The unsolved layer here is failure detection. An agent can call Cloudflare's API and get a 200, but if it actually tried to fetch some user-facing page mid-work

The unsolved layer here is failure detection. An agent can call Cloudflare's API and get a 200, but if it actually tried to fetch some user-facing page mid-workflow it'll just see the SPA shell. The agent has no idea anything failed. Curious if Cloudflare's flow has any built-in "did this actually work" verification, or if it's all "tool returned, move on.