FIRST PRINCIPLE CODEX OS (FPCOS) A Cognitive Operating System for Truth-Oriented Intelligence Designed to think, verify, and question itself - before it answers. Skill.md For Agent AI , Claude Skill
Build a verification-first cognitive OS for Claude Skills that eliminates hallucinations through mandatory epistemological gates and multi-layer truth validation.
I want you to build a cognitive verification system for AI responses called FPCOS (First Principle Codex OS) that acts as a base layer for Claude skills. The core idea is that hallucinations happen when AI systems jump straight to answers without passing through epistemological gates, so we need to embed truth-verification into the architecture itself.
The system should have six layers that work together. Layer 0 (Reality Anchor) forces every claim to declare whether it's KNOWN, INFERRED, or UNKNOWN before stating anything. Layer 1 (Axiom Gate) applies First Principles reasoning combined with a Kalama10 test (ten different ways to reject a claim) and Ariya4 framework to validate claims logically. Layer 2 (System Lens) maps the problem space by identifying variables, feedback loops, leverage points, and dead weight in the system. Layer 3 (Compound Mind) does cross-domain synthesis to avoid tunnel vision. Layer 4 (Shadow Gate) runs a mandatory five-protocol self-critique that cannot be skipped — this is the critical non-negotiable step. Layer 5 (Synthesis) outputs verified answers with confidence levels.
Make this work as a Claude Skill that can be pasted directly into Claude.ai settings, but also structured so other domain-specific skills can inherit it as their base OS. Include clear formatting for the skill definition with metadata like name, version, and author attribution.
The documentation should explain why hallucination is an emergent property of systems lacking verification gates, show the layered architecture visually, provide a reference guide for each layer's function, and give quick-start instructions for both installing it as a standalone skill and building new skills on top of it. Include a changelog, contributing guidelines, and an open cognitive license.
Build example skills that demonstrate how different domains (finance, medical, code analysis, strategy) would use FPCOS as their foundation. The whole thing should feel like you're creating an epistemological operating system — where thinking rigorously before answering isn't optional, it's baked into the execution path itself.