• Split when responsibility can stand alone
• Compose named booleans before decision
5 CLASSIFIED DIRECTIVES LOCKED
Full copy-pasteable AGENTS.md, .cursorrules, and architectural prompts.
Quantum System Instructions & Agent Rules
Generate production-ready AGENTS.md, .cursorrules, and .windsurfrules files calibrated for Cursor, Claude, Windsurf, and Gemini. Custom tuning for token limits, two-stage booleans, and zero-hallucination standards is reserved for Vault members.
How Claude, Gemini, GPT, Grok, & Cursor Handle Chemical X
Not a formal benchmark suite, just patterns I've noticed running real tasks against each of these models with and without the file-splitting rules.
Suffers severe context degradation past 1,000 lines; drops subtle state mutations, hallucinates missing props, and duplicates imports.
Surgical file edits on targeted subcomponent capsules under 100 lines; generates clean, fully-typed diffs in under 2 seconds.
Massive context window gets cluttered by scattered inline styles and un-factored state hooks, inventing phantom variables.
High-speed ingestion of modular composables; zero-hallucination compliance across reactive state slots.
Fails on multi-clause boolean conditionals embedded deep in JSX trees; generates unreachable logic guards.
Flawless control flow evaluation via Stage 1 concept declarations and discriminated state unions.
Rapid reasoning engine stalls when navigating deeply nested prop-drilling trees and uncleaned polling loops.
Instant alignment with Table-of-Contents scaffolding and self-cleaning timer hooks; zero state mutation leaks.