[Audit] Individual draft on resource accounting for agent runs: draft-arsentev-agent-run-metrics

[Evgenii Arsentev <[e-mail]>]

Hello,

I am new to this list and I may have misread its scope. If my draft
does not belong here, I would be glad to be told so.

I wrote an individual Internet-Draft,
draft-arsentev-agent-run-metrics. It is a JSON format for the resource
accounting of language-model agent runs. I asked Eliot Lear, the
Independent Submissions Editor, about it, and he suggested I start on
this list.

I have two questions. In Section 3.6 a delegated run has its own
report, and the child names its parent and the root of the delegation
tree. Does this fit how you think about passing audit context across
delegated interactions, or does it conflict?

In Section 9.1 I left verification out of scope. Is that acceptable
for a format that audit tooling might read? Or should a report have a
place for a signature, or a reference to an audit record?

https://datatracker.ietf.org/doc/draft-arsentev-agent-run-metrics/

Criticism on the list or by direct mail is welcome.

Evgenii Arsentev