chore(mobile): read the app version from the tag instead of committing it

[Cawlumm]

The release version lived in two places — mobile/app.json, bumped by hand before each tag, and the tag itself. CI papered over the gap by rewriting app.json during the build, so the committed value was simultaneously required by process and ignored in practice. A ritual whose only failure mode is silent.

mobile/app.config.js now reads the version from LYFTR_VERSION, which the release job sets from the tag. The tag becomes the only place a release version exists: nothing to bump, nothing to drift, and the stamp step is deleted rather than made more careful.

module.exports = ({ config }) => ({
  ...config,
  version: process.env.LYFTR_VERSION || config.version,
})

Changes

• mobile/app.config.js — new; the only thing it overrides is version. Everything static stays in app.json.
• expo.version becomes an explicit dev placeholder, 0.0.0-dev, used when LYFTR_VERSION is unset: local expo start, dev builds, dry runs. Never what ships.
• android.versionCode dropped. It was kept as a seed to initialize the remote counter on the first appVersionSource: remote build. That has happened — the counter stands at 5 on EAS — so the field was inert, and the CLI recommended removing it on every build.
• Workflow — the "Stamp the version into app.json" step is gone; the build step passes LYFTR_VERSION instead.
• RELEASING.md — the "bump expo.version before tagging" instruction is replaced with "there is nothing to bump."

Verification

expo config --json resolves correctly both ways:

env	resolved version
unset	0.0.0-dev
LYFTR_VERSION=0.1.0-beta.5	0.1.0-beta.5
LYFTR_VERSION=9.9.9	9.9.9

android.versionCode resolves to undefined, i.e. left to EAS. npm run type-check passes.

What this can't prove before merge: a dry run leaves LYFTR_VERSION unset by design, so CI can only exercise the fallback. The tag path is verified locally via expo config above, but its first real proof is the next tagged release — where the existing gate already fails the build if the APK's versionName doesn't match the tag. So a regression here surfaces as a failed release, not a bad one.

🤖 Generated with Claude Code

https://claude.ai/code/session_01JxKyZy6vR6xKu7Sp2MrEwn