Maintaining an AI-built property finance app: proven Codex testing, release and handoff workflows?
Maintaining an AI-built property finance app: proven Codex testing, release and handoff workflows?
I’m building a small family property-management app using ChatGPT for requirements and Codex desktop on Linux for implementation, with TypeScript, PostgreSQL/Supabase and Vercel.
It covers renovation forecasts, quotes, costs and payments, operating income and expenses, and family loans. The app is already in use. We have automated checks, isolated synthetic data, financial reconciliation and controlled releases.
I’m looking for practical experience from people who have maintained an AI-built financial or admin app beyond its prototype:
- How do you keep requirements, acceptance tests, candidate versions and deployed releases aligned across long Codex threads and workspaces?
- What browser-testing and release-evidence workflow has caught failures that builds and unit tests missed?
- Which GitHub skills or agent packages have improved your results without conflicting instructions or excessive process?
- What works well for translating approved mockup images into maintainable application code, including screenshot comparison and mobile behaviour?
- What broke after deployment, and what did you change?
Links to maintained repositories, runbooks, example acceptance records or fictional-data screen recordings would be especially useful.
The aim is to improve the existing application and workflow while keeping everyday tasks understandable for non-technical family users.
1 comment
Know an answer, or have the same question?
Log in to comment Create a free accountNot a finance app, so take this as one data point: I maintain a live AI-built web game with real users (Claude Code rather than Codex), and your five questions aren't finance-specific. What's worked for me: 1. Alignment: I stopped trusting the thread. The repo is the memory. Every job starts from a committed brief, every session ends with a decision-log entry, and long threads get retired and restarted from a "read these first" list. Releases are fast-forward merges only, and the live site shows its commit ID, so I can confirm what shipped is what was tested. 2. What tests missed: headless-browser tests and a real test deploy catch a lot, but my phone caught the rest. iOS sharing silently dropped the message text, and a button sat off-screen at one phone size. Both passed every automated check. 3. Agent packages: I've avoided them. Plain instruction files in the repo, one source of truth per fact. Conflicts mostly came from the same fact written in two places. I can't hold every doc in my head, but the agent reads all of them, every time, and takes each one literally. When two disagree, it doesn't ask; it quietly picks one, and I find out when something breaks! 4. Mockups: no images. The agent builds the mock from the real stylesheet, I approve it, and it becomes the spec the tests check against. 5. What broke: win played on a phone in airplane mode was lost silently, with no error anywhere. In your app that would be a lost payment record, so I'd test offline and flaky-network entry deliberately. Mine now goes through a retrying outbox. And agent reports drifted (one said players rated a puzzle "just right"; the log said "too easy"), so key claims get checked against real data before anything ships. Your financial reconciliation is the same instinct. Full write-up of the workflow: https://app.therundown.ai/community/posts/4205e015-7e73-4337-aa34-4b262c5f5d77. Happy hunting!
Posting guidelines