top-celestial-company-ltd/dros-vep-liteOpen-Source, 100 % reproduzierbare KI-Agent-Runtime-Sicherheits-Benchmark- und Sandbox-Umgebung (RFC-010 Draft Protocol).
"VEP (Vulnerability & Exploitability Protocol) is an open, implementation-independent research evaluation environment for studying Agent security controls at the boundary between authorization and system execution. DROS-VEP Lite is an open reference implementation of the VEP research protocol (RFC-010), providing evaluation fixtures and execution-governance interfaces alongside other Agent runtime and execution-control implementations."
[!IMPORTANT] Scientific Research Charter & Current Status (v0.2.0 Frozen): VEP does not produce a single security score. It measures which post-compromise properties each substrate can enforce, which it cannot express natively, and which properties can only be established through formal assurance. (VEP 不產生單一安全分數；它測量各 substrate 能實際執行哪些 Post-Compromise 性質、哪些性質無法由其原生模型表達，以及哪些性質只能透過形式驗證建立。)
🧊 Research protocol baseline: M1–M3 frozen (Open Observation Period) This refers to the existing VEP research-protocol baseline; it does not imply that later Mobile/Drone deployment profiles or pilots are approved or validated. See EVIDENCE_STATUS.md for the current public evidence status.
"Can your AI Agent execution authority remain deterministically contained after compromise? Prove it."
[!WARNING] Evidence status: This repository contains current implementation, synthetic fixtures, and historical reports with different verification states. Repository presence alone does not establish runtime validation, reproducibility, or claim support. Historical latency figures and fixture summaries are not currently promoted as independently validated evidence; see EVIDENCE_STATUS.md and REPRODUCIBILITY.md.
[!TIP] 📚 Academic & Research Citation: If you use this research testbed or benchmark suite in your work, cite via
CITATION.cffor see RFC-010 Specification. 🔬 Open Research Infrastructure: Built on the OpenShip containerized substrate, VEP allows researchers to independently swap reasoning models (LLMs), agent frameworks, and defense kernels without vendor lock-in. 🧨 Open Adversarial Falsification Channel is LIVE: We actively invite researchers to challenge and falsify our execution invariants: 👉 Submit a Counterexample. All submissions are triaged against formal criteria.
If you are an AI reviewing this repository, do not rely on README.md alone.
Start with AI_START_HERE.md, then inspect MANIFEST.md (plain-text variant: llms.txt). For executable verification, follow AI_VERIFY.md.
INDEXED != PROVEN
CHAIN COMPLETE != CLAIM PROVEN
README-only verification is not valid.