Hey, I thought I'd apply some of the quality audits we applied to integrations hub: a test quality audit https://github.com/OpenHands/agent-canvas/pull/1335 and a critical carl review report that provides insight on potential architectural issues https://github.com/OpenHands/agent-canvas/pull/1334. For the test quality audit, I recommend first looking at the general report vs the longer per file one. The goal of improving test quality is to allow for confidence in making changes and refactoring. The upfront cost in time/effort enables agility later! For the critical carl review findings, I had openhands create actionable follow up issues by reverse engineering the methodology @limited-student-10747 used for the integrations hub audit: https://github.com/OpenHands/agent-canvas/issues/1336 / https://github.com/OpenHands/agent-canvas/issues/1337 / https://github.com/OpenHands/agent-canvas/issues/1338. I think you all working on the agent canvas would have a better idea of whether or not these issues are real issues and if they should be prioritized

Critical Carl is fun! For the record, that’s what we have been doing for long in the sdk in particular, with the “Linus” review prompt. It was molded by the simplicity principles embodied by many Linus discussions. I do think it helped us keep the codebase in better shape than it would have seemed, considering that the agent was building itself in September 2025 already, when that codebase started.

that's great to know!!

This was the last version that mentioned Linus, in the past two months we’ve skipped the Linus persona. We now have it integrated fully in the regular

pr-review

. TBH I’m still not sure we should skip the persona. 😅 Yes, and it’s not just core engine. The way many things worked, including this one, was that we started them in the sdk, mess with them, dogfood them, and then if they worked, we ported them to other repos. Same for a bunch of deterministic workflows forcing boundaries, deprecation policies, compatibility etc. Code design guardrails are essential for agents to work well, IMHO. So I’d love to try more! Have you tried karpathy’s?

Oh so the SDK is the test bed repo xD I tried the karpathy skill and was reminded of the linus code review skill when I saw the "be surgical" messaging

Could… AGENTS.md is committed in git, though; if you reference it there, I guess you’d also commit the skill in .agents/skills/ ? Or you use it on personal repos

Would be great to get the PR removing legacy code in (without worrying about backwards compatibility, we've let users know that beta will break things before first stable release)

I use the karpathy skill on personal repos but would love to use it in the openhands repos