Launched this week
ShipGuard
Make Codex prove the diff before risky edits
6 followers
Make Codex prove the diff before risky edits
6 followers
ShipGuard is an open-source, local-first workflow kit for AI-assisted iOS maintenance. It makes Codex start with a scoped task contract and finish with diff, evidence, and claim verification. The goal is simple: let agents move fast around build, simulator, StoreKit, widgets, notifications, performance, and release work without losing reviewability.
Hey Product Hunt, I'm Jason. I built ShipGuard while working on Ringly because Codex is genuinely useful on real iOS product work, but raw agent speed near production app surfaces can get spicy fast.
The hard part was not "can the AI write Swift?" The hard part was making agent work reviewable before it touches notifications, StoreKit, widgets, App Intents, background modes, performance-sensitive UI, or anything that ends up in a release claim.
ShipGuard is my answer: let Codex cook, but make it bring receipts.
It is an open-source, local-first workflow kit around Codex for production iOS maintenance:
prepare a scoped task contract before Codex edits
identify risky iOS surfaces first
route the right proof lane: build/run, logs/debugger, simulator, SwiftUI preview, profiler, device/TestFlight/App Store/manual review
verify the exact diff, evidence, and claims after Codex works
keep local/simulator proof separate from release proof
grade reports before turning them into work
redact and share safely
audit ShipGuard's own commands, docs, skills, plugin metadata, package proof, and release evidence instead of trusting pretty guidance
prioritize the next useful action instead of dumping a giant checklist
keep private app observations out of public artifacts
The latest loop is the part I wanted from day one:
prepare the task -> let Codex work -> verify the diff, evidence, and claims.
The repo is public, MIT licensed, and CI-validated. I am still moving fast on it, so feedback from iOS devs and agent-heavy builders is exactly what I want.
Curious what you would require an agent to prove before you trust it near production iOS code.