@wooden-umbrella-84098 I was looking at github.com/OpenHands/OpenHands/issues/16302#… because I found it interesting. I am concerned about the security implications of importing skills from outside the organization. If it will be a marketplace like plugins, will there be guardrails for what skills should be allowed that an admin controls? disabled_skills won't be enough because new unsafe skills can be made with minimal modifications. Should RBAC be used for scoping skills? How about limiting which tools skills can invoke? If I was the administrator, I'd allow skills that connect to safe(ish) local tools without limit, and scrutinize external-facing tools like MCPs. It could be overengineering, but considering security usually means more planning up front, even if we don't build it right away. I wanted to share a prototype I made around sharing skills github.com/kimjune01/skilldex • skills declare required capabilities • user and organization policy resolve those requirements • only permitted MCP tools are registered • skill instructions never determine their own authority. For a staggered scope, a boolean switch from an administrator that allows internal skill imports and another switch for external skill imports would be a good start. Is there a roadmap doc or discussion that I should look at first before drafting up a spec?

We don't really have RBAC in agent canvas, so I think a lot of this is out of scope 🙂

when you said marketplace i was thinking about fetching resources

gotcha yeah

oh ok so it's not user generated, i misunderstood

Is this doable by splitting marketplaces a bit, into ‘core’ and ‘recommended’, and making the ‘large codebase’ marketplace non-default? I left a little comment to this effect, WDYT?