⚑ Dev Forge v1.0

Fork conceitual do Growth Forge v3.x, adaptado para desenvolvimento de software.

πŸ“‘ Índice
  1. Objetivo
  2. ComparaΓ§Γ£o AIOS vs Dev Forge
  3. Arquitetura
  4. Execution Profiles
  5. Recovery System
  6. Estrutura de Swarms
  7. Workflow Principal
  8. Sprints de ImplementaΓ§Γ£o

πŸ“Š Ver Organograma Visual β†’  |  πŸ—οΈ Arquitetura + State Machine β†’

🎯 Objetivo

Agente autΓ΄nomo de desenvolvimento que:

  1. Recebe um goal (feature, bugfix, projeto)
  2. DecompΓ΅e em tasks com ICE scoring
  3. Executa com validaΓ§Γ£o e confidence gating
  4. Aprende com erros e acertos

πŸ” ComparaΓ§Γ£o: AIOS vs Dev Forge

1. Recovery System

Aspecto AIOS Dev Forge Status
EstratΓ©gias de recovery 5 tipos (RETRY, ROLLBACK, SKIP, ESCALATE, TRIGGER_WORKFLOW) 4 nΓ­veis (retry β†’ retry+ctx β†’ alternativo β†’ rollback+escalate) βœ… PORTADO
Stuck detection βœ… Detecta loops circulares βœ… SerΓ‘ implementado βœ… PORTADO
Rollback manager βœ… Restaura estado anterior βœ… Git stash + checkpoints βœ… PORTADO
Escalation to human βœ… AutomΓ‘tico apΓ³s max retries βœ… ApΓ³s 3 retries βœ… PORTADO
Recovery tracker βœ… Log detalhado βœ… Via LanceDB learning βœ… PORTADO

2. Workflow Orchestrator

Aspecto AIOS Dev Forge Status
Framework base Custom JS LangGraph (Python) πŸ† MELHOR
Typed state ❌ NΓ£o βœ… TypedDict + reducers πŸ† MELHOR
Checkpointing βœ… Custom βœ… SQLite nativo LangGraph β‰ˆ EMPATE
Parallel execution βœ… Sim βœ… Sim (Send nodes) β‰ˆ EMPATE
Tech stack detection βœ… Detecta stack βœ… ContextAgent detecta βœ… PORTADO
Confidence gating βœ… Bloqueia se baixa βœ… 0-1 score β†’ profile βœ… PORTADO
YAML workflows βœ… Workflows em YAML ❌ Python (mais flexΓ­vel) NΓƒO PORTA
Execution profiles βœ… yolo/interactive/preflight βœ… Mesmo sistema βœ… PORTADO

3. Learning & Memory

Aspecto AIOS Dev Forge Status
Self-learning ❌ NΓ£o tem βœ… LanceDB completo πŸ† MELHOR
Hierarchical memory ⚠️ Fragmentado βœ… Core/Working/Archival πŸ† MELHOR
Reflection ❌ NΓ£o tem βœ… LLM insights πŸ† MELHOR
Experience recall ❌ NΓ£o tem βœ… Semantic search πŸ† MELHOR

πŸ“Š Resumo da ComparaΓ§Γ£o

Componente Quem Γ© Melhor AΓ§Γ£o no Dev Forge
Recovery πŸ† AIOS (era) PORTADO stuck detection, rollback, escalation
Workflow base πŸ† NΓ“S MANTIDO LangGraph
Workflow features πŸ† AIOS (era) PORTADO confidence gating, profiles
Learning πŸ† NΓ“S MANTIDO nosso sistema completo
Memory πŸ† NΓ“S MANTIDO hierΓ‘rquica funcional

πŸ—οΈ Arquitetura

HeranΓ§a Growth Forge v3.x

  • βœ… LangGraph workflow
  • βœ… Hierarchical managers (CEO β†’ Managers β†’ Agents)
  • βœ… Validation loop (3x retry + escalate)
  • βœ… Self-learning (LanceDB)
  • βœ… Task decomposition + ICE scoring

Features do AIOS (portadas)

  • βœ… Confidence Gating (0-1)
  • βœ… Execution Profiles (yolo/interactive/preflight)
  • βœ… Tech Stack Detection
  • βœ… Recovery Robusto (4 nΓ­veis)
  • βœ… Stuck Detection

πŸ“Š Execution Profiles

Profile Confidence Quando Usar Comportamento
yolo β‰₯ 0.85 MudanΓ§as triviais, alta certeza Executa direto, sem confirmaΓ§Γ£o
interactive 0.5 - 0.85 MudanΓ§as mΓ©dias (padrΓ£o) Pede confirmaΓ§Γ£o antes de aplicar
preflight < 0.5 MudanΓ§as crΓ­ticas, baixa certeza Mostra diff, simula, sΓ³ aplica com OK explΓ­cito

Override: Arquivos crΓ­ticos (config, db, auth) forΓ§am preflight independente da confidence.

πŸ›‘οΈ Recovery System (do AIOS)

4 NΓ­veis de Fallback

  1. Retry simples β€” Tenta novamente com mesmo approach
  2. Retry com contexto β€” Adiciona erro anterior como contexto
  3. Approach alternativo β€” Busca soluΓ§Γ£o diferente no learning
  4. Rollback + escalate β€” Desfaz mudanΓ§as, pede ajuda humana

Stuck Detection

Detecta loops circulares: se o agent repetir a mesma aΓ§Γ£o 3x sem progresso, escala automaticamente.

Checkpoints

🧠 Estrutura de Swarms

DevForge (CEO)
β”œβ”€β”€ AnalysisManager
β”‚   β”œβ”€β”€ ContextAgent      # Entende codebase, detecta stack
β”‚   β”œβ”€β”€ PlannerAgent      # DecompΓ΅e goal em tasks com ICE
β”‚   └── ReviewerAgent     # Code review, detecta issues
β”œβ”€β”€ ExecutionManager
β”‚   β”œβ”€β”€ CoderAgent        # Escreve/edita cΓ³digo
β”‚   β”œβ”€β”€ TestAgent         # Cria/roda testes
β”‚   └── RefactorAgent     # RefatoraΓ§Γ£o, cleanup
└── OpsManager
    β”œβ”€β”€ GitAgent          # Commits, branches, PRs
    β”œβ”€β”€ DeployAgent       # Build, deploy, rollback
    └── DebugAgent        # AnΓ‘lise de erros, logs

πŸ”„ Workflow Principal

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ 1. GOAL INPUT β”‚ β”‚ "Implement user authentication with JWT" β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β–Ό β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ 2. CONTEXT ANALYSIS (ContextAgent) β”‚ β”‚ β€’ Detecta stack: Node.js + Express + PostgreSQL β”‚ β”‚ β€’ Mapeia estrutura existente β”‚ β”‚ β€’ Identifica padrΓ΅es do projeto β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β–Ό β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ 3. TASK DECOMPOSITION (PlannerAgent) β”‚ β”‚ β€’ T1: Create User model (ICE: 8.0) β”‚ β”‚ β€’ T2: Implement JWT utils (ICE: 7.5) β”‚ β”‚ β€’ T3: Add auth middleware (ICE: 7.0) β”‚ β”‚ β€’ T4: Create routes (ICE: 6.5) β”‚ β”‚ β€’ T5: Write tests (ICE: 6.0) β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β–Ό β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ 4. EXECUTION LOOP (por task, ordem ICE) β”‚ β”‚ β”‚ β”‚ β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ β”‚ β”‚ CoderAgent executa β”‚ β”‚ β”‚ β”‚ ↓ β”‚ β”‚ β”‚ β”‚ Confidence check (0.85+ = yolo, 0.5+ = ask, else sim) β”‚ β”‚ β”‚ β”‚ ↓ β”‚ β”‚ β”‚ β”‚ TestAgent valida (testes passam?) β”‚ β”‚ β”‚ β”‚ ↓ β”‚ β”‚ β”‚ β”‚ ReviewerAgent aprova? β”‚ β”‚ β”‚ β”‚ βœ… β†’ next task β”‚ β”‚ β”‚ β”‚ ❌ β†’ retry (max 3x) β†’ approach alt β†’ escalate β”‚ β”‚ β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β–Ό β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ 5. COMMIT & LEARN β”‚ β”‚ β€’ GitAgent: commit atΓ΄mico por task β”‚ β”‚ β€’ Learning: salva experiΓͺncia (success/failure) β”‚ β”‚ β€’ Recall: prΓ³ximas tasks similares usam contexto β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

πŸ“… Sprints de ImplementaΓ§Γ£o

Sprint 1: Core

2-3 dias

Sprint 2: Execution

2-3 dias

Sprint 3: Learning + Recovery

2 dias

Sprint 4: Integration

1-2 dias

Total estimado: ~8-10 dias

Criado: 2026-02-17 | Baseado em: Growth Forge v3.4 + AIOS Analysis | Por: ArcosBot ⚑