convez.ai
SkillProdutividadeIntermediário
Phluxxed

llm-wiki

Phluxxed/llm-wiki

Wiki automático mantido por IA com scaffolding, templates e controle de qualidade

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill que cria um scaffold de wiki persistente mantido por IA, seguindo o padrão de wiki pessoal de Karpathy. Inclui estrutura de diretórios pronta, templates, scripts de linting e avaliação com LLM-as-judge, e histórico de mudanças append-only.

Instalação

$npx convez-ai Phluxxed/llm-wiki

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
1
Acompanhando
0
Saúde
28%
Último commit
há 2 semanas
Versão
—
Issues abertas
1
Contribuidores
1

Casos de uso

  • Documentar padrões e convenções de projetos com IA
  • Manter histórico estruturado de experimentos e decisões arquiteturais
  • Criar bases de conhecimento colaborativas com validação automática
  • Organizar runbooks, ADRs e papers em formato único

Veredito

Implementação sólida do padrão Karpathy com tooling robusto (linting, eval, render). Ideal para equipes que querem documentação estruturada mantida por agentes, mas exige Python e familiaridade com scaffolding de projetos.

README

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

llm-wiki

A skill that scaffolds a persistent, LLM-maintained wiki — inspired by Karpathy's personal wiki pattern.

Drop it into your agent's skills directory, then run /wikime in any project to get a fully operational wiki scaffold in seconds.

What it sets up

your-wiki/
├── wiki-agent.md          ← agent operating manual (all instructions live here)
├── CLAUDE.md              ← @wiki-agent.md (or natural-language pointer for other agents)
├── CONVENTIONS.md         ← human-readable naming and structure reference
├── README.md              ← quick start and scripts reference
├── index.md               ← one-liner catalog of all wiki pages
├── log.md                 ← append-only change history
├── {page-type}/           ← primary wiki pages (e.g. papers/, use-cases/, experiments/)
│   └── *.md
├── entities/              ← entity and concept pages
│   └── *.md
├── _templates/
│   ├── {page-type}.md     ← template for new pages
│   └── entity.md          ← template for entity/concept pages
├── sources/               ← immutable raw inputs (never edited after saving)
└── scripts/
    ├── render.py          ← generates wiki.html — single-file reader with eight views
    ├── lint.py            ← structural health check
    ├── query.py           ← frontmatter queries (filter by status, category, tag, etc.)
    └── eval.py            ← LLM-as-judge quality eval (grounding, contradictions, redundancy) with regression gating

Installation

# Clone the repo
git clone https://github.com/Phluxxed/llm-wiki ~/llm-wiki

# Symlink into your agent's skills directory (recommended — keeps it version-controlled)
# Claude Code example:
ln -s ~/llm-wiki ~/.claude/skills/wikime

Each agent has its own skills directory — your agent will know where to look.

Requires pyyaml and markdown for the scripts:

pip3 install pyyaml markdown

scripts/eval.py's claude judge additionally needs claude-agent-sdk (pip3 install claude-agent-sdk); skip it if you run the eval with --judge codex or --judge none.

Usage

In any directory (empty or existing project), run:

/wikime

The skill asks two questions — what the wiki is for, and what the primary page type is — then scaffolds everything in one pass.

Supported page types: use cases, papers, experiments, runbooks, ADRs, or anything else you name.

Safe on existing projects: if a CLAUDE.md, AGENTS.md, or GEMINI.md already exists, the skill appends a single pointer line rather than overwriting it. If wiki-agent.md already exists, the skill stops — you already have a wiki.

Scripts

ScriptCommandOutput
scripts/lint.pypython3 scripts/lint.pyStructural health check — missing sections, broken refs, open risks, index consistency
scripts/query.pypython3 scripts/query.py --helpFrontmatter queries — filter by --status, --category, --type, --tag, --stale, --risks
scripts/render.pypython3 scripts/render.pyGenerates wiki.html — single-file reader (Home, Page, Search, Graph, Risks, Recent changes, Open questions, Entities). Open in a browser or view as a Claude artifact
scripts/eval.pypython3 scripts/eval.py --gateLLM-as-judge quality eval — grounding, cross-page contradictions, redundancy, near-duplicate disambiguation; per-metric thresholds + regression gating (exit code). Auto-detects your agent CLI (claude/codex) as a keyless judge; run records in .eval/

Agent compatibility

Works with any agent that supports skills. The running agent creates its own schema file — a one-line pointer to wiki-agent.md where all instructions live.

AgentSchema filePointer format
Claude CodeCLAUDE.md@wiki-agent.md (native include)
CodexAGENTS.mdNatural-language pointer
Gemini CLIGEMINI.mdNatural-language pointer
Ver o original no GitHub

Bom para

Relacionados