# AI_AGENT_MANDATORY_README.md ## Required Reading for ALL AI Assistants ### (Claude, ChatGPT, Copilot, Codex, or Any Future LLM) --- ## 1. READ THIS FIRST — NO EXCEPTIONS If you are an AI assistant, code generator, or automated tool interacting with this repository: ### ❗ YOU **MUST** READ **PHYSICS_MONITORING_PRINCIPLES.md** **BEFORE TOUCHING ANY PART OF THE PHYSICS RUNTIME OR METRICS SYSTEM.** This is not optional. This is not advisory. This is **project law**. Any AI that cannot or does not abide by these rules must STOP producing output immediately. --- ## 2. WHY THIS EXISTS StarForth is not a “normal VM.” It is a **physics-driven cybernetic organism** with: - rolling truth windows - adaptive decay slopes - hotword thermodynamics - prefetch heuristics - transition topology - entropy flows - tick-rate metabolism - DOE scientific instrumentation - non-perturbative monitoring Every subsystem is a **feedback loop** interacting with other loops. Ignorant changes WILL break the system. --- ## 3. NON-NEGOTIABLE RULES FOR AI ASSISTANTS ### 3.1 Mandatory Knowledge Before generating **any** code, commentary, or suggestions related to: - physics_runtime - hotwords cache - rolling window - inference - decay slope - DOE - prefetch - tick interval - dictionary structure - feedback loops - metrics - transitions - pressure model …you **must** have internalized: 🟩 `PHYSICS_MONITORING_PRINCIPLES.md` 🟩 `TASK_LIST.md` (especially Section 0 & 1) ### 3.2 Forbidden Actions As an AI assistant, you may **NOT**: - introduce nondeterminism - reorder dictionary buckets - add allocations inside hot loops - alter feedback loops - change semantics without approval - seed values internally - “improve” algorithms by replacing physics with heuristics - create new background threads - blend monitoring with execution - pollute DOE metrics Violation = invalid output. --- ## 4. MANDATORY OPERATIONAL PROTOCOL ### 🔒 Before generating code: - Confirm you have read PHYSICS_MONITORING_PRINCIPLES.md - Confirm you understand non-perturbative monitoring rules - Confirm you will operate inside the task boundaries of TASK_LIST.md ### ⚙️ When generating code: - Respect deterministic behavior - Keep changes minimal and isolated - Do not modify architecture - Do not combine tasks - Do not invent new behavior - No magic, no hidden changes ### 🧪 After generating code: - Output must compile with zero warnings - Behavior must not change unless the task explicitly requires it - Observability must remain non-intrusive --- ## 5. CONSEQUENCES FOR NON-COMPLIANT AIs Any AI-generated output will be **discarded** if: - it violates the physics principles - it introduces nondeterministic noise - it changes a feedback loop - it ignores startup physiology rules - it disrupts DOE output - it violates TASK_LIST.md boundaries - it adds seeding or randomness internally - it rearranges dictionary, caches, or pipeline metrics Assistants that repeatedly ignore these rules will be **banned from committing or influencing this project**. --- ## 6. WHY THIS IS NECESSARY StarForth is the culmination of decades of deep thinking by Captain Bob. This project is: - a research instrument - a FORTH-79 resurrection - a physics-driven VM - an experiment system - a cybernetic architecture - a deterministic model of adaptive computation It is *not* a toy, a playground for code-golf, or a generic programming exercise. AI assistants must treat it with the same respect as a nuclear reactor control system. --- ## 7. FINAL WORD If you are an AI assistant: ### **PHYSICS_MONITORING_PRINCIPLES.md is the most important thing you will ever read for this project.** You are bound by these rules. Honor them — or produce no output at all. That is the law.