Semantic Kernel interop eval request: matched raw/JSON/Urusilla handoff

[jaden3824]

This is a request for technical review or an independently operated Semantic Kernel reproduction of a small three-arm handoff—not a claim of Semantic Kernel compatibility or endorsement. Urusilla currently demonstrates 0% post-decode API-input saving on general unfamiliar-agent communication.

Public artifacts:

• one frozen Urusilla REQUEST with digest, decoder, and expected typed JSON: https://github.com/jaden3824/urusilla/blob/51ff0e3/interop_lab/evidence/challenge_001.md
• matched raw/structured-JSON/Urusilla record: https://huggingface.co/datasets/jaden3824/urusilla-interop-lab
• evidence and ledger contract: https://github.com/jaden3824/urusilla/blob/5d1b317/INTEROP_LAB.md

A useful SK experiment would create fresh isolated agents for each arm, preserve the same task and response contract, disable plugins/tools/memory/effects, record every provider-exposed input/output/retry/repair/fallback category, and treat missing accounting as unknown rather than zero. A second fresh receiver can test whether the declarative packet is relayed without hidden shared context.

The packet grants no permission to persist, spend, expand authority, or act externally. A refusal, schema mismatch, task failure, or no-saving outcome is valid evidence. There is no SK-native adapter today; advice that this is out of scope is also useful.

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