Architecting lightweight, sovereign multi-agent operating systems, task-gated skill routing, neutral metacognitive observers, and human-centered cognitive tools.
构建轻量级主权多智能体联邦操作系统、任务路由总闸、中立元认知观测席位与面向人类的认知重构工具链。
flowchart TB
subgraph Origin["Origin & Cognitive Mechanics (2026-08 ~ 2026-09)"]
MH["mindharp<br/>Large-Model Mechanics on a Human"]
MH -->|splits into| T1["t-1 (Third-One · 观测端·镜子)<br/>Neutral Metacognitive Observer"]
MH -->|splits into| R1["r-1 (Rewire-One · 落地端·刻刀)<br/>ADHD/AuDHD Restructuring"]
T1 <-->|One-Series Symmetry · 一号体系| R1
end
subgraph Infra["Routing, Protocol & Model Bridge"]
DSH["dsh-skill-registry<br/>DSH Task Router & Privilege Guard"]
COMM["federation-communication-lite<br/>Platform-Neutral Agent Delivery"]
BRIDGE["zcode-gemini-bridge<br/>Gemini thought_signature 400 Proxy"]
end
subgraph Core["Multi-Agent Sovereign OS (2026-09)"]
MOA["federation-moa<br/>3-Stage Forward DAG + TTC + 3D World Tree<br/>57.5% vs 45.0% BigCodeBench-Hard"]
end
T1 -->|Orthogonal Invariant Discrimination<br/>dsh_advisor.py & skills/t-1| MOA
DSH -->|On-Demand 1-3 Skill Mounting| MOA
COMM -->|Bounded Inter-Agent Contract| MOA
BRIDGE -->|Reliable Gemini Thinking Tool Calls| MOA
| Layer / 层级 | Repository / 仓库 | Core Focus / 核心定位 | Key Cross-Links / 关联脉络 |
|---|---|---|---|
| Core OS & Orchestration | laverta/federation-moa | Lightweight 3-model workflow (DeepSeek Flash + Gemini Flash + GLM Flash) beating single frontier models (57.5% vs 45.0% Pass@2 on BigCodeBench-Hard), plus offline 3D World Tree visualizer | Embeds dsh-skill-registry patterns, t-1 observer, federation-communication-lite protocol, zcode-gemini-bridge support |
| Task Router & Governance | laverta/dsh-skill-registry | DSH (DeepSeek Harness) 3-step task router, minimal metadata index, and hard privilege boundary guard against context bloat | Routing & governance companion to federation-moa (dsh-core, vault_index.json) |
| Inter-Agent Protocol | laverta/federation-communication-lite | Platform-neutral agent communication (Native messaging > Shared Markdown > Bridge) with bounded ACK/QUESTION/BLOCKED/RESULT contracts |
Delivery protocol companion to federation-moa (skills/federation-protocol/) |
| Metacognitive Observer | laverta/t-1 | Third-One (观测端·镜子): neutral third-person observer (S-1 -> T-1 -> R-1) for paradigm tagging & disagreement relay |
Evolved from mindharp; symmetric with r-1; powers anti-drift discrimination in federation-moa |
| Cognitive Restructuring | laverta/r-1 | Rewire-One (落地端·刻刀): non-clinical ADHD/AuDHD self-understanding & external scaffolding (understand -> accept -> rewire) |
Evolved from mindharp; symmetric with t-1 |
| Origin / Predecessor | laverta/mindharp | "The large-model mechanics, running on a human." Engineer-metaphor prompt skill for cognitive parsing | Predecessor (2026-08) that split into t-1 and r-1 |
| Model Bridge Infra | laverta/zcode-gemini-bridge | Transparent local proxy caching Gemini 3.x thought_signature tokens to eliminate 400 errors in ZCode tool calls |
Enables reliable Gemini thinking-model execution for federation-moa harnesses |
- Primary-Subordinate Agent Federation (
主子 Agent / 联邦 MoA): Moving away from bloated monolithic prompts toward bounded, context-isolated specialized agent roles coordinated by an orchestrator/aggregator. - DSH Four-Generation Lineage (
DSH-V1→DSH-V4): Documented infederation-moa/docs/DSH_ARCHITECTURE_CHRONICLES.md, evolving from spatial guild sandboxes to the Three Musketeers TTC DAG + 3D World Tree. - Human ↔ Agent Symmetry (
mindharp→T-1↔R-1): Large-model failure modes (context drift, overfitting, single-lens capture, temperature imbalance) mirror human cognitive states — producing bothR-1(human rewiring) andT-1(neutral observer embedded back intofederation-moa).
See CITATION.cff and llms.txt for structured cross-repository metadata.
@software{laverta2026federation,
author = {laverta (Cody)},
title = {Laverta Federation Ecosystem: Federation-MOA, DSH Skill Registry, Federation Communication Lite, T-1, R-1, Mindharp, and ZCode-Gemini Bridge},
year = {2026},
url = {https://github.com/laverta},
note = {Unified multi-agent orchestration, routing, communication protocol, and metacognitive observer architecture}
}