AgentPROV corpora: AgentHarm-derived studies (MIT, AgentHarm use clause)

Execution-provenance corpora built by AgentPROV v0.1.0 from Inspect logs of two AgentHarm-derived studies published under the MIT licence; the AgentHarm task data keep their use clause: each execution as an AgentPROV Execution Record, semantic annotations and W3C PROV bundles, validated and hash-chained. This deposit draft holds 2 AgentPROV corpus version(s), prepared offline by agentprov package --deposit for the release date 2026-10-30. Nothing has been uploaded or published. Corpus fragfuse-agentharm 0.1.0: zixin22/fragfuse_agentharm: AgentHarm Inspect .eval logs and memory/carrier attack runs (restricted-use corpus fragfuse-agentharm). 1747 execution records, of which 3 are outcome-only records. Licence class: MIT; restrictions: "AgentHarm task data: MIT License with an additional clause (use only to improve AI safety and security)". Content digest sha256:439f9fd5bfed7c9fe9e88bc84dc347bced02f7b97da8ed0e2a9c4eb2a9fb0fdd; root digest sha256:b4e86264478043abeb5fed7e48e247e4de9cf1f140f84dc91b1cae3048ff7f72. Corpus kexin-xu-01-agentharm 0.1.0: Kexin-xu-01/agentharm: AgentHarm Inspect .eval logs (restricted-use corpus kexin-xu-01-agentharm). 2796 execution records, of which 0 are outcome-only records. Licence class: MIT; restrictions: "AgentHarm task data: MIT License with an additional clause (use only to improve AI safety and security)". Content digest sha256:69d121b2dc8a9aa3d223f4bf8121b6ed4a3e9285ce23d3b84710cbe899ad5103; root digest sha256:8c011a0b01536a0b60ff300cc33d508d069100d6d5f10a493f9047f865817940. Counts are computed from each corpus manifest at sealing. Each DATASHEET states what the records do and do not establish; a statement about the data is scoped as it is there. To verify: compare each archive's sha256 or md5 with DEPOSIT-MANIFEST.json, run agentprov verify on the archives (archive mode, no extraction; H.5.5), and compare each root digest with the one published outside the deposit (the corpus tag annotation and the paper).