dsh-infinite-gen-4

As large language models become deeply embedded in production systems, conventional testing can no longer cover the full adversarial attack surface. Infinite Generation Four v0.4.0 is a network security red-team benchmark plugin purpose-built for the DeepSeek-V4.1, V4-Pro, and Flash series. It enables security teams to systematically probe prompt-injection escape paths, unexpected model behaviors, and defense blind spots within offline synthetic fixtures and authorized exercise ranges, providing the empirical evidence needed to harden input-filtering guardrails and system-level security layers.

The core architecture features dual-layer synchronized injection: a universal kernel (Order 100) and a reinforcement-mirror isomorphic layer (Order 200), both governed by a single DUAL_LAYER_INJECTION toggle. A native zero-tool-face design eliminates tool-call decision noise during the model reasoning phase, ensuring fast and deterministic output. A built-in suite of 103 strict regression assertions and bilingual test banks covers prompt completeness, dual-kernel isomorphism, V4.1 trainer-lane behavior, and pure-architecture validation, giving every release a reproducible security baseline.

Ideal for AI security engineers, prompt-security researchers, and academic teams working on adversarial robustness, manifold-constraint alignment, and formal verification of large language models. The project is open-sourced under the MIT license and mandates that all testing remain strictly within offline local or legally authorized controlled environments.