convez.ai
AgenteDevOps & InfraestruturaAvançado
plwp

Chief Wiggum

plwp/chief-wiggum

Orquestração de SDLC completo: requisitos → arquitetura → TDD → revisão → ship

Ver no GitHub
InicialPluginScore · de 0 a 100

Agent que orquestra o ciclo completo de desenvolvimento de software (SDLC) para Claude Code, transformando tickets vagos em um fluxo disciplinado de captura de requisitos, planejamento, arquitetura, implementação TDD, revisão estruturada e entrega via PR. Integra múltiplos modelos de IA, análise estática, testes e validação antes do merge.

Instalação

$npx convez-ai plwp/chief-wiggum

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
2
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 dias
Versão
—
Issues abertas
34
Contribuidores
2

Casos de uso

  • Transformar gravações de reuniões em requisitos estruturados
  • Planejar epics com análise de dependências e riscos de integração
  • Implementar tickets com loop TDD e consulta multi-modelo (Codex, Gemini, Opus)
  • Gerar PRs com contexto arquitetural e artefatos de suporte
  • Executar quality gates em nível de epic com testes de mutação

Veredito

Ferramenta robusta para equipes que buscam rigor em fluxos de IA coding, com orquestração explícita de contratos, verificação independente e traceabilidade. Exige configuração não-trivial e entendimento profundo de workflows de engenharia, mas entrega valor real em projetos que demandam qualidade e segurança no delivery.

README

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

Chief Wiggum

Agentic SDLC orchestration for Claude Code. It turns vague tickets into a disciplined delivery loop: requirements capture, epic planning, architecture, test-first implementation, structured review, and PR-ready output.

Why This Exists

  • Most AI coding workflows optimise for fast patches, not for system quality, traceability, or safe delivery.
  • Chief Wiggum adds explicit contracts, multi-model consultation, worktree isolation, independent verification, and GitHub-integrated shipping steps.
  • The goal is not "let the model code unsupervised". The goal is a repeatable engineering workflow that still holds up under review.

Core Capabilities

  • Epic planning: group issues into execution waves with dependency and integration-risk analysis
  • Architecture: define contracts, invariants, state machines, ADRs, and integration tests before implementation
  • Implementation: run a TDD loop with Codex, Gemini, and Opus as inputs rather than a single-model guess
  • Verification: require static analysis, test execution, browser validation, and structured review before shipping
  • Shipping: generate PRs with architecture context and supporting artifacts

Quick Start

# 1. Clone and verify
cd ~/repos/chief-wiggum
claude /setup

# 2. Add as skill source to your target project
# In your-project/.claude/settings.local.json:
{
  "commandDirs": ["~/repos/chief-wiggum/.claude/commands"]
}

# If chief-wiggum is not checked out at ~/repos/chief-wiggum:
export CHIEF_WIGGUM_HOME=/path/to/chief-wiggum

# 3. Use from your target project directory (not chief-wiggum)
claude /transcribe ~/recordings/client-call.mp4
claude /plan-epic owner/repo
claude /implement owner/repo#42

Harness-portable skills are stored under skills/. To install the Claude interactive delegate skill in Codex:

ln -sfn ~/repos/chief-wiggum/skills/claude-interactive-delegate ~/.codex/skills/claude-interactive-delegate
python3 ~/.codex/skills/claude-interactive-delegate/scripts/claude_delegate.py start

Important: Run skills from your target project directory, not from chief-wiggum itself.

Skills

Product Level

SkillPurpose
/designProduct design stage: divergent rendered mockups → human choice → extracted tokens in docs/design/

Epic Level

SkillPurpose
/plan-epicGroup related issues into an epic with dependency graph and integration risks
/architectDefine contracts, invariants, state machines, ADRs, and integration tests for an epic
/implement-waveImplement an epic in dependency-ordered parallel waves
/close-epicEpic-level quality gate: integration tests, mutation testing, stitch-audit, retrospective

Ticket Level

SkillPurpose
/implementTDD implementation loop: test-first → multi-AI consultation → structured review → verify

Supporting

SkillPurpose
/setupVerify and install all dependencies
/transcribeWhisper transcription → structured requirements
/seedArchitecture brainstorm and issue seeding for new projects
/create-issueCreate well-structured GitHub issues
/shipPR creation with mermaid architecture diagrams
/stitch-auditCross-layer data flow analysis
/updateRefresh AI model IDs and library versions

Portable Skills

SkillPurpose
$claude-interactive-delegateDelegate bounded agent tasks to a persistent interactive Claude Code tmux session

Pipeline

%%{init: {'theme': 'base', 'themeVariables': {'primaryColor': '#003f5c', 'primaryTextColor': '#fff', 'primaryBorderColor': '#2f4b7c', 'secondaryColor': '#665191', 'tertiaryColor': '#a05195', 'lineColor': '#2f4b7c', 'textColor': '#333'}}}%%
graph TD
    subgraph "Input"
        A["/transcribe"]:::entry
        B["/seed"]:::entry
        C["/create-issue"]:::default
        D["Requirements / Issues"]:::default
    end

    subgraph "Product Design"
        N["/design"]:::new
    end

    subgraph "Epic Flow"
        E["/plan-epic"]:::modified
        F["/architect"]:::new
        G["/implement<br/>(per ticket)"]:::modified
        W["/implement-wave<br/>(parallel)"]:::modified
        H["/close-epic"]:::new
    end

    A --> D
    B --> C
    B --> E
    B --> N
    N --> E
    C --> E
    E --> F
    N -.->|"docs/design/"| F
    F --> G
    F --> W
    G --> G
    G --> H
    W --> H

    classDef entry fill:#ff7c43,stroke:#ffa600,color:#fff
    classDef default fill:#003f5c,stroke:#2f4b7c,color:#fff
    classDef modified fill:#665191,stroke:#a05195,color:#fff
    classDef new fill:#d45087,stroke:#f95d6a,color:#fff

/implement — Orchestration Detail

%%{init: {'theme': 'dark'}}%%
sequenceDiagram
    participant U as User
    participant O as Orchestrator
    participant AI as Codex / Gemini / Opus
    participant S as Sonnet (worktree)

    U->>O: /implement owner/repo#42
    O->>O: Resolve paths, load epic context

    rect rgba(102, 81, 145, 0.25)
        note right of O: Step 4 — Multi-AI consultation
        par Approach gathering
            O->>AI: Codex consultation
            O->>AI: Gemini consultation
            O->>AI: Opus exploration
        end
        AI-->>O: Three approaches
        O->>AI: Opus reconciliation (+ epic contracts)
        AI-->>O: Implementation plan
    end

    O-->>U: Approach summary (checkpoint)

    rect rgba(212, 80, 135, 0.25)
        note right of O: Step 5 — Test-first specification
        O->>S: Write failing tests (TDD red phase)
        S-->>O: Tests written, all failing
    end

    rect rgba(212, 80, 135, 0.25)
        note right of O: Step 6 — Implementation
        O->>S: Make tests pass + enforce contracts
        S->>S: Code, lint, fix
        S-->>O: All tests green
    end

    rect rgba(102, 81, 145, 0.25)
        note right of O: Step 7 — Structured review
        par Review (with checklist)
            O->>AI: Codex review
            O->>AI: Gemini review
        end
        AI-->>O: Checklist scorecard + findings
    end

    O->>O: Static analysis gate
    O->>O: Apply fixes & verify independently
    O->>O: Verify contract enforcement
    O->>O: Browser-use / E2E validation
    O->>O: Create PR with mermaid diagrams
    O-->>U: PR link

Architecture

%%{init: {'theme': 'base', 'themeVariables': {'primaryColor': '#003f5c', 'primaryTextColor': '#fff', 'primaryBorderColor': '#2f4b7c', 'secondaryColor': '#665191', 'tertiaryColor': '#a05195', 'lineColor': '#2f4b7c', 'textColor': '#333'}}}%%
graph TD
    subgraph "Chief Wiggum"
        Skills[".claude/commands/"]:::entry
        Scripts["scripts/"]:::modified
        Templates["templates/"]:::modified
    end

    subgraph "AI Backends"
        Codex["OpenAI Codex"]:::existing
        Gemini["Google Gemini"]:::existing
        Opus["Claude Opus"]:::existing
    end

    subgraph "Target Repo (worktree)"
        Code["Source code"]:::new
        Tests["Test suite"]:::new
        BU["Browser-use / E2E"]:::new
    end

    subgraph "Infrastructure"
        GH["GitHub CLI"]:::existing
        Keyring["System Keyring"]:::existing
        Whisper["Whisper"]:::existing
    end

    Skills --> Scripts
    Skills --> Templates
    Scripts -->|consult_ai.py| Codex
    Scripts -->|consult_ai.py| Gemini
    Scripts -->|sub-agent| Opus
    Scripts -->|repo.py| GH
    Scripts -->|keychain.py| Keyring
    Scripts -->|transcribe| Whisper
    Scripts -->|implement in| Code
    Code --> Tests
    Code --> BU

    classDef entry fill:#ff7c43,stroke:#ffa600,color:#fff
    classDef existing fill:#003f5c,stroke:#2f4b7c,color:#fff
    classDef modified fill:#665191,stroke:#a05195,color:#fff
    classDef new fill:#d45087,stroke:#f95d6a,color:#fff

Requirements

  • Python >= 3.11
  • GitHub CLI (gh) and git for the core workflow
  • Optional provider CLIs: Claude Code (claude), OpenAI Codex (codex), Google Gemini (gemini)
  • Optional delegate support: tmux for $claude-interactive-delegate
  • Optional transcription support: ffmpeg, openai-whisper
  • Secrets stored in system keyring (managed via python3 scripts/keychain.py)

Check only the profiles you intend to use:

python3 scripts/check_deps.py --for core
python3 scripts/check_deps.py --for core --provider claude-interactive
python3 scripts/check_deps.py --for core --provider codex --provider gemini
python3 scripts/check_deps.py --for transcription
python3 scripts/check_deps.py --for browser-validation

Provider Roles

AI backends are configured by role in config/providers.json. Workflows can still call a provider directly:

python3 scripts/consult_ai.py codex prompt.md -o response.md

They can also consult a role quorum:

python3 scripts/consult_ai.py --role reviewer prompt.md --output-dir ~/.chief-wiggum/tmp/reviews

Role config controls which providers are required or optional. Optional providers can fail or be disabled without blocking the quorum; required providers must be enabled and return successfully.

Ver o original no GitHub

Bom para

Relacionados