sínteses independentes (A cautelosa, B prescritiva) é rigorosa; a modularidade permite adoção incremental. Exige cópia manual para `~/.claude/skills/`, sem automação de instalação.

README

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

agentic-development

A lab for turning research into reusable Claude Code skills (and tools) for agentic development.

The loop: dump deep-research on an agentic-development topic, read it critically, then distill it into a self-contained, tested skill that improves day-to-day work with coding agents.

Layout

research/YYYYMM/<topic>-A.md   # two independent research syntheses per topic
research/YYYYMM/<topic>-B.md
skills/<skill-name>/SKILL.md    # skills distilled from the research

The two-source convention

Every research topic has an A and a B synthesis from independent researchers. A tends to be cautious and evidence-grounded; B tends to be confident and prescriptive (and B's specific citations are sometimes unreliable — verify before trusting). Skills are designed from where the two agree.

Skills

SkillDistilled fromWhat it does
solidify-repoagent-human-repoAudits a repo for human/agent collaboration — instruction files, deterministic checks, a one-command verify harness, and a three-layer security gate (SAST + dependency/supply-chain audit + secret scanning) — and applies fixes with per-category approval. Pulls stack-specific tool picks from a curated tooling reference.
architecture-criticagent-driven-architecture-designFresh-context adversarial review of a written design before code: a 7-dimension rubric, severity-tagged findings, and a PROCEED/REVISE verdict.
feature-specagent-driven-feature-specTurns a vague feature idea into a right-sized spec before implementation — triage tiers to avoid over-production, fixed checklists for uniform coverage, and a machine-parseable handoff block for autonomous pipelines.
autonomous-build-looplong-running-agentsConductor that drives a whole wishlist to verified completion in one long/unattended run — composes the other skills per phase, keeps state on disk for resumability, and scales pipeline depth to each feature's tier.

Using a skill

Skills are authored and version-controlled here, but only take effect once copied into your global Claude Code skills directory:

cp -r skills/<skill-name> ~/.claude/skills/

Then invoke it from Claude Code (e.g. "solidify this repo", or "review this design before I build it").

How skills are built

Each skill is authored with the superpowers plugin for Claude Code: brainstorming to turn research into an approved design, then writing-skills (test-driven — baseline the behavior with subagents, write the skill to close the gap, refactor until solid).

License

MIT

Ver o original no GitHub

Bom para

Relacionados