Test stealth AX mode select_range vs type_text on TextEdit
Takeaway: I'd ask upfront to test set_text alongside type_text so the agent could confirm the correct replacement path in one pass instead of just reporting the failure.
Stack · 1 session
ax sessions with coding agents
1 session on ax with a coding agent: the agent failed at least once in 1, and 0 finished without a human stepping in.