[andrewpark-scaleai]

Why

The goal is faster agentenv-framework CI with no test weakened. Two unit tests of a site that never says it's ready each waited out the full 30s WEBSITE_READY_MARGIN_S. That's 60s of unit-tier time. The tests next to them already set the margin to 0.

What

• test_a_site_that_never_says_it_is_ready_fails_the_deploy_naming_why and test_a_site_whose_ready_answers_no_website_readiness_is_never_ready_and_the_deploy_says_so set WEBSITE_READY_MARGIN_S to 0. Together they take 2s instead of 63s.

Coverage

I broke the readiness wait in env_gateway_provider.py eight ways:

• The two tests still fail when a not-ready site counts as ready, when the error leaves out the reason, when a bad readiness answer counts as ready, when there's no deadline, and when the timeout_s argument is ignored.
• They miss changes to the margin itself. So do main's versions of these tests, which never checked how long the wait was. This change loses no coverage.

Verification

• All 266 tests in env_gateway_provider_test.py pass.

Confidence Score: 5/5

The PR appears safe to merge.

Summary

Two never-ready-site tests set WEBSITE_READY_MARGIN_S to zero, removing the extra wait without changing production code.

• Never-ready website tests skip the default readiness margin.

_{Reviews (1) · Last reviewed commit: "test(env_gateway): the never-ready site ..." · Reviewed by Greptile}