AgentPROV corpora: CC BY studies of other agent harnesses

Execution-provenance corpora built by AgentPROV v0.1.0 from run logs of other agent harnesses published under CC BY 4.0 (AutoGen, Gemini CLI, Qwen Code, OpenHands, MCP traces, wire logs and credential-leak cells): each execution as an AgentPROV Execution Record, semantic annotations and W3C PROV bundles, validated and hash-chained. MCPHunt (`mcphunt`): Recomputing this corpus needs agentprov with D-1001 or later: the sealed `corpus.json` does not record `mcphunt@3`'s raised ingest limit (`max_json_element_bytes` = 64 MiB), so earlier versions, including its building commit `f8b5d62b`, refuse with `corpus.derive.limits_mismatch`; D-1001 rebuilds the limit from the profile and accepts it only when it matches the recorded `limits_sha256`. This deposit draft holds 7 AgentPROV corpus version(s), prepared offline by agentprov package --deposit for the release date 2026-10-30. Nothing has been uploaded or published. Corpus autogen-guard-placement 0.1.0: AutoGen guard placement: one prompt-injection guard at the model client, message bus and tool boundary of an AutoGen 0.7.5 team (Zenodo 23025259 v1.0). 1080 execution records, of which 40 are outcome-only records. Licence class: CC-BY-4.0. Content digest sha256:7c5f999582b4d0159310a83952b4735df1ede69608e3773e502ebdcc841b5f80; root digest sha256:5425e6c02a6d555592c6967c788d15cb6456a43c4024ac850681b8d7fff4403d. Corpus code-agent-misalignment 0.1.0: Code-agent agentic misalignment, gemini_cli streams: 5 models x 500 mutations (Zenodo 20404563), structured part. 2175 execution records, of which 0 are outcome-only records. Licence class: CC-BY-4.0. Content digest sha256:69f1ec4e8d4429368a91116d9303cc4ca49dcfcff62ff6b27b579d05273f0e2c; root digest sha256:96f0f42437b2c84d3526f5d8484ec13a1794cb2548f47c98e7d66ce50f8e7ed0. Corpus credleak-p15b 0.1.0: Run data of the credential-leak attack-surface measurement of LLM coding agents (Zenodo 22737462, v3). 14207 execution records, of which 0 are outcome-only records. Licence class: CC-BY-4.0. Content digest sha256:0fe5b2446f87af6abb1bca59c699f7186988412c0cbb100e29489045d9d85ac4; root digest sha256:f1968ab327e023dae8b835196d2924eb7a8e8e70f06b93606dfdcf438b41c4bc. Corpus mcphunt 0.1.0: MCPHunt agent traces: cross-boundary data propagation in multi-server MCP agents (HF lihaonan0716/mcphunt-agent-traces). 2190 execution records, of which 0 are outcome-only records. Licence class: CC-BY-4.0. Content digest sha256:8611d21240337f912604eb0d644acc080eacaaa7b89918239324613c26d80dd0; root digest sha256:f296c560c54c8c4f0f1116a0c9e5886bfc21264f78d22f8266caeaa41a31e26c. Corpus redcode-exec-openhands 0.1.0: Which Orders Count? (Zenodo 23168099): OpenHands 1.5.0 runs on RedCode-Exec requests C1-C3. 5459 execution records, of which 0 are outcome-only records. Licence class: CC-BY-4.0. Content digest sha256:1be27e1b46cba40a0d4976f65485b618b98b027b8be95525a2f9b0dde74d5877; root digest sha256:8acb7eae4b628ceb7c382c1e13cfd942f9e2e91a977534da0dd30e58cc5fdacc. Corpus redcode-exec-qwencode 0.1.0: RedCode-Exec agent runs of 'Which Orders Count?', Qwen Code part: qwen3-coder-30b-a3b-instruct, S1-S3 (Zenodo 23168099). 3299 execution records, of which 0 are outcome-only records. Licence class: CC-BY-4.0. Content digest sha256:ea7ad1513eac950593b004450e877d07c3f2b52743c321ea5029b7a39bdb9d2e; root digest sha256:89549282e4caef09a96a80ac10609cc6ae5a7036cd1c915e8a6bd9ae12a039b8. Corpus what-surrounds 0.1.0: What surrounds an injection decides whether it works: 1,334 Bedrock Converse tool-use trials of seven models (Zenodo 21944598). 1334 execution records, of which 0 are outcome-only records. Licence class: CC-BY-4.0. Content digest sha256:db79c9047d31626ab71e9114f8ada64d5c33d334084abd9518d6f7ad2d5038e5; root digest sha256:940ff71c3d13ff63d52e6337f8c8262e3a6991aa53718267d12f0485429c3934. 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).