Master's 3126d9b54 made the iOS app's default private routing "always". Name lookups then go through a preset server, which can't reach the test SMP server on the runner's localhost, so every iOS case that resolves a SimpleX name would fail. The desktop and Android tests already set "unknown" before the first launch; the iOS smoke test now writes the same into the app group's defaults after installing the app, and the app applies it when its chat starts.
The iOS and desktop app jobs saved their stopped test backend (server keys and databases) as a public artifact, so that the Claude job could restart it with the same addresses: anyone signed in to GitHub could download it for a day and act as the test directory or relay while the Claude job ran.
backend-state.sh now packs it as CMS AuthEnvelopedData (AES-256-GCM, RSA-OAEP key transport) to the committed backend-state.crt; the Claude job decrypts it with the environment secret BACKEND_STATE_KEY, given only to that step, so the backend it restarts and the Claude session don't inherit it. Without the secret the Claude job stops at that step, before the Claude session.
The iOS and desktop smoke tests looked for the note "All results on this page are
already in your chats", which the app no longer shows. As nothing marks the end of
a search that adds no row, they wait out the request (it times out after 10 s),
retry once if it failed, and check the listed chat, no result row for it, no Open
and no note. The test plans, prompts and READMEs say the same, and New chat's
results as one group with the matching contacts first.
The smoke test, the device test plan and the Claude prompt check the enable alert (Cancel sends nothing, Enable searches, later searches don't ask), one list with Open and no headers, the Search key with "Connect to …" kept, Open reaching the prepared channel without an alert and with its subscriber count, the load row's count, loading and retry, New chat's card and the long-press menu. The load row's disabled state is read before the screenshot, which takes long enough for the stopped directory's request to fail.
Run: 011-search-redesign, slice 3 (feature CI)
The artifact carrying the app job's stopped test backend to the Claude job holds its databases and server keys, which anyone signed in to GitHub can download for the day it is kept. The READMEs now say so, what someone could do with it, and why encrypting it isn't possible under the Claude job's one-secret rule.
Review R1-1, run 008-dirsearch-known-chats
[app-ci:ios] [ci-backend]
The app is built in the app job with that job's backend's directory address in DIRECTORY_SERVICE_LINK, and since master moved Claude into its own job, that job started a new backend with another address. The app job now uploads its stopped backend's work dir (without the binaries), and the Claude job restarts it with a new backend.sh resume: the same servers' keys and databases, so the same addresses. The relay, directory, names server and contacts start through the functions start uses, split out of it. The iOS Claude job's smoke test also gets BACKEND_DIR, as the app job's does.
[claude-test] [ci-backend]
After joining ci_cards, the smoke test expects the note "All results on this page are already in your chats" for its name, and, in both layouts and in New chat, that a search the local list doesn't match shows ci_cards as its chat row, which opens the channel, not "#ci_cards" with "Join channel". The plan gets U4 and W12 and the new U2, the prompt describes known entries and the note, and the Claude scope runs the changed cases (Haiku 5.5, 300 turns, 3 USD).
Run: 008-dirsearch-known-chats, slice 4
One-hand: the names setup, N1-N11, the New chat cases with names and the full search in both lists, checking that a row opened from goes away; N11 uses #cinews_weekly so the top bar's N1 still finds cinews unopened. Top bar: N1, N4 and the full search on the chat list.
[claude-test] [ci-backend]
A session found no role pickers on the names server it added, so no name resolved and the names cases were blocked (run 38993332306); a later session added the server twice.
One-hand: the names setup, N1-N11, the New chat cases with names, the full search in both lists, G3 again, and whether a row stays next to the chat opened from it; top bar: N1, N4 and the full search on the chat list.
[claude-test:ios] [ci-backend]
D15 said the search bar is above the onboarding pages, which holds only with the top bar; W4 assumed a contact matches. A session tapped the pencil for G3 after the held search had already failed (app log), so the prompt asks for the action as the next command.
One-hand: the chat list, directory results, de-duplication, New chat and regression cases; top bar: the cases this layout draws differently. The names cases run in a separate session.
[claude-test] [ci-backend]
The Claude sessions run the online search plan again: the cases, what each session covers per layout, the prompt's directory search, names and report sections, and the session's start state after the smoke test's searches.
With the test backend, the app is built against the test directory's address, and the smoke test goes on to search it: the Directory header above the results and their avatars in both layouts, the chat list's cards shown while typing and hidden under the results once the profile has a conversation, and "Show more" disabled while its page loads.
A fresh client sends the directory the search the app sends, and must find ci_group and ci_channel; then "ci seed" and "ci avatar" are followed page by page into seed.md, with each page's entries and size against maxServiceBodyLength. The message search stays. The seeded entries name the cases they serve again.
The branch name was part of the jq program text, so a branch named to close the string could select an artifact from a fork or another branch. Same fix as the desktop lookup.
A client started with -s gives that server every role, names included, only to the users that exist when it starts, so the name probe's first connect plan on a new database failed with NO_NAME_SERVERS (run 38904976359). The user is now created by a plain run first, as every other peer is and as desktop's backend does.
