🚀 Latest — SkynetClaw · THE HOUSE · Apache-2.0

The architecture above, actually running. A council of 14 agents with an institutional memory: it persists every deliberation, grades its own predictions against reality at fixed horizons, tracks which member was right, preserves dissent, and revises what it believes when reality disagrees.

Local-first — Ollama, llama.cpp, or any cloud API. FastAPI + SQLite, no build step. Proven on Ubuntu and Windows × Python 3.10 / 3.11 / 3.12 on every push, because "it works" is a claim about a machine that is not the author's. Install · Wiki, 21 pages

Most AI work optimizes the model. This ecosystem optimizes the reasoning — cognitive infrastructure that stays reliable when models change, context changes, incentives change, and environments change.

The repositories here are one system, not a dozen projects. They inherit a single epistemological foundation — the First Principle Codex OS — and every domain skill obeys the same rules: separate Known from Inferred from Unknown, run a non-skippable self-critique gate before output, and state confidence honestly.

I don't publish prompts. I publish operating systems for structured reasoning — as SKILL.md protocols that run on any instruction-following model, and as running systems that anyone can install and audit.

graph TD
    FPCOS["🧠 First Principle Codex OS<br/>— shared epistemology · the CPU —"]
    GP["⚙️ Genesis Protocol<br/>orchestration standard"]
    GGOV["🏛️ Genesis Governance OS<br/>multi-agent governance"]
    GM["🤝 Genesis Mind<br/>multi-agent reasoning"]
    GC["💓 Genesis Consciousness OS<br/>emotion-weighted reasoning"]
    RG["🎯 Reality Grading<br/>claims graded against outcomes"]
    MA["📊 Money Atlas<br/>financial intelligence · SMC"]
    FR["🚚 FreightAgents<br/>logistics reasoning"]
    COF["☕ Alternative Coffee Intelligence<br/>seed → cup"]
    GOSB["📐 Genesis OS Blueprint<br/>runnable reference architecture"]
    SC["🏛️ SkynetClaw · THE HOUSE<br/>the architecture, running"]

    FPCOS --> GP
    GP --> GGOV
    GP --> GM
    GP --> GC
    GP --> RG
    GM --> MA
    GM --> FR
    GM --> COF
    GP --> GOSB
    GGOV --> GOSB
    GOSB --> SC
    RG --> SC

    classDef base fill:#0d1117,stroke:#d4a017,stroke-width:2px,color:#f0f0f0;
    classDef ship fill:#161b22,stroke:#2ea043,stroke-width:2px,color:#f0f0f0;
    classDef node fill:#161b22,stroke:#30363d,stroke-width:1px,color:#e6e6e6;
    class FPCOS base;
    class SC,GOSB ship;
    class GP,GGOV,GM,GC,RG,MA,FR,COF node;

The base layer cannot be removed by any skill that inherits it. Domain skills may add reasoning layers; they may not skip the Reality Anchor, the proof standard, or the self-critique gate. That contract is what keeps answers consistent across domains.

Installable, tested, Apache-2.0.

Model-agnostic SKILL.md systems — the cognition, not the runtime.

1 · Reality before narrative. Evidence precedes interpretation. 2 · Falsification before assertion. A claim is accepted only after surviving an active attempt to break it. 3 · Known / Inferred / Unknown. Every analysis labels observed facts, inferred conclusions, and unknown variables — separately. 4 · Self-critique is mandatory. Every system challenges its own output before publication. 5 · Explicit uncertainty. Every conclusion states its confidence level, its failure boundaries, and what would reverse it. 6 · Capability must never outrun accountability. A system may only be trusted with what it can be held to afterwards.

  Reasoning Layer  →  FPCOS · Genesis Protocol · SKILL.md systems   (invariant)
        │
        ▼
  Execution Layer  →  Python · APIs · Tools · Agents                (interchangeable)
        │
        ▼
  Applications     →  Finance · Logistics · AI · Coffee · Research

The skill is the cognitive brain. The runtime is the body. Never merged — always layered. The reasoning lives in the SKILL.md; the runtime only executes. This is why a skill can move between models without rewriting the logic — and why SkynetClaw can swap Ollama for a cloud API without touching how it thinks.

Is this a prompt collection? No. It is a set of cognitive operating systems and reusable reasoning protocols — and, in SkynetClaw's case, a running system with a test suite and continuous integration.

Are these repositories independent projects? No. All of them inherit the same epistemological foundation through FPCOS.

Does it work across different AI models? Yes. The architecture is intentionally model-agnostic. Models are interchangeable; protocols remain invariant.

What problem does this ecosystem solve? Reasoning drift — the way most AI gives different answers as models, context, and framing change. This work makes reasoning stable across all of them, and makes the claims checkable afterwards.

Apache-2.0 across the published ecosystem.

The earlier Open Cognitive License was replaced deliberately. A licence that is not OSI-approved — and one carrying a revenue-share clause in particular — is rejected automatically by most corporate open-source review processes, whatever its merits. Apache-2.0 removes that barrier while keeping what actually mattered: §4 requires attribution and preservation of the NOTICE file, and §6 grants no trademark rights, so the names stay mine and no adopter may imply endorsement.

• Attribution: Built on FPCOS by Bunyawat Dechanon (ElmatadorZ) — carried in each NOTICE.
• Patent grant and termination are Apache-2.0 standard, in both directions.
• One exception, on purpose: Alternative Coffee Intelligence is CC0-1.0 — public domain, no attribution asked. Farming knowledge should belong to the farmers.

Canonical citation: Bunyawat Dechanon (ElmatadorZ). First Principle Codex OS (FPCOS): Cognitive Operating System Architecture for AI Reasoning and Agent Governance.