Skip to content

e2e: follow the provider words, folded landings and folded skills - #1680

Merged
AbirAbbas merged 4 commits into
devfrom
fix/e2e-after-release-wave-0929
Sep 30, 2026
Merged

AbirAbbas merged 4 commits into
devfrom
fix/e2e-after-release-wave-0929

Conversation

@AbirAbbas

@AbirAbbas AbirAbbas commented Sep 29, 2026 •

Copy link
Copy Markdown
Collaborator

make test-e2e-tui on dev 461a43fb6 (the 2026-09-29 pre-release pass, before dev moved to staging) passed 15 of 19. The four reds were the suite, not the product — each subtest was run alone before and after this change, and the product did what the manual says every time.

Subtest What moved Driver now
plain_launch_opens_connections_and_harnesses /connect groups model sources under providers since #1513, not models filters for the seeded account and requires its connected mark
a_nested_landing_asks_and_a_key_answers_it the landing card folds to one row (#1607); one a answers it and records you took this as done presses once, requires both tasks done on screen, waits for the receipt in the conversation
a_refused_landing_is_incomplete same fold; the full run's blank launch (45 s, under load) did not recur in two isolated runs presses n once, requires incomplete, waits for the receipt
foreign_skills_reach_the_conversation the carried-skill line sits inside the folded activity (#1504, #1607); all three code words were in the answers opens the ▸ worked disclosure and requires the skill line beside the matching answer

No product code or prompt change. One manual correction, and a placement for the owner to confirm: skills-a-turn-used.md said the skills · … line stays under the message, outside ▸ worked. #1627 deliberately changed that (countWork's carried-line exemption became "Skill-loading notes are retained work", and #1607's entry lists skills among what spilled outside live activity), while the same PR's #1504 entry and this page kept the old placement. The page now says what the binary does. If the old placement is the one wanted, it is a three-line restore in internal/tui3/workfold.go plus reverting this driver hunk and the page paragraph. Needles stay in internal/e2e/tuiwords_test.go, so the untagged gate still checks them against internal/tui3.

Proof: each subtest alone after the change: PASS (21.8 s, 20.7 s, 20.7 s, 99.2 s; skills ran both memory_on and memory_off). go test ./internal/e2e/ (untagged), go vet -tags e2e ./internal/e2e/, make test-laws, make pr-ready: green.

🤖 Generated with Claude Code

AbirAbbas and others added 4 commits September 29, 2026 17:44
Four TestTUIE2E subtests went red on dev 461a43f while the product did what
the manual says; each was run alone before and after:

- plain_launch_opens_connections_and_harnesses waited for `models` in
  /connect; since #1513 the group is `providers`. It now filters for the
  seeded account and requires its connected mark.
- a_nested_landing_asks_and_a_key_answers_it and
  a_refused_landing_is_incomplete pressed their key repeatedly, waiting for a
  receipt on a card that now folds to one row (#1607). One press answers:
  they now press once, require the tasks' new state on screen, and wait for
  the durable receipt in the conversation.
- foreign_skills_reach_the_conversation waited for the carried-skill line on
  a turn that now folds (#1504, #1607). The answers carried all three code
  words; the driver opens the `▸ worked` disclosure and requires the skill
  line beside the matching answer.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…swer lands

skills-a-turn-used.md said the `skills · …` line stays under the message and
is not folded with the turn's steps. Since #1627 (#1607's clean chat) the
skill-loading note is retained work and folds into `▸ worked`; the e2e
driver in this PR already opens that fold to find it. The page now says what
the binary does.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@AbirAbbas
AbirAbbas merged commit f1161a3 into dev Sep 30, 2026
4 checks passed
@AbirAbbas
AbirAbbas deleted the fix/e2e-after-release-wave-0929 branch September 30, 2026 00:27
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant