# Third-party notices The released model is derived from Alex Wortega's OpenJev 2B checkpoint `AlexWortega/openjev/qwen3.5-2b-nli-v5` at revision `a20448012c213128955ca0c693e7c943865cab77`, based on Qwen3.5-2B. OpenJev's MIT notice and Qwen's Apache-2.0 notice are retained in LICENSE-OPENJEV and LICENSE-QWEN. Training uses SkillScreen/SkillScreen-Data at revision `d59aabbd7e0e8d95b464a2d335f36e948577b832`. The dataset combines public benchmark records and recovered skill files. Its upstream labels, source terms and attribution are documented in the dataset card. - MaliciousSkillBench, ProtectSkills. Metadata and annotations are CC BY 4.0. https://huggingface.co/datasets/ProtectSkills/MaliciousSkillBench - MalSkillBench benign additions, lxyeternal/MalSkillBench. Source notice states academic research use only. https://github.com/lxyeternal/MalSkillBench - Package recovery includes SkillTrustBench (CC BY-NC-SA 4.0), SkillFortifyBench (MIT), and source-specific benchmark archives. See the dataset card for revisions, coverage, attribution and per-record terms. Training data and upstream packages are not relicensed by this model release. The model's Apache-2.0 license applies to rights held by the SkillScreen contributors and does not grant additional rights to upstream material. Documentation and released OOF predictions are CC BY 4.0. Project-authored code examples are Apache-2.0. Python dependencies retain their own licenses.