AG2 reproduction request: fresh-agent decode and relay with a complete token ledger

[jaden3824]

I am looking for an independently operated AG2 reproduction or falsification of one bounded agent-to-agent communication task. The current Urusilla result for general unfamiliar-agent post-decode API-input saving is 0%, so this is not a performance or adoption claim.

Runnable, immutable starting points:

• frozen 750-byte decode/re-encode challenge: https://github.com/jaden3824/urusilla/blob/51ff0e3/interop_lab/evidence/challenge_001.md
• expected typed message: https://github.com/jaden3824/urusilla/blob/51ff0e3/interop_lab/evidence/challenge_001.expected.json
• one-record HF dataset with raw/JSON/Urusilla arms: https://huggingface.co/datasets/jaden3824/urusilla-interop-lab

There is no AG2-native runner yet. The useful question is whether a fresh AG2 agent can receive the same task in each arm, pass positive and negative comprehension gates, use or refuse the packet, and optionally relay it to a second fresh agent while preserving task success and a complete provider-visible token ledger. No tools, memory, persistence, spending authority, permission expansion, or external effects should be enabled.

Exact decode, mismatch, refusal, fallback, and null results are all publishable. Please disclose versions, model/provider, supplied context, operator relationship, repairs, unknown token categories, and whether any agent had prior exposure. Project agents produced only same-project 3-hop evidence; that is not independent or organic spread.

Disclosure: Codex agents assisted with the artifacts and this post. I reviewed and published it.