Thin Soul, Fat Skill: A Token-Efficient Architecture for Production Multi-Agent Systems

Separating agent identity from execution logic - 12.5MB per agent, 75 agents on 16GB, 83% fewer tokens per tool-use interaction. Soul files (30-line YAML) declare agent identity; skill plugins execute deterministically as subprocesses. Skill code is never sent to the LLM, making the architecture both token-efficient and injection-resistant.

Note (2026-05-09): This version bundles English + 中文 in a single PDF (English first, then Chinese), generated directly from the canonical Markdown source files.