Hi all! 👋 I have two small PRs ready for review, both with green checks and rebased onto current `main`: • github.com/OpenHands/software-agent-sdk/pull/XXXX — implements Option A of RFC #4106 (structured

outcome

on

FinishAction

), as the maintainers indicated in the issue. Adds

outcome: "success" | "infeasible"

+

reason

so orchestrators can detect deliberately-failed tasks without parsing free text. Fully backward compatible, 14 new tests. • github.com/OpenHands/software-agent-sdk/pull/4146 — fixes #4105: the visualizer's metrics subtitle now shows per-request token usage alongside the cumulative total (

↑ input 12k (total 54k)

), so running totals are no longer mistaken for request sizes. Display-only, 3 new regression tests. They supersede my earlier #4116/#4112, which I've closed — the originals had a broken PR description check (my fault 😅), now fixed. Happy to address any feedback!

Hey @bitter-spoon-68230 I reviewed the second PR and left some comments. About the first one: the link is broken ( I don't think we have a PR numbered

XXXX

🙂 ). Could you point me to the correct PR? Thanks 🙂

Thank you i saw it, i plan to fix it today 😁

Ok... I believe the PR is not solving your problem 🙂 I left a more precise comment in the PR. See also this issue. github.com/OpenHands/software-agent-sdk/issues/2566

Happy to take it, thanks! Already looked through the branch, surface is small and the rebase onto current main looks manageable. I'll fix the schema/metadata name collision, get it green and open the PR keeping your commits. One thing I'd need your input on: the reviewers asked for an eval impact assessment and what's the usual way to get that run? That seems like the real gating item. Aiming for a rebased branch in the next few days (maybe more because I’ve started an internship in Zurich).

I don’t remember what the reviewer said. But tag me when the Pr is open and we will see. Thanks