The future is not predicted. It is compiled.

Cognitive OS is a universal cognitive operating system specification for AI coding agents. It defines a production-grade, 16-section cognitive architecture that governs how LLM-powered agents think, plan, verify, and execute across real-world systems.

Quick Links

Repository: github.com/aegntic/cognitive-os
Master Spec: AGENTS.md (699 lines, 16 sections)
Full Documentation: README
Architecture: System Diagrams

Supported AI Agent Harnesses

Harness	Config File
Claude Code	`CLAUDE.md`
Cursor	`.cursorrules`
Codex CLI	`CODEX.md`
Gemini CLI	`GEMINI.md`
Factory Droid	`.factory/AGENTS.md`

Key Features

10-Layer Cognitive Stack: Intake triage, reasoning modes, context budgeting, diff-aware editing, tool blindness prevention, semantic search, generated file awareness, phased execution, rollback planning, dependency discipline
Multi-Model Orchestration: Task complexity routing (T0-T4), ensemble verification, cost-aware execution
4-Layer Memory Stack: Working, episodic, semantic, procedural memory layers
Autonomous Verification: Mandatory 7-step verification chain with self-healing pipeline
Swarm Coordination: Sub-agent swarming, worktree isolation, DAG execution
Adversarial Security: Pre-commit checklist, prompt injection defense, circuit breakers
Economic Intelligence: Token cost tracking, ROI-aware routing
Self-Improvement: Continuous learning loop, instinct evolution

Get Started

git clone https://github.com/aegntic/cognitive-os
cd cognitive-os
npm install
npm test

Copy AGENTS.md and your harness adapter into any project. The AI agent auto-detects and loads the spec at session start.

License

Proprietary. See LICENSE.