[flagged]
Credential scoping handles the "what can the agent do" part, but after running a few agents on my own data for a while I think the harder problem is what they can see. Prompts turned out to be pretty much useless as a…
[flagged]
Credential scoping handles the "what can the agent do" part, but after running a few agents on my own data for a while I think the harder problem is what they can see. Prompts turned out to be pretty much useless as a…