The avatars are a two-shade gradient with a large white letter, noisy enough to stay near the size the apps make. backend.sh starts an SMP server built from the pinned simplexmq with its [NAMES] resolver on a local registry, and ci_contact claims seeded.testing for its address there; the entries' descriptions now read "CI seeded ..." so the name is also a directory search.
In run 38737084248 the credential scrub made Claude Code 2.1.292 require its whole Bash sandbox (bubblewrap and socat) and force the default permission mode, so every ui call failed. The scrub and bubblewrap are dropped: the PreToolUse guard, which runs in -p sessions, keeps Bash to single ui commands, and now logs each decision outside Claude's directory as evidence. The model is Claude Sonnet 5.5 for the last Claude run the session allows. Third Claude run: [claude-test]
A wrapper in /usr/local/bin runs the helper from the checkout, so Claude's Bash tool finds ui on PATH whatever environment it builds for its shell. First Claude run: [claude-test]
In the app's font they look alike, and run 38729120356 read the descriptions "CI seed group" and "CI seed channel" as "Cl seed ...", failing four checks.
A claude job (environment app-sim-claude, only on workflow_dispatch or a head commit with [claude-test]) starts its own test backend, builds the patched app, runs the scripted onboarding, and gives claude -p (pinned 2.1.292, Claude Haiku 5.5, at most 60 turns and 4 USD) the same search in both layouts to do with the ui helper, reading screenshots. Claude gets Bash, Read and Write only: a PreToolUse hook lets through nothing but single ui commands, credentials are scrubbed from its subprocesses, files are limited to its output directory, no MCP. The job passes when Claude reports CLAUDE-TEST: PASS, and uploads its report, screenshots, transcript and usage (turns, tokens, cost estimated at API prices).
ui do runs several helper commands in one call and stops at the first that fails, so an agent on the runner needs fewer turns for predictable sequences. The helper unsets CLAUDE_CODE_OAUTH_TOKEN first, so nothing it starts inherits the agent's credentials.
After the search in the default layout, the smoke test turns off Settings > Appearance > Reachable app toolbars, checks that the search field moved to the top, and searches again with the same checks, this time confirming the alert with "Search and don't show again".
The app is built with DIRECTORY_SERVICE_LINK replaced by the test directory's address (in the workflow only). After onboarding the smoke test types "ci seed" in the default layout (toolbar at the bottom), confirms the first-search alert, and pages with Show more until #ci_group and #ci_channel are both shown. It then checks, without stopping at the first failure: two pages, no further page offered, the Directory header above the results, each avatar's colour, each row's description and Join label, and the member counts. Checks go to checks.txt and the job summary.
A phrase that is exactly the text now wins over one that only contains it (the alert's Search button under "Search in Directory?", the Directory header next to the Search in Directory row), and words are compared by their joined characters, so OCR splitting "#ci_group" still matches. New commands: ui find prints a text's centre and box; ui avatar measures the profile image left of a list row's name, saves that area, and with a colour checks that it dominates.
Both get a JPEG profile image made as the apps make them (quality 85, scaled down until the data URI is at most 12500 characters), red for ci_group and green for ci_channel. With them the search results no longer fit in one service response (10737 bytes, compressed if needed), so the directory splits them; the probe follows the cursor, prints each page's entries and sizes, and requires each result once, with its whole image, over at least two pages.
A backend job builds the CLI and the directory service without client_library, reused across runs as an artifact named by its sources (plan job, as on the iOS lane). The app job starts a relay and the directory on the preset servers, registers and approves ci_group and ci_channel from a CLI admin, and checks that a search request from a fresh client finds both. The app is not pointed at it yet; its build and desktopTest still run when the backend fails.
Run 38716815535 passed both smoke tests; its last screenshot (09-final) repeated 08-search-typed. The exit trap now takes it only on failure, where it shows the screen the test stopped at. This run also checks that the Gradle cache saved by that run is restored.
Run 38715611090 still stopped at the profile screen: averaging 4x4 blocks and taking a 5x5-block median leaves large bold text in the estimate (the title read as a dark background) and lets the white labels lift a grey disabled button over the threshold. A map coarse enough for titles would swallow 55-pixel buttons, so each word's background is now the median of the rows 2 to 4 pixels above and below its box, read from the 8-bit grayscale screenshot, which also feeds both OCR passes. Checked offline on the screenshots and TSVs of runs 38712854507 and 38714550750: the first screen, the profile screen with its title and grey button, and the branch chat list with the typed text and the directory row all read correctly.
In run 38714550750 the background map judged the white title area dark and the grey "Create profile" button light, so both were dropped, and onboarding stopped at the profile screen on both apps. ImageMagick's Q16 build writes PGM samples as 16 bits unless told -depth 8, while the reader takes one byte per pixel. The reader now also refuses a map without 8-bit samples.
Run 38712854507 showed that the released app (v7.1.0-beta.6) has no search field after onboarding: with no chats its left pane is empty and the centre shows "Talk to someone", while the branch shows the search bar in that state. The chat list is now recognised by "Talk to someone", which both show; the release job stops there (onboarding mode), and the branch run again requires the "Search in Directory" row after typing.
Run 38712854507 got through onboarding on both apps, but on the branch app the typed "ci smoke" was lost: the dark-text pass read it as "ci smokel" with confidence 6, below the cutoff, while the light-text pass read noise ("OKC!") at the same place, and its "i)" split "Search in Directory". The light pass reads the gaps between dark letters as text, and the dark pass reads noise inside the coloured buttons, so neither confidence nor a pass order decides well. A median of the downscaled screenshot gives the background luminance; a word counts only from the dark pass on light backgrounds and from the light pass on darker ones, with no confidence cutoff. Phrases are now built from neighbouring words, so a stray word elsewhere on the same row no longer splits one ("Be" and "free" were).
Run 38712060819 showed that one image for both polarities smears page text: the morphological open treats the white gaps between dark letters as light text. Dark text now comes from the local threshold of run 38707461053, which read the grey and dark onboarding text; light text, such as labels on coloured buttons, from the difference from an open, which leaves no ring at the button edge. The words of both passes are merged by position, keeping the more confident reading.
In run 38711351160 the normalised image had no rings and showed the button labels, grey and dark text all dark on white, but the fixed 80% threshold turned every pixel with over 20% contrast black, so letters merged into blobs ("Be free" read as "Be tires"). With the background now uniformly white, tesseract's own threshold separates the text, so the image goes to it in grayscale.
Run 38707461053 read the grey onboarding text after the local threshold, but not the white labels on the blue buttons: in the negative image the threshold drew a thick ring around each button, and tesseract output nothing inside it. The text image is now the difference of the screenshot from its morphological close and open, so dark and light text both become dark on white and step edges between regions cancel; one tesseract pass instead of two. Each shot keeps that image and tesseract's TSV in UI_OUT/ocr.
In run 38702315040 the app started under Xvfb and showed onboarding, but tesseract's single global threshold, pulled down by the black screen outside the window and the illustration, lost the grey text and the white labels on the blue buttons, so "Create your profile" was never found. A local adaptive threshold (-lat) on the enlarged grayscale image and on its negative turns both into dark text on white. Each shot now also keeps the two images OCR read, in UI_OUT/ocr.
The branch's Kotlin does not compile yet (DirectorySearch.kt and DirectorySearchView.kt import chat.simplex.res.MR instead of chat.simplex.res.*), so the app job stops before the UI steps. A new release job downloads the desktop .deb of the branch's latest release tag for the runner's Ubuntu and runs the same smoke test on it, which exercises Xvfb, openbox, xdotool and the OCR helper independently of the branch build. The smoke test now checks that the typed text appears in the search field and reports whether the Search in Directory row is shown, since releases before directory search don't have it. This run also checks that the core library cache is hit.
First iteration on f/ci-desktop: the core job builds libsimplex.so with scripts/desktop/build-lib-linux.sh on ubuntu-24.04 (GHC 9.6.3, direct-sqlcipher +openssl, no VLC) and caches it by source hash, with the cabal store cached separately. The app job builds :desktop:createDistributable with Corretto 17 and the repo's assets, runs :common:desktopTest, then starts the app under Xvfb and openbox with software rendering and drives onboarding to the chat list and typing in the search field with the ui helper (xdotool input, tesseract OCR to find text), saving numbered screenshots, the text read from them and a step log as artifacts.
