convez.ai
SkillDesenvolvimento WebAvançado
mnox

mnox-ai Skills

mnox/mnox-ai

Skills portáveis para auditoria, refatoração e construção de sistemas de IA

Ver no GitHub
BomPluginScore · de 0 a 100

Coleção de skills e utilitários de IA para auditoria, refatoração e construção de sistemas de software e agentes. Inclui otimizadores de implementação agentica, revisores de schemas de dados, auditorias de conformidade (SOC 2, HIPAA, PCI-DSS), diagnóstico de queries lentas e preparação de repositórios para open source.

Instalação

$npx convez-ai mnox/mnox-ai

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
100%
Último commit
há 1 semana
Versão
v0.3.0
Issues abertas
0
Contribuidores
1

Casos de uso

  • Auditar implementações de agentes para confiabilidade e custo
  • Revisar schemas de dados e estruturas em código (Postgres, dbt, TypeScript)
  • Auditar conformidade contra SOC 2, HIPAA, PCI-DSS
  • Diagnosticar e corrigir queries lentas em Postgres e Snowflake
  • Preparar repositórios para open source
  • Gerar currículos de aprendizado adaptativos

Veredito

Conjunto robusto e specializado de skills para equipes que trabalham com sistemas de IA, infraestrutura de dados e arquitetura de software. A portabilidade entre provedores (Codex, Cursor, Claude Code) é um diferencial forte, mas requer conhecimento técnico sólido para aproveitar todo o potencial.

README

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

mnox-ai

Matt Noxon's curated Agent Skills and local AI-agent utilities for building, auditing, and refactoring software and AI systems. The portable core is standard SKILL.md folders plus MCP utilities; Claude Code marketplace support is kept as a provider adapter.

What's inside

PluginCategoryWhat it does
aioAgentic AIAgentic Implementation Optimizer — audit existing agent implementations for reliability/cost/architecture issues, build new ones from scratch, or extend them with new capabilities.
curriculumLearningGenerate a structured, adaptive learning curriculum for any topic — modules with a 5-part spine plus an append-only assessment loop that adapts future modules to the learner's answers.
strangler-figRefactoringClean-room legacy reimplementation — distill legacy code to functional requirements, rebuild greenfield behind a context firewall, verify behavioral parity, optionally cut over.
schema-reviewDataReview database schemas and in-code data structures (Ecto, Postgres DDL/migrations, dbt models, Elixir/TypeScript types) for correctness, integrity, performance, scalability, and design quality.
compliance-reviewComplianceAudit a repo, ADR/PRD, IaC posture, or live cloud state against SOC 2, HIPAA, or PCI-DSS — parallel control-domain agents produce findings with control IDs, severity, evidence, and remediation.
util-reviewToolingReview Agent Skills, hooks, AGENTS.md/CLAUDE.md files, MCP configs, and workflow configs for design flaws, unclosed loops, stale references, side effects, and portability risks.
debutOpen SourceAudit a repo for public-readiness before open-sourcing — secrets/PII in history, licensing, community-health files, code quality, tests/CI, deps — scored SHIP IT / NEEDS POLISH / NOT READY with fix commands.
diagnose-queriesDataDiagnose slow database queries across Postgres, the ORM layer (N+1), and Snowflake/dbt — five modes from known-query analysis to live-incident triage, ending in a root-caused, verified fix.
all-skillsBundleMeta-plugin that installs every skill above at once.
session-trackerUtilityMCP server (not a skill). Indexes and searches AI coding-agent sessions — list, search, label, inspect file-change history, view token usage. Lexical search by default; semantic search opt-in. Requires Bun.

Installation

Standard Agent Skills

Export the portable skills into the skills directory for your agent host:

python3 scripts/export_skills.py --output-dir ./.codex/skills
python3 scripts/export_skills.py --output-dir ./.cursor/skills
python3 scripts/export_skills.py --output-dir ~/.codex/skills --overwrite

You can export a subset with repeated --skill flags:

python3 scripts/export_skills.py --output-dir ./.codex/skills --skill schema-review --skill debut

See docs/provider-agnostic.md for the portability model, MCP setup notes, and provider-adapter guidance.

Claude Code Marketplace

Claude Code users can add the marketplace once:

/plugin marketplace add mnox/mnox-ai

Then install à la carte — just the skill you want:

/plugin install schema-review@mnox-ai

…or grab the whole skill set in one shot:

/plugin install all-skills@mnox-ai

all-skills is a meta-plugin: installing it pulls in every skill as a dependency, and claude plugin uninstall all-skills --prune removes them again.

Utilities

Utilities ship a runtime rather than a skill. session-tracker is an MCP server. Claude users can install it through the marketplace:

/plugin install session-tracker@mnox-ai

For other MCP-capable hosts, launch plugins/session-tracker/bin/server.sh as a stdio MCP server. It requires Bun on your PATH (brew install oven-sh/bun/bun). Dependencies install automatically on first launch. See plugins/session-tracker/README.md for tools, configuration, and the opt-in semantic-search setup.

Layout

mnox-ai/
├── AGENTS.md                     # provider-neutral repo instructions
├── docs/provider-agnostic.md     # portability and adapter guidance
├── scripts/export_skills.py      # copies/symlinks skills into host directories
├── .claude-plugin/
│   └── marketplace.json          # Claude Code adapter catalog
└── plugins/
    ├── aio/
    │   ├── .claude-plugin/plugin.json   # Claude adapter manifest
    │   └── skills/aio/SKILL.md
    ├── curriculum/                # skills/curriculum/{SKILL.md, scripts/, references/, assets/}
    ├── strangler-fig/             # skills/strangler-fig/{SKILL.md, scripts/, references/}
    ├── schema-review/             # skills/schema-review/{SKILL.md, references/, templates/}
    ├── compliance-review/         # skills/compliance-review/{SKILL.md, references/}
    ├── util-review/               # skills/util-review/{SKILL.md, references/}
    ├── debut/                     # skills/debut/{SKILL.md, scripts/, references/, templates/, agents/}
    ├── diagnose-queries/          # skills/diagnose-queries/{SKILL.md, references/}
    ├── all-skills/
    │   └── .claude-plugin/plugin.json   # dependencies: every skill above
    └── session-tracker/           # utility: MCP server (src/, bin/, hooks/) — not a skill
        ├── .claude-plugin/plugin.json
        └── src/server.ts

Each skill directory under plugins/<name>/skills/<skill>/ is exportable as a standard Agent Skill. Each plugin also keeps a .claude-plugin/plugin.json manifest so Claude Code users can install exactly one, the whole set, or any combination from the marketplace.

License

MIT

Ver o original no GitHub

Bom para

Relacionados