mzyx-skills
Nikitzu/mzyx-skills
Pipeline estruturado de engenharia: brainstorm → spec → plan → build → ship
Ver no GitHubConjunto curado de skills para Claude Code que integra três bibliotecas upstream (superpowers, agent-skills, waza) em um pipeline de engenharia diária. Cobre camadas de processo (brainstorm → plan → execute → verify → review), domínio (TDD, debugging, performance) e disciplina (think, design, check, hunt, research).
Instalação
npx convez-ai Nikitzu/mzyx-skills1 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
- Estruturar fluxo de trabalho de engenharia com comandos slash (brainstorm, spec, plan, build, ship)
- Implementar TDD e verificação automática antes de completar tarefas
- Conduzir revisões de código, auditorias de segurança e testes em paralelo durante deploy
- Root-cause debugging hipótese-first com skill 'hunt'
- Pesquisa estruturada em seis fases e auditoria de configuração do Claude Code
Veredito

Excelente para times de engenharia que buscam automatizar pipeline estruturado. Sem duplicação desnecessária das upstream, mas requer familiaridade com Claude Code para aproveitar toda potência. Recomendado para fluxos rigorosos (TDD, review-gate, ship-pipeline).
README
A tradução para o português deste README ainda está na fila.
mzyx-skills
A curated, trimmed mix of three upstream Claude Code skill libraries —
superpowers,
agent-skills, and
waza — assembled into one plugin tuned for
daily engineering use.
Why
Both upstreams are excellent, but installing both side by side produces heavy duplication (review/plan/test all appear in both, plus alias commands and deprecated shims). This repo cherry-picks one canonical skill per concern and drops everything covered by personal rules files.
Three-layer model:
- Process layer (superpowers). How work flows. Brainstorm → plan → execute → verify → review → finish. Lifecycle scaffolding.
- Domain layer (agent-skills). How to do the work in each phase. TDD, debugging, simplification, performance, shipping. Content checklists.
- Discipline layer (waza). Sharper opinions on think/design/check/hunt
plus standalone utilities for research (
/learn), URL fetch (/read), prose polish (/write), and config audit (/health). Eight skills, each a single habit with a clear trigger.
Slash command flow
The pipeline. Each command invokes one or more skills. Commands ship in .claude/commands/; specialist agents in agents/.
/brainstorm → /spec → /plan → /build → /ship
(sp) (as+sp) (sp) (asks!) (3 agents+sp)
| Command | Skills / agents invoked | Source |
|---|---|---|
/brainstorm | mzyx-skills:brainstorming | sp |
/spec | mzyx-skills:brainstorming (if fuzzy) → mzyx-skills:spec-driven-development | sp + as |
/plan | mzyx-skills:writing-plans | sp |
/build | Asks the user which mode first: parallel teammates / separate session / inline solo. Each mode pairs mzyx-skills:test-driven-development + mzyx-skills:verification-before-completion per task; the parallel mode layers mzyx-skills:using-teams on top. | sp + as |
/ship | Parallel fan-out: code-reviewer + security-auditor + test-engineer agents → mzyx-skills:verification-before-completion → GO / NO-GO + rollback | 3 agents + sp |
Review, TDD, and simplification flow through skill auto-trigger on natural language ("review my changes", "write a failing test", "simplify this") — no wrapper commands needed.
(sp = superpowers, as = agent-skills.)
Discipline-layer skills (waza) ship without command stubs — invoke directly via the namespaced skill name or rely on auto-trigger:
| Skill | When |
|---|---|
mzyx-skills:think | rough idea → decision-complete plan before code |
mzyx-skills:design | UI direction-lock + screenshot iteration |
mzyx-skills:check | solo diff review + verification (lighter than /ship) |
mzyx-skills:hunt | hypothesis-first root-cause debugging |
mzyx-skills:learn | six-phase research → publish workflow |
mzyx-skills:read | fetch URL/PDF as clean Markdown |
mzyx-skills:write | strip AI-pattern prose |
mzyx-skills:health | Claude Code config-stack audit |
Skills also auto-trigger on natural language as in the original repos — e.g. saying "fix this bug" still fires hunt. Commands are explicit verbs for the structured pipeline, not the only entry point.
Lifecycle map
| Stage | Skill | Source |
|---|---|---|
| Session start | using-skills | superpowers |
| Explore intent | brainstorming | superpowers |
| Spec (multi-file, ambiguous) | spec-driven-development | agent-skills |
| Plan (clear-enough) | writing-plans | superpowers |
| Implement (this session) | subagent-driven-development | superpowers |
| Implement (separate session) | executing-plans | superpowers |
| Write code (logic / bug / change) | test-driven-development | agent-skills |
| Hit a bug / unexpected error | hunt | waza |
| UI / frontend | frontend-ui-engineering | agent-skills |
| Refactor / clean up | code-simplification | agent-skills |
| Pre-claim done | verification-before-completion | superpowers |
| Self-review | code-review-and-quality | agent-skills |
| Perf concern | performance-optimization | agent-skills |
| Record decision | documentation-and-adrs | agent-skills |
| Write / edit a skill | writing-skills | superpowers |
| Plan / pressure-test design | think | waza |
| UI direction lock | design | waza |
| Solo review + verify | check | waza |
| Hypothesis-first debug | hunt | waza |
| Polish prose | write | waza |
| Multi-source research | learn | waza |
| Fetch URL / PDF | read | waza |
| Audit Claude Code config | health | waza |
Install
Recommended — install as a plugin
This is the right install path. All skill cross-references in this repo use the mzyx-skills: prefix, which only resolves correctly when the plugin is loaded by name.
/plugin marketplace add github.com/Nikitzu/mzyx-skills
/plugin install mzyx-skills@mzyx-skills
After install, skills appear as mzyx-skills:brainstorming, mzyx-skills:test-driven-development, etc. in the listing, and these slash commands become available:
- Pipeline commands (sp/as):
/brainstorm/spec/plan/build/ship - Discipline skills (wz):
mzyx-skills:think·design·check·hunt·write·learn·read·health(no command stubs — invoke namespaced or rely on auto-trigger)
Alternative — symlink into ~/.claude/skills/
Skills only, no commands. Bare names (no mzyx-skills: prefix) — cross-references inside skills will fail to resolve unless the prefix matches your install path, so prefer the plugin install above.
git clone https://github.com/Nikitzu/mzyx-skills.git ~/Code/mzyx-skills
ln -s ~/Code/mzyx-skills/skills/* ~/.claude/skills/
Updating from upstream
This repo is a curated fork. To pull selective updates from upstream:
git remote add upstream-superpowers https://github.com/obra/superpowers.git
git remote add upstream-agent-skills https://github.com/addyosmani/agent-skills.git
git remote add upstream-waza https://github.com/tw93/waza.git
git fetch upstream-superpowers
git fetch upstream-agent-skills
git fetch upstream-waza
# See what changed in skills you actually use
git log --oneline HEAD..upstream-superpowers/main -- skills/
git log --oneline HEAD..upstream-agent-skills/main -- skills/
git log --oneline HEAD..upstream-waza/main -- skills/
# Cherry-pick selectively (one file at a time keeps drift minimal)
git checkout upstream-superpowers/main -- skills/brainstorming/SKILL.md
git checkout upstream-agent-skills/main -- skills/test-driven-development/SKILL.md
Cherry-pick over merge. A merge would drag in the skills and behaviours explicitly removed here.
Waza re-strip on update. After cherry-picking from upstream-waza, re-run the strip pass to remove the 🥷 emoji preamble and Chinese trigger phrases before committing — the original waza skills carry both, and they have no purpose in this fork.
Excluded — and why
See NOTICE for the full exclusion list.
License
MIT — see LICENSE. All upstream attributions preserved.