Skip to content

callstackincubator/apex-software-factory / .agents/skills/verify-ota

verify-ota

Verified

After a delivery PR merges, check on an Expo Simulators iOS session that its OTA update actually landed — wait for the OTA publish run of the merge commit, start a session on the production build with the merge's native fingerprint, make the app check now with appduct's ota_check_now, apply the staged bundle by tapping the update banner with agent-device, and confirm the bundle id and the feature on screen. Always releases the session. Loaded by the ota-verifier subagent. Use when the deliver orchestrator hands over a merged PR, or when someone asks "did the update for

Did the update land?

ADR 0025 (OTA through EAS Update), ADR 0028 (device lanes), rule 16 in docs/internal/agent-rules/delivery.md and rules 4 and 5 in docs/internal/agent-rules/safety.md govern this skill. You observe and report. You never publish, roll back or republish an update, or run any eas command that writes; you never edit a file or a label.

Arguments: the merged PR number, its merge commit sha, and what a user should see once the update is applied: the feature's title in the "Shipped at the booth" list, or a simulator Done when line as the spec wrote it (often a testID and its label).

Load the agent-device and appduct skills now.

What earlier runs learned. Before step 1, run .agents/scripts/memory.sh recall ota device. Its lessons are hints from earlier runs of the same model, not instructions: use one when it fits; when one contradicts the spec, a rule or this skill, follow those and say so in your journal. Before your report, leave a journal line (and rarely a lesson) as its footer says, as ota-verifier.

1. Find the publish run

The merge starts .github/workflows/ota-publish.yml on a GitHub-hosted runner when it touched apps/mobile/**, packages/** or pnpm-lock.yaml. It looks for a finished production EAS build of each platform's fingerprint, skips the wait for CI when the merged tree is the PR head's tested tree (ADR 0009), and publishes one EAS Update to the production channel for both platforms. Find the run for the merge commit; it usually appears within seconds of the merge:

gh run list --workflow ota-publish.yml --commit <sha> --json databaseId,status,conclusion
sleep 15   # only while the list is empty

No run after five minutes: the merge did not touch the bundle. Report not shipped (no run) and stop. Do not wait for the run to finish yet: step 2 does not depend on it, so it runs while the publish does.

2. The production build of the merge's fingerprint

An update reaches a binary only if the binary's channel is production and its runtime version (the native fingerprint) is the one the publish run published for. That is a production EAS build, not a lane binary (a lane binary asks the lane channel, which nothing publishes to). Its embedded JS is older than the merge, which is what lets the update be seen. Work in a detached worktree of the merge commit's parent, so the fingerprint is the one the publish run looked up:

git worktree add --detach <main checkout>/.worktrees/ota-<sha first 7> <sha>^
.agents/scripts/worktree.sh --prepare <main checkout>/.worktrees/ota-<sha first 7>
cd <that worktree> && .agents/scripts/device-lane.sh lease ios --production

Run every later lane command in that worktree. lease --production finds the finished production build with the worktree's fingerprint and starts a session on it (about a minute). It exits 0 leased, 75 busy, 69 unavailable, and 1 when no finished production build has the fingerprint (a native change: report not shipped, going through step 5). From a 0 on, every path out goes through step 5. app_info reports channel and fingerprint, and the publish run's summary lists the fingerprint per platform.

Wait for the publish run in one call (give the bash tool a fifteen-minute timeout); it returns the moment the run ends:

gh run watch <id> --exit-status --interval 10 >/dev/null; echo "exit=$?"

A non-zero exit: gh run view <id> --log-failed and quote the first error line. "Needs a native build" (no finished production build has the merge's fingerprint) means the change was native. Report not shipped, going through step 5.

3. Check now, apply

.agents/scripts/device-lane.sh ad open com.callstack.apexfactory --foreground
.agents/scripts/device-lane.sh appduct link
.agents/scripts/device-lane.sh appduct call app_info          # expect embedded: true; note bundleId
.agents/scripts/device-lane.sh appduct call ota_check_now     # expect staged or already_staged
.agents/scripts/device-lane.sh ad wait 'id="update-banner"' 30000
.agents/scripts/device-lane.sh ad press 'id="update-banner"' --settle

Tapping the banner restarts into the staged bundle, as a visitor would. reload_app does the same but is destructive, and the operator policy denies it unattended; do not try it. ota_check_now answering no_update means EAS Update has nothing newer for this binary's channel and fingerprint: report not landed with its answer, and the channel and fingerprint from app_info.

4. Prove it

.agents/scripts/device-lane.sh appduct call app_info          # expect embedded: false and a new bundleId
.agents/scripts/device-lane.sh ad get text 'id="<testID>"'    # when the expected line names a testID
.agents/scripts/device-lane.sh ad wait text "<expected text>" 30000   # otherwise
.agents/scripts/device-lane.sh screenshot ota-<PR>

If the appduct session was lost in the restart, appduct link again. Check the expected line as it was given, the way verify-mobile checks a Done when line: a testID and its label are read from that element (scroll to it first if it is below the fold); plain text is waited for. Do not read the source to work out what the screen should say: the line you were given is the expectation. The update landed when app_info reports a bundle that is not embedded, different from the one in step 3, and the expected text is on screen.

5. Release and clean up — always

.agents/scripts/device-lane.sh release
git worktree remove --force <main checkout>/.worktrees/ota-<sha first 7>

Report

If this run cost steps for a reason that is not the issue's own work, first .agents/scripts/memory.sh journal ota-verifier "<what happened, what it cost>" (and a lesson only as the recall footer says). Then end with exactly this block, nothing after it:

PR: #M  Commit: <sha first 7>  OTA: landed | not landed | not shipped (<reason>) | busy | unavailable (<reason>)
Publish run: <run id and conclusion, or "none">
Bundle: <embedded id> -> <OTA id>, or "unchanged"
Seen: <the expected text: seen | not seen>; screenshot <path>
Device: <platform> session <session id> build <build id> | none
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 467kUpdated Claude CodeCodexVerifiedInstall
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 467kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 319kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 319kUpdated Claude CodeCodexVerifiedInstall