AI Skill Kitchen
maxeckemoff/ai-skill-kitchen
Skills modulares para orquestração de agentes Claude em arquitetura Cowork
Ver no GitHubColeção de skills reutilizáveis para agentes LLM, projetada primariamente para Claude (Code, Cowork, Agent SDK). Inclui 6 skills prontos para arquitetura multi-sessão: gerenciamento de contexto, handoff entre agentes, estruturação de tarefas, validação de transferências, check-off de entrega e detecção de padrões reutilizáveis.
Instalação
npx convez-ai maxeckemoff/ai-skill-kitchen1 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
Casos de uso
- Coordenar múltiplas sessões de agentes especializados (Manager, Architect, Developer, Schema Expert)
- Preparar transferências quentes de contexto entre agentes com prompts scaffolded
- Monitorar pressão de contexto e aplicar estratégias de intervenção progressivas
- Estruturar requisições de implementação com constraints semânticos e tracking
- Fechar entrega com validação pós-integração e logging em tracker de projeto
Veredito

Toolkit maduro para times que operam arquitetura multi-agent coordenada. Padrões bem documentados e altamente portáveis; exige compreensão de workflows multi-sessão para máximo valor.
README
A tradução para o português deste README ainda está na fila.
AI Skill Kitchen
A small, growing collection of LLM agent skills and shared markdowns that I've built and found useful enough to share. Skills here are designed primarily for Claude (Claude Code, Cowork, the Claude Agent SDK), but the patterns and prompts often translate to other LLM frameworks with light adaptation.
What's inside
These six skills form a toolkit for operating a multi-session Cowork architecture, where one human dispatches work across role-specialized agent sessions (Manager, Architect, Developer, Schema Expert, specialists) that coordinate through shared markdown bridge files.
| Skill | What it does | Status |
|---|---|---|
| cowork-progressive-disclosure-helper | Alerts Cowork sessions on context pressure and walks you through a 6-level intervention ladder (recap, checkpoint, dispatch, multi-way split, full reorganization) before automatic compaction takes over. Decides when a scope is ready to split off. | Released |
| bridge-handoff-authoring | Scaffolds the warm-transfer kickoff prompt and initial bridge file for a new subordinate session: role assignment, sidebar identity code, scope, dataset orientation, recipes, standing rules, expected first response. | Released |
| architect-to-developer-ask-authoring | Turns a freeform description into a structured, ticket-numbered implementation ask in the bridge: scope, semantic constraints, in/out-of-scope files, deliverable format, pre-written tracker update. | Released |
| pre-launch-warm-transfer-review | Reviews a drafted kickoff prompt before launch across five categories (completeness, tone rules, dataset specificity, open-question framing, standing-rules clarity) and returns a pass / needs-revision verdict. | Released |
| check-off-protocol | Runs the post-integration close-out after you confirm a delivery is live: read the bridge entry, log to the project tracker, clear or update the bridge ask, surface follow-ups. | Released |
| skill-extraction-spotter | Watches an active session for work worth codifying as a reusable skill, then surfaces a short recommendation with sketched SKILL.md frontmatter and hands an accepted candidate to skill-creator. | Released |
More to come. The intent is to expand this kitchen over time with skills, prompts, and patterns that have earned their keep in real work.
Installing a skill
Each skill ships as a .skill bundle at the repo root, alongside its source folder:
- Download the
[skill-name].skillfile from the repo root. - Open Claude Desktop (or your Claude environment of choice).
- Either drag the file onto the Claude window, or open Settings → Skills → Install from file.
- The skill registers and becomes available in matching sessions automatically.
If you want to read or modify a skill before installing, the source (SKILL.md, README.md) lives in the matching [skill-name]/ folder.
Contributing or forking
Each skill is independent. Fork the repo, modify the skill folder you care about, repackage with your preferred tooling, and use. If you build improvements that are broadly useful, pull requests are welcome.
License
MIT. Use, modify, redistribute. Attribution appreciated but not required.
About
Maintained by @maxeckemoff. Skills are products of real workflow problems, not synthetic demos.