Shipdeck

Shipdeck is a native macOS app (SwiftUI) for solo developers running many apps. Each product has one or more components — an iOS app, a backend, a landing page — spread across one or more repos. Open it and you see which app needs a release, which has uncommitted work, which has gone quiet, and exactly why.

Landing page: shipdeck.taiphanvan.dev. In beta; the source is not open source. The story behind it: Shipdeck: one person, many apps.

Background#

With QuickSpend, Examino, PES Arena and this site running in parallel, every return to a product turned into archaeology: git log, App Store Connect, old notes. Shipdeck puts code state, store state and each product's Claude Code sessions in one place.

Tech stack#

SwiftUI + Swift — native macOS app; product data lives in a local products.json, nothing is written into repos.
AI-free overview — reads git, manifest versions, Sparkle appcasts and iTunes Lookup to sort products into Needs a release / In progress / Quiet for a while / All good. Websites and links are checked on a schedule too.
Claude Code survey — adding a product means picking a folder; a headless, read-only session finds components, commands, services and docs. Sync with Code re-surveys from new commits and saves only the fields you tick.
App Store Connect API — versions, review state, phased release, TestFlight, user reviews, App IDs and capabilities; release or reply to reviews in the app. Keys stay in the keychain.
Release checklist — per-platform checklists (iOS, Android, Web) that tick what can be read from the repo and App Store Connect.
SwiftTerm — embedded terminals for each product's Claude Code sessions, which keep running when you switch products.
Action layer + MCP — every change is an action with a stable id and a tier (read, local, remote, publish, owner); agents call it over a CLI or MCP and stay inside their session's product.
Sparkle — signed, notarized auto-updates.

Challenges#

Letting agents use the tool without losing control: remote and publish actions wait in an approval inbox, publish always asks, and every run is journaled. AI fills in product data but never overwrites it on its own, and the survey runs without a shell and with git-ignored and secret files blocked.

Lessons#

Let boring tech handle what must be fast and reliable; bring AI in only where understanding code is the job. When Claude Code writes most of the code, my job is product owner: say clearly what I want, use it daily, and catch what feels off.

Tech stack#

SwiftUI + Swift — native macOS app; product data lives in a local products.json, nothing is written into repos.

AI-free overview — reads git, manifest versions, Sparkle appcasts and iTunes Lookup to sort products into Needs a release / In progress / Quiet for a while / All good. Websites and links are checked on a schedule too.

Claude Code survey — adding a product means picking a folder; a headless, read-only session finds components, commands, services and docs. Sync with Code re-surveys from new commits and saves only the fields you tick.

App Store Connect API — versions, review state, phased release, TestFlight, user reviews, App IDs and capabilities; release or reply to reviews in the app. Keys stay in the keychain.

Release checklist — per-platform checklists (iOS, Android, Web) that tick what can be read from the repo and App Store Connect.

SwiftTerm — embedded terminals for each product's Claude Code sessions, which keep running when you switch products.

Action layer + MCP — every change is an action with a stable id and a tier (read, local, remote, publish, owner); agents call it over a CLI or MCP and stay inside their session's product.

Sparkle — signed, notarized auto-updates.