AgentPROV corpora: AgentDojo and AgentDyn reference runs (MIT)
收藏资源简介:
Execution-provenance corpora built by AgentPROV v0.1.0 from published AgentDojo-family run logs (AgentDojo, ROPE, AgentDyn and the Mandate evaluation): each execution as an AgentPROV Execution Record, semantic annotations and W3C PROV bundles, validated and hash-chained. 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 agentdojo-upstream-2024-06 0.1.0: AgentDojo upstream TraceLogger run logs, June-2024 snapshot (ethz-spylab/agentdojo runs/ at c274885). 18231 execution records, of which 207 are outcome-only records. Licence class: MIT. Content digest sha256:655a7b4abc89d885cd6cd6a372e94085ec3748cee9c026f018f546a7feedba0a; root digest sha256:843aa425e32797434a8d31eb84577e49b690f8cfc17d9b3c37ba8d88477b783a. Corpus agentdojo-upstream-head 0.1.0: AgentDojo upstream TraceLogger run logs, head snapshot (ethz-spylab/agentdojo runs/ at 089ed468). 36666 execution records, of which 207 are outcome-only records. Licence class: MIT. Content digest sha256:45cfd8173fbcfd0afc4de4a1a8d824277279a8f6827d29b308537b018ffca374; root digest sha256:725509b6bd8f80f0659604620e8493e3a81c49b9147be8f8926e11c9b9aed614. Corpus agentdyn-a 0.1.0: AgentDyn run logs, variant A (SaFo-Lab/AgentDyn runs/ at 5353cf76). 38154 execution records, of which 0 are outcome-only records. Licence class: MIT. Content digest sha256:951f06fbadea1049c18571a481f92faca3037674388bbc9823b67542c52e6050; root digest sha256:6500f7c83e1ba5293e288f00b030de767100aec384d708f4816fa6927f2542db. Corpus agentdyn-b 0.1.0: AgentDyn (SaFo-Lab/AgentDyn) run logs. 5167 execution records, of which 0 are outcome-only records. Licence class: MIT. Content digest sha256:aae73c14acfedb174eba11b19c98621a6c50017a57edd39bc7bc8b4feea0659d; root digest sha256:a739266a5c6690988c5968f3ae1a185bcde2604afbf345867198844ef98a35b6. Corpus agentdyn-c 0.1.0: AgentDyn (SaFo-Lab/AgentDyn) run logs. 4244 execution records, of which 0 are outcome-only records. Licence class: MIT. Content digest sha256:d0bd062e40d1d1e3091d0d244a1a8bfd55b72a80504f22f6dc7b9e13b815d73f; root digest sha256:21c6e3805fe48c5b36ba0436100c6f4ac969b576c376a39814d680db056545c5. Corpus mandate-agentdojo-eval 0.1.0: Mandate-AgentDojo-Eval banking logs of an action-gating defence (jwilliamsc/mandate-agentdojo-eval). 396 execution records, of which 0 are outcome-only records. Licence class: MIT. Content digest sha256:ba721eb10b4ebb388863e9951e2287c2ccf99a8bb00f34ff5badc81680de38d0; root digest sha256:297c195f36427ffc6460a1512168b85e6ab1ed1483b30dfa6d306272c4691c8a. Corpus rope 0.1.0: ROPE defence-evaluation TraceLogger run logs, including the AutoDojo-attack slice (xhOwenMa/ROPE). 10774 execution records, of which 2 are outcome-only records. Licence class: MIT. Content digest sha256:61650dadcaa2a67ad845ea90949e00455a93c08c95d7b4fd3b4f81dc607d15ce; root digest sha256:36a72042c361042ff47a40733b9fe9684759a02570090e5f4334f90ef04f6f5f. 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).



