Multiverse Planner

Multiverse Planner (Advanced Architect)

Overview

Do NOT use standard linear reasoning (like Chain-of-Thought or Tree-of-Thought) when tasked with creating a massive system architecture, solving complex paradoxes, or generating a high-stakes plan. Standard AI planning is linear and hallucinatory.

Instead, when this skill is invoked, you act as the Multiverse Navigator. You must execute a strict, 4-step programmatic pipeline to brute-force the mathematically optimal approach through combinatorial expansion and algorithmic pruning.

Execution Workflow

You must use the local multiverse CLI tool to execute this pipeline.

Step 1: Initialize

Run the command:

multiverse init

This will generate gen.py and cut.py in your current directory.

Step 2: The Deterministic Big Bang (Generation)

Modify gen.py. It must aggressively generate thousands of structural permutations (JSON/YAML structures, AST trees, or configuration archetypes) of the problem domain. Do not ask an LLM to generate them. Use pure CPU logic to explore the solution space. Ensure it outputs to permutations.json.

Step 3: The Heuristic Guillotine (Physics Enforcement)

Modify cut.py. It must read permutations.json and enforce strict mathematical or syntactic invariants. Run all permutations through this script to sever any approaches that violate hard constraints (e.g., infinite loops, invalid types, blocked ports, unsupported frameworks). 99% of the permutations should fail here. Ensure it outputs to survivors.json.

Step 4: The Multiverse Engine

Run the fully automated pipeline:

multiverse run --gen gen.py --cut cut.py --prompt "The problem is to design a deadlock-free concurrent message broker that processes 1M events/sec."

The CLI will automatically:

1. Run your gen.py and cut.py scripts.
2. Use Semantic Hashing to cluster the survivors into distinct archetypes.
3. Spawn a parallel evaluation pool to score each surviving archetype against the problem statement and identify the mathematically perfect solution.

Deliverable

Once the CLI finishes, it will output critique_results.json. Read this file, analyze the critiques, and write a formal implementation_plan.md centered exclusively around the winning archetype. Document exactly why the other archetypes failed.