Agent-browser iOS Simulator is good enough for Slate local open plus initial snapshot, but not yet post-input proof

What happened

We needed a smaller setup/proof step before committing to broader transport architecture for slate-browser.

The question was whether agent-browser on iOS Simulator Safari was real enough to justify further investment, or whether it was still just transport theater.

What was proved

Against a live local slate-v2 example:

• agent-browser -p ios can open Safari on iOS Simulator
• it can open a local Slate example URL
• it can return the resolved URL
• it can produce an initial interactive snapshot with useful refs

That is enough for:

• transport setup proof
• human-assisted local proof setup
• pre-action artifact capture on iOS Simulator

What did not prove cleanly

The post-input path was still flaky:

• keyboard type was not reliable enough
• keyboard inserttext got further, but post-input snapshot and screenshot were not stable enough to call a true automated proof lane

So the honest read is:

• open + initial snapshot are green
• post-input proof is not green yet

Reusable rule

For slate-browser transport spikes:

• treat agent-browser iOS Simulator as a valid setup/proof transport for opening local examples and capturing initial browser state
• do not call it a stable automated proof backend until post-input artifact capture is reliable

This is good enough for the first setup spike. It is not good enough for the final iOS parity claim.