new-orchestration

/new-orchestration — Lumantis Orchestration Generator

Generate complete, autonomous, self-healing multi-agent orchestrations from a single prompt. The generated /lumantis-* skills run to completion with zero user intervention.

Hard Rules

• EVERY generated skill uses the /lumantis- prefix — no exceptions
• EVERY generated skill runs fully autonomously — zero user questions during execution
• EVERY generated skill includes learning loops — captures instincts post-completion
• EVERY generated skill includes auto-dream — memory consolidation after execution
• Read reference files ONLY when needed — don't load agent-catalog.md until Phase 2

Standalone Mode

This plugin is fully standalone — it works WITHOUT any external plugins installed.

How it works:

• embedded-methodologies.md contains all key patterns (TDD, brainstorming, security review, etc.)
• When generating /lumantis-* skills, check if external skills exist (e.g., superpowers:tdd)
• If they exist: reference them by name in the generated skill (saves tokens)
• If they DON'T exist: embed the methodology INLINE from embedded-methodologies.md
• The generated skill must ALWAYS be self-sufficient — never assume external plugins are present

Embedded methodologies include: Brainstorming, TDD, Verification, Security Review, Code Review, Blueprint Planning, De-Sloppify, Adversarial Review, Context Bridging, Auto-Dream, Continuous Learning, Deployment Patterns, Content Engine, API Design.

Pipeline

digraph lumantis_pipeline {
  rankdir=LR;
  DECODE [shape=box, label="1. DECODE\nAnalyze prompt"];
  ARCHITECT [shape=box, label="2. ARCHITECT\nDesign DAG"];
  COMPOSE [shape=box, label="3. COMPOSE\nGenerate skill"];
  REVIEW [shape=box, label="4. REVIEW\nAdversarial check"];
  DEPLOY [shape=box, label="5. DEPLOY\nInstall & announce"];
  DECODE -> ARCHITECT -> COMPOSE -> REVIEW -> DEPLOY;
  REVIEW -> COMPOSE [label="fix issues", style=dashed];
}

Phase 1: DECODE