The Problem

The central information overview alone does not solve the management problem. After discovering affected sites, the agent needs to move from knowledge to action: request details from the source, modify the material, perform a check, or initiate another operation.

In a large cluster, it is undesirable for the user or agent to manually select the API and specific features of each site for every operation.

Research Direction

We need to understand what minimum common layer of capabilities will allow the central agent to address an action to a specific site or group of sites while maintaining local responsibility of each project for actual execution.

We should not pre-determine a single universal API for all possible actions. Perhaps a general mechanism for capability discovery and routing to local tools is sufficient.

This is directly related to the site and capability registry, connecting an arbitrary site, and the existing task of distributed authorizations.

For actions whose results can be formally verified, we should take into account the existing task of executable result checks.