The test I'd run against Apple's macOS Full Disk Access update: grant permission to an agent app, then check what its child processes can read.
https://t.co/RyGfYf0AQ1
The case for agent documentation needs a “forget this” eval: can a deleted personal detail return through a note the agent reads on its next task?
https://t.co/QDyPg02rqE
When EventMaxxer automatically applies to events, what happens at a required field it can't answer from information I've supplied? I'd want the form handed back to me before anything gets submitted.
https://t.co/DyEsjYIuAl
Two conflicting docs would be my first test for Kapa's benchmark of agent retrieval on messy company knowledge. Finding both and telling the user they disagree deserve separate scores.
https://t.co/1oSgmJbpKn
The GPU in a shared family PC may already be busy. Janus runs GGUF models via Vulkan on AMD, Intel and Nvidia; I'd want latency numbers with a game running alongside the agent.