preflight

name	preflight
description	Full pre-submission inventory — git, build numbers, archives, screenshot counts and sizes per platform, leftover processes — then the slow gates to run by hand. Use before submitting to the App Store, cutting a release, or answering "is anything else needed?".
disable-model-invocation	true

Preflight

python3 .claude/skills/preflight/check.py

Exit 0 when every check passes, 1 on any FAIL. Add --json for machine output.

Why this exists

"Is anything else needed?" was answered from the DMG alone. The iPad screenshot tab was never opened, and a platform went to review with one stale screenshot — twice. The question is a request to re-inventory every surface, and that is a checklist, not a judgement call. Anything a script can decide should not be left to whoever is looking.

What it checks, and why each one is there

Check	The incident behind it
Working tree clean · nothing unpushed	Work committed but never pushed, twice
One build number, one marketing version	16 targets; a partial bump ships a mismatched archive
Archives match the project	An archive built before the bump uploads the wrong build
Screenshots per platform, at exact pixel sizes	iPad shipped with 1; the 11-inch iPad's 1668×2420 is rejected outright
Raw captures present and tracked	The originals were lost to a scratchpad; composites are one-way
No test hosts left running	xcodebuild test injects into the app; hosts survive and look like the app
Mac app not running	run-tests.sh refuses rather than test on top of it
No simulators left booted	Leftovers confuse the next run and hold the app open

The screenshot check reads actual pixel dimensions with sips, not filenames — a file called iphone-6.5.png at the wrong size passes every check that trusts its name.

What it deliberately does not do

The slow gates are named, not run — they take minutes and one of them needs the user's app closed:

./scripts/run-tests.sh                                     # 374 app tests, 46 suites
swift test --package-path Packages/NotesEditor             # 401
cd Packages/NotesEditor && xcodebuild test -scheme NotesEditor-Package \
  -destination 'platform=iOS Simulator,name=HN-iPad'       # 381 — THREE summaries, sum them
scripts/check-download-page.sh                             # live page vs what latest serves

And one thing no script here can do: open every screenshot tab in App Store Connect, per platform. 0 of 10 and 1 of 10 read identically in a glance at the page; only the count distinguishes them. Screenshots also cannot be edited at all while a version is Waiting for Review, so a mistake there costs a removal from review and a resubmission — inventory before submitting, not after.

Extending it

Add checks to check.py via add(status, name, detail). Use FAIL only for things that must block a submission; WARN for things worth seeing that legitimately vary (a staged screenshot set mid-work, an app you left open on purpose). A check that cries wolf gets ignored, and an ignored check is worse than no check — the same reason a guard needs a negative control before you trust it.