e2e: five TestTUIE2E needles rot on clean dev #1014

replication

On a clean origin/dev tree (d906707, #1012):

make build
go test -tags e2e -count=1 -timeout 30m ./internal/e2e/ -run 'TestTUIE2E/(one_figure_on_every_spend_surface|a_nested_landing_asks_and_a_key_answers_it|a_refused_landing_is_incomplete|a_refused_task_proposal_draws_no_schema_sentence|space_in_the_task_room_pages_the_card)'

All five FAIL on the clean tree, ~6 min, a few cents. Full suite run (make test-e2e-tui) fails the same five and passes twelve.

needles that never land (20–30s waits each)

• one_figure_on_every_spend_surface — spendws (tui_e2e_test.go:1570)
• a_nested_landing_asks_and_a_key_answers_it — nobody could check it / [a] accept [s] tell it / your call / ? (tui_e2e_test.go:255)
• a_refused_landing_is_incomplete — nobody could check it / [n] not right (tui_e2e_test.go:307)
• a_refused_task_proposal_draws_no_schema_sentence — 1 yes, set it up · 0 no · c change (refusedargs_e2e_test.go:58)
• space_in_the_task_room_pages_the_card — enter open its room / · waiting in this conversation · alt+a then new conversation (tui_e2e_test.go:1928, :1960)

evidence

Two full runs (branch with unrelated session-economy fixes, and clean baseline) both fail the same five; twelve pass in both. This is the #184 class again — needles rotting against the surface. Filed from a public CI-reachable assertion; needs OPENROUTER_API_KEY or a profile key per liveKey.

acceptance

The five subtests pass on a clean origin/dev tree end to end, either because the needles track the surface again or because the surface is repaired.

Drafted with CodeAF · reviewed and owned by the author