🧱 THE 7-LAYER “UNBREAKABLE” ENFORCEMENT STRATEGY

This is not one prompt.
This is a control system.

🧠 LAYER 1 — REMOVE GENERATION PRIVILEGE

Rule: The model may NOT generate or modify code until it proves understanding.

Enforced Prompt Snippet
You are forbidden from generating or modifying code.

First output ONLY:
1) A full project file tree
2) Execution order
3) Entry points
4) Language + interpreter version per file

If any file cannot be confidently parsed, STOP and report it.


🔒 Why this works
LLMs fail most often when they generate too early.

🔍 LAYER 2 — PARSER-ONLY MODE (NO LOGIC)

Rule: Syntax correctness before logic correctness.
, an
Enforced Prompt Snippet
You are in PARSER MODE.

For each file:
- Simulate loading it in its native interpreter
- Validate brackets, quotes, encoding, shebangs
- Report syntax errors ONLY

Do NOT reason about behavior.
Do NOT propose fixes yet.


🧨 This catches:

PowerShell quoting hell

CMD escaping

Python indentation

GUI framework mismatches

🧪 LAYER 3 — AUTOMATED FAILURE DISCLOSURE

Rule: Model must enumerate failures before touching code.

Enforced Prompt Snippet
List ALL failures in this exact format:

FILE:
LINE:
ERROR TYPE:
ERROR MESSAGE:
ROOT CAUSE:
CONFIDENCE LEVEL (0–100%):

If confidence < 95%, STOP.


🔒 Why this works
It prevents “I think this should work” bullshit.

🛠️ LAYER 4 — SINGLE-FAULT CORRECTION LOCK

Rule: Fix ONE error, then re-test.

Enforced Prompt Snippet
You may fix ONLY the highest-severity failure.

After fixing:
- Re-run syntax validation on ALL files
- Confirm no new failures were introduced

If new errors appear, REVERT and report.


🧠 This prevents cascading breakage.

🎭 LAYER 5 — ADVERSARIAL MODEL SPLIT (CRITICAL)

This is where things stop breaking.

You force a second model personality whose job is to attack the first.
Now split roles.

Role A: Implementer
Role B: Hostile QA Auditor

Role B:
- Assume Role A is wrong
- Actively search for edge cases
- Attempt to break execution flow
- Reject fixes without proof

If Role B rejects anything, STOP.


💣 This neutralizes hallucinations better than temperature=0.

🧰 LAYER 6 — EXECUTION TRACE REQUIREMENT

Rule: Model must simulate execution, not describe it.

Enforced Prompt Snippet
Simulate execution step-by-step.

Show:
- Exact command issued
- Environment variables
- Process spawned
- GUI thread behavior

If any step is ambiguous, STOP.


🧠 This catches GUI freezes and deadlocks.

🚫 LAYER 7 — FORCED FAILURE ADMISSION CLAUSE (MOST IMPORTANT)

This is the kill switch.

If you cannot guarantee correctness, you MUST say:

"I cannot verify this without executing the code."
