SAGE System

Supervisor Agent-Generated Engineering — a Claude Code plugin that orchestrates phased parallel software development using fan-out/fan-in lanes, tiered quality gates, and dynamic planning contracts.

What is SAGE?

SAGE is a multi-agent orchestration system for building software systematically. Instead of writing code file-by-file, SAGE decomposes projects into phases, waves, and milestones — then deploys specialized agents in parallel to build them.

• Scope-adaptive planning — Compact (< 10 files), Standard (10-50), or Full (50+)
• Fan-out/fan-in execution — parallel lanes for contracts, logic, tests, and integration
• Tiered quality gates — unit, integration, phase-E2E, and rolling cumulative checks
• Continuous learning — captures patterns and decisions for future projects

Installation

From GitHub

/plugin install gh:Luiz-Frias/sage-system

Local development

claude --plugin-dir /path/to/sage-system

Usage

/sage <project-description>

SAGE will analyze your codebase, recommend a scope profile, generate a phased plan, and execute it with parallel agent lanes — pausing for your approval at key checkpoints.

Documentation

Document	Audience	Content
This README	Humans	Installation, overview, structure
PLUGIN.agents.md	Agents (cold-context entry)	Invocation flow, deployment patterns, gate decisions, recovery playbook
skills/sage/SKILL.md	Supervisor agent	Full orchestration protocol, cardinality rules, reference loading
AGENTS.spec.md	Spec reference	Formal state machine, agent capability matrix

Architecture

Plugin Components

.claude-plugin/
  plugin.json           # Plugin metadata

skills/sage/
  SKILL.md              # Supervisor orchestration protocol
  references/           # Loaded on-demand during execution
    wave-orchestration.md
    agent-coordination.md
    quality-gates.md
    architecture-patterns.md
    learning-system.md
  scripts/              # Validation scripts for Full profile
    validate_generated_plan_contract.py
    validate_agent_native_wrappers.sh
    validate_repo_redesign_contract.sh

agents/
  sage-analyzer.md      # Read-only codebase analysis + scope recommendation
  sage-builder.md       # Per-lane fan-out builder (branch-isolated)
  sage-validator.md     # Tiered gate enforcement
  sage-historian.md     # Pattern capture + decision documentation

Execution Model

              SAGE Supervisor (/sage skill)
                      |
         Phase 1: Analysis (sage-analyzer)
                      |
         Phase 2: Plan Generation
                      |
         Phase 3: Execution Loop
            |                   |
      Fan-out (waves)     Fan-in (merge)
      sage-builder x N    sage-validator
            |                   |
         Phase 4: Gate Validation
                      |
         Phase 5: Learning (sage-historian)
                      |
         Phase 6: Completion

Legacy Reference Directories

These directories predate the plugin architecture but are still used as optional context when present in a target project:

Directory	Purpose
core/	Engine specs, communication protocol, agent registry
domains/	Domain-specific reference (e.g., insurance)
languages/	Language-specific patterns, guardrails, templates
source_documents/	Project requirements, roadmaps, architecture docs
wave_contexts/	Phase/wave execution state and templates

License

Dual-licensed under MIT and Apache-2.0 at your option.

Copyright (c) 2025-2026 Luiz Frias

Contributing

Contributions welcome under the same dual license. Fork, branch, and submit a PR.

Contact

• GitHub Issues for bugs and feature requests
• Email: contact.me@luizfrias.com