convez.ai
SkillDesenvolvimento WebIntermediário
TuckerMoses

Claude Materia

TuckerMoses/claude-materia

Skills portáveis para Claude Code — encaixe e leve para qualquer lugar

Ver no GitHub
InicialPluginScore · de 0 a 100

Conjunto de skills e agents portáveis para Claude Code que funcionam como matéria cristalizada de conhecimento — você os encaixa em qualquer ambiente e ganham novas capacidades reutilizáveis. Cada skill é autocontido e se adapta a convenções locais quando disponíveis, ou usa padrões sensatos como fallback.

Instalação

$npx convez-ai TuckerMoses/claude-materia

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
28%
Último commit
há 1 dia
Versão
—
Issues abertas
1
Contribuidores
1

Casos de uso

  • Revisar código com múltiplos agentes especializados (coerência, design, detalhes)
  • Criar e reorganizar sessões tmux estruturadas a partir de listas de tarefas
  • Salvar e recuperar contexto de sessões de investigação interrompidas
  • Abrir abas paralelas de pesquisa sem interromper o fluxo principal

Veredito

Projeto criativo que traz padrão de design elegante (matéria como metáfora) para reutilização de skills em Claude Code. Bem documentado e pronto para uso, mas ainda em estágio inicial com agents prometidos para o futuro.

README

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

claude-materia

Portable skills and agents for Claude Code — slottable materia that extend any environment with reusable abilities. Equip them once, carry them anywhere.

What is materia?

In Final Fantasy VII, materia are crystallized orbs of knowledge. You slot them into your equipment and gain new abilities — magic, summons, support techniques. Swap them between characters, combine them for new effects, carry them across the entire game.

This plugin works the same way. Each skill is a self-contained orb of capability. Slot it into any Claude Code environment and it just works. If you have an environment set up, the materia adapts to your local conventions. If you don't, it runs with sensible defaults. No configuration required either way.

Install

claude plugin marketplace add TuckerMoses/claude-materia
claude plugin install claude-materia@claude-materia

Equipped materia

Skills

MateriaTypeWhat it does
adversarial-reviewCommandMulti-agent review loop for a scope (one file, many files, or a YAML manifest declaring intent + locations). Ships with three default reviewers (coherence, design, detail) and accepts user-supplied reviewers via env config or --reviewers <ref>. Dispatches reviewers, triage to synthesize and route, fixer to apply changes, scribe to author the final summary. Loops until clean, then promotes to opus for final verification.
sidechatSupportSpins off a tangent into a new tmux window with its own Claude Code session. The new session gets a context briefing so it hits the ground running. Your current conversation continues uninterrupted.
session-plannerSupportTurns todos into a live tmux workspace OR reorganizes an existing one. Five modes: create (todos → fresh session), reorganize (existing session → restructured), extend (existing session + todos), audit (analysis only), reannotate (migration). Confidence-weighted pane-type inference, sentinel-titled panes, an approval gate before any destructive op, and a non-transactional failure path with checkpoint logs and an incident breadcrumb.
park-sessionCommandBookmarks a Claude Code session by writing a structured pointer (session ID, when, what, next move) into a destination file you control. Lets you tear down ephemeral environments without losing the thread of mid-investigation work. Subcommands: park (default), init, unpark, list, audit.

Agents

No portable agents migrated yet. Coming soon: researcher, architect, debugger, house-wrecker, house-cleaner, teacher, expert.

Environment extension

Materia skills are designed to work anywhere, but they get smarter when they find an environment. If ~/.claude/env/index.md exists, each skill reads it on invocation, builds a relevance map of available heuristics, and derives task-specific criteria from the relevant entries.

This means:

  • With an environment: adversarial-review knows your kind definitions and can validate artifacts against their spec checklists. Sidechat forwards your conventions to the new session.
  • Without an environment: everything still works, just without the local context. The skill tells you no environment was found so you know the extension point exists.

To create an environment, add ~/.claude/env/index.md pointing to your heuristic files. See the environment convention below.

The environment convention

Every portable skill includes an ## Environment section that follows this protocol:

  1. Check if ~/.claude/env/ exists
  2. Read index.md to discover available heuristics
  3. Produce a relevance map (every entry gets an explicit disposition — no silent dropping)
  4. For relevant entries, derive task-specific criteria
  5. Include derived criteria in agent dispatches as appropriate

The environment is yours to design. The index is the only hardcoded path — everything else is discovered dynamically.

Writing a custom reviewer for adversarial-review

The adversarial-review skill accepts user-supplied reviewer agents. Drop a *.md agent file anywhere accessible and reference it via --reviewers <ref> or env config.

Minimum contract: an agent file needs only two things:

  • name (frontmatter field)
  • A non-empty prompt body describing what the agent does

Everything else is optional. The skill audits each candidate semantically — it evaluates whether your agent's described purpose will produce review-shaped output. Agents shaped like reviewers (find/identify/detect issues, emit findings with locations and severities) will pass; agents shaped like generators or orchestrators will be rejected with explanation.

Recommended shape (used by the bundled defaults):

---
name: my-reviewer
description: "What this reviewer checks for"
required: true               # or false with a trigger
trigger: null                # required if required: false
precondition: "..."          # describes when this agent should run
severity_guidance:           # hint to triage; optional
  - finding_type: my_issue
    typical_severity: high
---

# My Reviewer

You are reviewing an artifact for X. Your job is...

## What you check
- ...

## How to report findings
For each finding, emit: finding_type, severity, files (list), location, description, suggestion.

Where to put it:

  • ~/.claude/agents/<name>.md — referenced by bare name (run my-reviewer)
  • Anywhere else — referenced by path (run /abs/path/to/my-reviewer.md or run ./relative/path.md)
  • Configured in ~/.claude/env/index.md under adversarial-review.reviewers_dir for always-on inclusion

License

MIT

Ver o original no GitHub

Bom para

Relacionados