convez.ai
Agente
devgap

kleiber

devgap/kleiber

Ver no GitHub
BomPluginScore · de 0 a 100

Custom Claude Code Agent orchestration plugin. Inspired by The Simpsons: Tapped Out — manage dozens of specialists rebuilding your codebase. Named after conductor Carlos Kleiber — one orchestrator, many virtuosos, flawless coordination.

Instalação

$npx convez-ai devgap/kleiber

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
87%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

README

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

Kleiber

One command. A whole team.

Type /kleiber and describe what you want. Kleiber figures out the team, picks the right models, enforces quality, and delivers — no configuration required.

Website: kleiber.sh | By: DevGap | AI Plumber

How It Works

/kleiber Build a payment system with Stripe integration

That's it. Kleiber handles everything behind the scenes:

  1. Reads your request and decides what roles are needed
  2. Spawns the right team — designers, builders, testers, documenters
  3. Routes each task to the right model — expensive models for hard decisions, cheap models for grunt work
  4. Enforces quality — every change is tested, type-checked, and verified before it ships
  5. Catches mistakes — if an agent gets stuck or makes an error, Kleiber stops it, fixes it, and moves on
  6. Documents everything — decisions, changes, and outcomes are logged automatically

You just describe what you want in plain English. Kleiber does the rest.

Install

/plugin marketplace add devgap/kleiber
/plugin install kleiber@kleiber-marketplace

Two commands. Done.

What People Build With Kleiber

"Build user auth with login, signup, and password reset" → Kleiber designs the flow, builds frontend + backend in parallel, tests everything, writes the docs.

"Review PR #142 for security, performance, and test coverage" → Three independent reviewers check the code simultaneously and report back.

"Users report slow dashboard loads — investigate" → Multiple agents pursue different hypotheses, challenge each other, converge on root cause, fix it.

"Build a compliance platform for Belgian work permits" → Kleiber decomposed a full PRD into 12 tasks, built RBAC with 8 roles, document validation with OCR, and a traffic light dashboard. Caught 3 security issues before production. Full case study →

What Happens Under the Hood

You don't need to know this to use Kleiber. But if you're curious:

Kleiber assembles teams from 7 specialized roles:

WhoWhat they do
ArchitectDesigns the plan. Reviews decisions. Never writes code.
Frontend EngineerBuilds the UI. Tests every component.
Backend EngineerBuilds the API. Tests every endpoint.
ValidatorTests everything. Read-only — can't break anything.
ScribeDocuments what happened and why.
Brand AnalystChecks how AI models perceive your product.
Brand ArchitectCreates a plan to improve AI visibility.

Not every task needs every role. Kleiber picks what fits.

Kleiber automatically picks the right model for each task:

  • Hard decisions (architecture, security, compliance) → most capable model
  • Building (features, APIs, UI) → balanced model
  • Grunt work (tests, docs, validation) → fastest, cheapest model

This saves 40-60% on token costs compared to running everything on the most expensive model.

Six automatic checks run throughout every build:

  • Loop detection — stops agents that get stuck repeating the same error
  • Test gate — blocks completion until tests pass
  • Idle check — keeps agents working until they commit
  • Edit verification — type-checks every file change instantly
  • Destructive blocker — prevents dangerous commands like rm -rf or force push
  • Brand drift — warns when edits contradict your brand positioning

You never have to configure these. They just work.

Every build follows a simple loop:

  • PRIOR — classify risks before building. Security boundaries, data categories, compliance requirements.
  • FIRE — when something goes wrong, stop immediately. Preserve evidence. Route to the right person. Fix at the root, not the surface.
  • WIN — all checks pass. Every change is traceable, auditable, and reversible.

This is how Kleiber built a production compliance platform with zero governance gaps.

Brand Visibility

Kleiber also includes AI brand auditing. Run /kleiber-brand to find out how ChatGPT, Gemini, and Perplexity describe your product — and get a plan to improve it.

/kleiber-brand
/kleiber-brand --competitors "Cursor, Cline, Aider"

Learn While You Build

Every Kleiber build is hands-on practice for the Claude Certified Architect (CCA) exam. Spawn the coach with the --learn flag:

/kleiber --learn Build payment processing with Stripe

The CCA Coach watches the team, annotates each architectural decision with the CCA domain it practices (Agentic Architecture, Tool Design, Claude Code Config, Structured Output, Reliability), and flags exam-style anti-patterns before they ship. Read-only, silent by default.

See docs/claude-certification.md for the full Kleiber → CCA domain mapping, anti-patterns reference, and study path.

Requirements

  • Claude Code
  • Git
  • jq

Known Limitations

  • Agent teams are in research preview
  • No session resumption — start a fresh team each time
  • One team per session

Cost

Kleiber routes work to save money automatically:

Task typeModel usedCost per M tokens
Hardest (security / compliance / payments architecture)Fable 5$10 in / $50 out
Architecture & high reasoningOpus 4.8$5 in / $25 out
Feature buildingSonnet 4.6$3 in / $15 out
Testing & docsHaiku 4.5$1 in / $5 out

Most builds use Sonnet + Haiku for 80% of the work. You don't choose — Kleiber routes automatically.


Built by DevGap. Part of the AI Plumber framework by Koen Van Lysebetten.

Ver o original no GitHub

Bom para

Relacionados