claude-engineering-kit
nic611/claude-engineering-kit
My daily Claude Code engineering setup — agents, skills (BDD generator, code review, verification gate), commands, rules, hooks
Instalação
npx convez-ai nic611/claude-engineering-kit1 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
README
A tradução para o português deste README ainda está na fila.
claude-engineering-kit
A practical Claude Code engineering kit — the agents, skills, commands, rules and hooks I use daily. Grown from real enterprise frontend work: React 16→18 migrations, BDD testing in a banking environment, and AI-assisted workflow design.
Companion repo: nic-opencode-config — the OpenCode equivalent of this setup (my team environment runs OpenCode; I run Claude Code personally, and keep both portable).
What's inside
| Asset | What it does |
|---|---|
agents/ | Subagent definitions: explorer, reviewer, tdd-agent, refactor-agent, build-fixer |
skills/bdd-generator | Generates Cucumber/Gherkin BDD suites from component code |
skills/code-review | Structured review workflow with severity-tagged findings |
skills/skill-creator | Meta-skill: scaffolds and evaluates new skills |
skills/smart-reflect, note-connect, weekly-review | Personal knowledge-vault loop: capture learnings as they happen, link notes, weekly review |
skills/verification-before-completion | Forces evidence-based "done" — no claim without a passing check |
skills/webapp-testing | Browser-level testing workflow with example scripts |
commands/ | /tdd, /daily-update, /daily-patch slash commands |
rules/ | Auto-loaded coding standards: code style, testing, security, git; effort & planning conventions |
hooks/ | Session memory persistence + strategic context compaction |
gpt-prompts/ | Two-phase enterprise Copilot workflow (Part 1 review → Part 2 generate) for locked-down environments |
Design principles
- Spec first, generation second — every workflow writes the rules down before letting the model execute.
- Verification is a gate, not a vibe —
verification-before-completionblocks "done" without evidence. - Portable across assistants — same mental model runs on Claude Code (personal) and OpenCode/Copilot (enterprise).
Usage
Drop the folders you want into your project's .claude/ directory, or symlink the whole kit:
git clone https://github.com/nic611/claude-engineering-kit.git
cp -R claude-engineering-kit/{agents,commands,rules,hooks,skills} your-project/.claude/
MIT licensed. Issues and PRs welcome.