A Hermes Agent skill repository for the Five-Layer World Collaboration Model: a fractal intelligence architecture that maps AI systems from small execution units to multi-node intelligent worlds.
Core idea:
Do not solve every AI system problem by adding more agents. Place each capability at the lowest sufficient layer, require evidence from lower layers, and promote reusable patterns upward only when they prove stable.
FIVE-LAYER WORLD COLLABORATION
Direction / governance / memory / promotion flows downward
Evidence / traces / validation / health flows upward
┌─────────────────────────────────────────────────────────────────────────────┐
│ 5. Cosmic / 宇宙层 │
│ multi-device, multi-node, multi-world collaboration │
│ "Do we need universal brain + multi-node hands?" │
└──────────────────────────────────────▲──────────────────────────────────────┘
│ evidence / boundary feedback
┌──────────────────────────────────────┴──────────────────────────────────────┐
│ 4. Planetary / 地球层 │
│ one coherent intelligent world: workspace, gateway, memory, docs, cron │
│ "Does this runtime world run coherently?" │
└──────────────────────────────────────▲──────────────────────────────────────┘
│ health / registry / canonical docs
┌──────────────────────────────────────┴──────────────────────────────────────┐
│ 3. Societal / 社会层 │
│ multi-agent society: L0/L1/L2 roles, handoffs, reviews, ownership │
│ "Are coordination and accountability clear?" │
└──────────────────────────────────────▲──────────────────────────────────────┘
│ review evidence / handoff state
┌──────────────────────────────────────┴──────────────────────────────────────┐
│ 2. Biological / 生物层 │
│ single agent/app as lifeform: role, memory, tools, state, health │
│ "Does this agent have stable role and growth loop?" │
└──────────────────────────────────────▲──────────────────────────────────────┘
│ tool results / traces
┌──────────────────────────────────────┴──────────────────────────────────────┐
│ 1. Cellular / 细胞层 │
│ JSON action executors, validators, policy gates, adapters, scripts │
│ "Can structured intent become safe observable evidence?" │
└─────────────────────────────────────────────────────────────────────────────┘
Cellular execution pattern from JSON Tool Runtime:
human/event input -> LLM strict JSON -> schema validation -> policy gate
-> deterministic executor adapter -> observable evidence
| Layer | Name | Unit | Runtime question |
|---|---|---|---|
| 1 | Cellular / 细胞层 | JSON action executor, tool adapter, script, API, cron, log, artifact, function call | Can this structured intention be executed reliably and produce observable evidence? |
| 2 | Biological / 生物层 | single agent as lifeform | Does this agent have role/DNA, memory, tools, state, health, and growth? |
| 3 | Societal / 社会层 | multi-agent society | Are L0/L1/L2 roles, handoffs, review, and ownership clear? |
| 4 | Planetary / 地球层 | one intelligent instance/world | Do workspace, gateway, memory, docs, cron, skills, and registries run coherently? |
| 5 | Cosmic / 宇宙层 | multi-device / multi-node / multi-world network | Does this need universal brain + multi-node hands, and are boundaries/feed-back loops clear? |
.
├── README.md
├── LICENSE
├── .gitmodules
├── scripts/
│ ├── install.sh
│ └── validate_skills.py
└── skills/
├── five-layer-world-collaboration/
│ └── SKILL.md
└── llm-json-tool-runtime-skill/ # git submodule
└── skills/
└── llm-json-tool-runtime/
├── SKILL.md
└── references/
This repository contains the Five-Layer World Collaboration skill directly.
llm-json-tool-runtime-skill is included as a git submodule because it is a reusable lower-level cellular execution pattern:
LLM -> JSON -> validator -> executor adapter
Relationship:
- Five-Layer World Collaboration: macro architecture / world model.
- LLM JSON Tool Runtime: cellular execution unit pattern for turning model JSON into validated side effects or evidence.
Clone with submodules:
git clone --recurse-submodules https://github.com/MarsDoge/five-layer-world-collaboration-skill.gitIf already cloned:
git submodule update --init --recursiveInstall both skills:
./scripts/install.shThis installs:
skills/five-layer-world-collaboration
-> ~/.hermes/skills/autonomous-ai-agents/five-layer-world-collaboration
skills/llm-json-tool-runtime-skill/skills/llm-json-tool-runtime
-> ~/.hermes/skills/software-development/llm-json-tool-runtime
Then restart Hermes or run /reload-skills if available.
/skill five-layer-world-collaboration
/skill llm-json-tool-runtime
Or start Hermes with the skills preloaded:
hermes -s five-layer-world-collaboration -s llm-json-tool-runtimepython3 scripts/validate_skills.pyExpected output includes:
OK skills/five-layer-world-collaboration/SKILL.md five-layer-world-collaboration
OK skills/llm-json-tool-runtime-skill/skills/llm-json-tool-runtime/SKILL.md llm-json-tool-runtime
The submodule is intentionally not flattened or copied into this repository. The canonical source for JSON Tool Runtime remains:
https://github.com/MarsDoge/llm-json-tool-runtime-skill
Do not rely on symlinks for installation. Use scripts/install.sh so Hermes receives real skill directories at the correct category paths.
To update the submodule later:
git submodule update --remote skills/llm-json-tool-runtime-skill
git add skills/llm-json-tool-runtime-skill
git commit -m "Update llm-json-tool-runtime submodule"The Five-Layer skill includes a longer portable model note:
skills/five-layer-world-collaboration/references/five-layer-model.md
Use it when writing public docs, explaining the model, or deciding whether a capability belongs at the cellular, biological, societal, planetary, or cosmic layer.
This repo is a skill package, not a full runtime implementation. It teaches agents how to reason about architecture placement, promotion paths, evidence, memory, and multi-layer collaboration.
The LLM JSON Tool Runtime submodule provides the lower-level pattern for typed JSON action execution.
MIT